<<

Spanish Velarinsertion and Analogy: A Usagebased Diachronic Analysis

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree of Philosophy in the Graduate School of The Ohio State University

By

Steven Fondow, B.A, M.A

Graduate Program in Spanish and Portuguese

The Ohio State University

2010

Dissertation Committee:

Dr. Dieter Wanner, Advisor

Dr. Brian

Dr. Terrell Morgan

Dr. Wayne Redenbarger

Copyright by

Steven Richard Fondow

2010

Abstract

The theory of Analogical and Exemplar Modeling (AEM) suggests renewed

discussion of the formalization of analogy and its possible incorporation in linguistic

theory. AEM is a usagebased model founded upon Exemplar Modeling (Bybee 2007,

Pierrehumbert 2001) that utilizes several principles of the Analogical Modeling of

Language (Skousen 1992, 1995, 2002, Wanner 2005, 2006a), including the

‘homogeneous supracontext’ of the Analogical Model (AM), frequency effects and

‘randomselection’, while also highlighting the speaker’ central and ‘immanent’ role in

(Wanner 2006a, 2006b). Within AEM, analogy is considered a cognitive

means of organizing linguistic information. The relationship between input and stored

exemplars is established according to potentially any and all salient similarities, linguistic

or otherwise. At the same time, this conceptualization of analogy may result in language

change as a result of such similarities or variables, as they may be used in the formation

of an AM for the input. Crucially, the inflectional paradigm is argued to be a possible

variable since it is a higherorder unit of linguistic structure within AEM.

This investigation analyzes the analogical process of Spanish velarinsertion

according to AEM. Velarinsertion involves the presence of a nonetymological voiced

velar // appears in the first person singular present indicative and all forms of

the present subjunctive, .g. 1s PRS IND pongo , 1s PRS SBJ ponga , cntr. 2s PRS IND pones . It affects a dynamic, yet heavily restricted lexical subclass of second and third ii conjugation characterized by synchronic variation and diachronic change. The diachronic development of the velarinsert class requires a sufficient level of synchronic detail so as to properly define the AM and identify changes in its multiple phonological, morphological, lexical and paradigmatic variables.

Previous analyses of velarinsertion, in Spanish and other Romance , are limited and/or ambiguous because they rely on the traditional notions of proportional analogy and leveling. Thus, several crucial details of the analogical process are assumed or overlooked in order to more broadly describe the diachronic extension of the velarinsert across a series of verbs having different stemfinal segments. Relatedly, the imprecise defining of the AM has led to numerous hypotheses concerning the original model for velarinsertion including leader verbs like decir , 1s PRS IND digo , conocer , 1s

PRS IND conozco or tañer , OSp. 1s PRS IND .

Several other theoretical issues are also addressed as they relate to AEM and

velarinsertion. The role of frequency in the salience and resting activation levels of

exemplars explains the traditional ‘resistance’ to regular in high frequency,

irregular verbs. Also, the paradigmatic structure that emerges from the allomorphic

distribution of the velar is argued to be a crucial variable in the diachronic development

of the velarinsert class. Finally, the preceding points in conjunction with the changes

affecting the composition of the AM lend themselves to a very weak predictive power in

identifying potential targets for continued extension of the velar, some of which are

attested dialectally or are accepted as standard variants in Modern Spanish, as well as

Catalan and Italian.

iii

For Meghan.

iv

Acknowledgements

The following examination of Spanish velarinsertion is the culmination of seven years of investigation that began in a historical class taught by Dr. Mary

Beckman. Subsequent independent study under the supervision of Dr. Wayne

Redenbarger provided me with plenty of food for thought and wonderful discussions of not only velarinsertion but also , analogy and, perhaps more interestingly, baseball. At the same time, various classroom projects led me to refine my views on

linguistic theory. I would like to thank Dr. Fernando Gil for letting me argue

some rather bold points and for giving me ‘clases particulares’ on Optimality Theory and . Thank you to Dr. Brian Joseph who not only taught me to think about what

morphology is but has always been a comforting presence in all of my more recent

academic endeavors. Your classes were amazing. My teaching duties at Ohio State also

let me take full advantage of my linguistic training. Thanks to Dr. Terrell Morgan for

letting me teach for so long and for being so willing to meet with me

to discuss so many issues.

To my advisor Dieter Wanner, you have shaped my views on language and philosophy more than you could possibly imagine. I want to thank you for taking me on

as a student and for asking the tough , even when it took me a while to

understand exactly what it was you were asking. Your patience and work are apparent in

this and many other of my projects.

A special thanks to everyone at The Ohio State University, especially faculty and

staff in the Departments of Linguistics and Spanish and Portuguese. Treating a graduate

student like me as a colleague made a world of difference in my eyes.

My interest in linguistics traces back, interestingly enough, to my hometown of

Antigo, Wisconsin. Special thanks go to my grandfather, Richard E. Fondow I, Mr.

Everett Ison and the Santín family for making language so fascinating to me.

At UWMadison, Dr. Juan Carlos Temprano helped guide me as an undergraduate

and nominated me for a scholarship which I used during my academic year abroad at the

Universidad de la Complutense in . My time there assured me that language and

lingusitics were fields that I wanted to pursue. As a result, Dr. John Nitti was gracious

enough to give me the opportunity to transcribe a copy of , as well as

hire me as a Research . In Graduate School, Dr. Ray HarrisNorthall taught me

the foundations of Linguistics that I needed to reach this point.

Finally, I want to express my unending appreciation for all my family and friends.

Your patience and love has kept me going even when I was unsure if I could finish this project. Of course, if you were not so fun and giving, I might have been done years ago!

To my parents and brother, you have always given me a helping hand and been accepting

of the countless ventures I have undertaken. The biggest thanks without a doubt go to my

wife, Meghan, and daughter, , you keep me going and wash away all the stresses in

life. Thank you for being so extraordinary.

vi

Vita

January 2, 1976 ...... Born Antigo, Wisconsin

1999 ...... B.A. Spanish and Linguistics, University of

WisconsinMadison

1999 ...... Research Assistant, University of

WisconsinMadison

19992001 ...... Graduate Teaching Assistant, Spanish and

Portuguese, University of

WisconsinMadison

2001...... M.A. Spanish, University of

WisconsinMadison

20022006 ...... Graduate Teaching Associate, Spanish and

Portuguese, The Ohio State University

2006present ...... Associate Lecturer, Spanish and Portuguese,

University of WisconsinMadison

Fields of Study

Major Field: Spanish and Portuguese

vii

Table of Contents

Abstract ...... ii Acknowledgements ...... v Vita ...... vii Fields of Study ...... vii Table of Contents ...... viii List of Figures ...... xiii List of ...... xiv Chapter 1: Analogy, Theory and AEM ...... 1 1.1—Analogy and Spanish velarinsertion...... 1 1.1.1—Analogy in lingustic history...... 3 1.2—Analogy in modern linguistic theory...... 10 1.2.1—Rulebased approaches...... 11 1.2.2—Constraintbased theories...... 15 1.3—Usagebased theories...... 18 1.3.1—General implications of usagebased analogical approaches ...... 18 1.3.2—Exemplar Modeling...... 27 1.3.3—Analogical Modeling of Language in Analogical Exemplar Modeling...... 31 1.3.4—Immanence...... 38 1.4—Conclusions and overview of the following investigation...... 39 Chapter 2: The History of VelarInsertion ...... 42 2.1—The velarinsert in Spanish...... 42 2.1.1—The Spanish verbal system...... 42 2.1.2—The Spanish velarinsert subclass...... 44 2.1.3—Overview of chapter 2...... 48 2.2—Velarinsertion across Romance...... 50 2.2.1—Portuguese...... 51 viii

2.2.2—Catalan...... 53 2.2.3—Italian...... 55 2.2.4—The linguistic boundaries of velarinsertion...... 58 2.3—(Late) developments leading to velarinsertion in Spanish...... 60 2.3.1—Conjugational class reduction via shifts in the CL system...... 60 2.3.2—Yod effects and palatalization...... 62 2.3.2.1—Palatalization by yod in the verbal paradigm...... 63 2.3.2.2—The –NGERE set...... 72 2.3.2.3—The decir , hacer set...... 75 2.3.3—The relevance of yod effects and palatalization...... 79 2.4—The analogical basis for Spanish velarinsertion...... 82 2.4.1—Velarinsertion and linguistic independence...... 86 2.4.2—Possible AM for velarinsertion...... 87 2.4.2.1—The –NGERE set, part II...... 88 2.4.2.2—The decir , hacer set, part II...... 91 2.4.2.3—An alternative AM for velarinsertion...... 93 2.4.2.4— of possible AM...... 96 2.5—Previous examinations of velarinsertion in Spanish...... 97 2.5.1—Synchronic rule and constraintbased approaches...... 98 2.5.2—An analogical perspective on velarinsertion...... 102 2.5.2.1—Traditional analyses...... 103 2.5.2.1.1—Synchronic competition between velarinsertion and palatalization...... 103 2.5.2.1.2—‘Multiple causation’ and a revised view of analogy...... 106 2.5.2.1.3—Reinforcement of 1s/SBJ allomorphy...... 112 2.5.2.1.4—Velarinsertion in lieu of palatalization...... 114 2.5.2.1.5—Inter and intraparadigmatic influence...... 117 2.5.2.1.6—Stem allomorphy and phonological balance...... 117 2.5.2.2—Contemporary analyses of velarinsertion across Romance...... 121 2.5.2.2.1—The velarinsert as an inflectional component...... 121 2.5.2.2.2—The ‘indexical’ function of the velar...... 123 2.5.2.2.3—The implications of velarinsertion in Catalan...... 124 2.5.2.2.4—The 1s/SBJ ‘morphome’...... 125

ix

2.6—Diachrony, analogy, paradigmatic function and immanence...... 128 Chapter 3: A Usagebased Approach to VelarInsertion ...... 131 3.1—Velarinsertion as an analogical process...... 131 3.1.1—Immanence and ‘multiple causation’ in AEM...... 131 3.1.2—Synchrony in diachrony...... 133 3.1.3—Outline of chapter...... 136 3.2—Defining ‘multiple causation’ in the AM...... 137 3.2.1—The phonological variables of the AM...... 137 3.2.1.1—Preceding conditioning environment...... 138 3.2.1.2—Postvelar conditioning environment...... 141 3.2.1.3—Additional phonological factors...... 142 3.2.2—Morphological characteristics of the AM...... 143 3.2.3—The morphomic function of the paradigm...... 144 3.2.4—Immanence and linguistic categorization...... 146 3.3—Looking back at velarinsertion...... 147 3.4—Orthographic problems...... 151 3.5—Velarinsertion: A diachronic analysis through synchronic examination...... 155 3.5.1—Stage 1: Preliterary developments...... 156 3.5.2—Stage 2: Lateral (and rhotic) expansion...... 160 3.5.3—Stage 3: Sonority and the AM...... 164 3.5.4—Stage 4: Contrasting diachronic results...... 171 3.5.5—Stage 5: Prescriptive variation...... 173 3.5.6—Stage 6: Extreme velarinsertion and idiosyncratic analogy...... 176 3.6—Dialectal AM and potential directions of change...... 182 3.6.1—The of outside influence...... 183 3.6.2—Stage 7: Dialectal velarinsertion in Modern Spanish...... 183 3.6.3—Analogical change across ...... 186 3.7—Theoretical consequences and conclusions...... 188 Chapter 4: AEM in Theory and Practice ...... 190 4.1—Theoretical and practical issues...... 190 4.1.1—The elegant complexities of analogy...... 191 4.2—The cognitive paradigm...... 193

4.2.1—Paradigm theory...... 195 4.2.2—Paradigmatic processes...... 198 4.2.3—Paradigmatic function...... 203 4.2.4—Paradigmatic morphologization...... 205 4.2.5—A broader view of the 1s/SBJ template...... 206 4.2.6—Homophony...... 208 4.3—Frequency effects...... 209 4.3.1—Type frequency...... 213 4.3.2—Token frequency...... 218 4.4—...... 225 4.5—Where to ‘go’? Analogy and directions of change...... 230 4.5.1—Potential directions of change...... 231 4.5.2—Catalan...... 232 4.5.3—Italian...... 234 4.5.4—Potential, competition and frequency in Spanish...... 236 4.6—Conclusions...... 242 Appendix A: Modern Spanish Regular Conjugational Classes () ...... 244 Appendix B.1: Modern Spanish (NonEtymological) VelarInsert Verb Class ...... 245 Appendix B.2: Modern Spanish VelarInserts Derived by Prefixation ...... 246 Appendix : Present Paradigms of Standard Modern Spanish VelarInsert Verbs ...... 247 Appendix D: Attested Forms of remanir ...... 249 Appendix E: Verbs with Attested Velar in Synchronic Variation ...... 250 Appendix F: Romance Patterns corresponding to Spanish tener , venir ...... 251 Appendix G: Modern Verbs corresponding to Standard Modern Spanish VelarInserts ...... 254 Appendix : Standard Modern corresponding to Standard Modern Spanish VelarInserts ...... 256 Appendix I—Standard Modern Italian Verbs corresponding to Standard Modern Spanish VelarInserts ...... 258 Appendix —Classical Latin Regular Conjugational Classes (Present Tense) ...... 260 Appendix K—Some Potential Targets for VelarInsertion in Modern Spanish ...... 261 Appendix —1s/SBJ Template in Standard Modern Spanish ...... 264 Bibliography ...... 267

xi

Endnotes ...... 285

xii

List of Figures

Figure 2.1—Distribution of the velarinsert, as seen in poner “put, place” …………….44

Figure 2.2—CL TENĒRE and MSp. tener ………………………………………………..46

Figure 2.3—Examples of Extension of VelarInsertion in MCat………………………..54

Figure 2.4—Velar/PalatalFinal Dialectal Variation in MIt…………………………….57

Figure 2.5—Sp. verb forms not showing stemfinal palatalization by yod ……………..67

Figure 2.6—Diachronic development of MSp. tañer , from CL –NGERE verb set ………74

Figure 2.7—Diachronic development of MSp. decir and hacer ………………………..77

Figure 2.8—Examples of the CL –NGERE verb set ……………………………………...91

Figure 2.9—Example of the ‘inchoative’ verb set ………………………………………94

Figure 3.1—Example of various paths for change in Romance …………………….....132

Figure 3.2—Evidence in favor of palatalization prior to velarinsertion ……………....150

Figure 3.3—Variation of –uir verbs in S. Juan Bautista de la Concepción (17 th c.) ...... 179

Figure 4.1—Most frequent verbs by token (Juilland & ChangRodríguez 1964) ……..211

Figure 4.2—Frequency of verb classes (Davies 2006) ………………………………...215

Figure 4.3—Token frequency of 3s/3p SBJ synchronic variants of roer (per CORDE) 219

Figure 4.4—Frequency of standard velarinserts (Juilland & ChangRodríguez 1964) .221

Figure 4.5—Token frequency of synchronic variants of salir (per CORDE) ………....223

Figure 4.6—Token frequency of synchronic variants of valer (per CORDE) ………...224

xiii

List of Abbreviations

1s first person singular present indicative (unless otherwise specified) 2s second person singular 3s third person singular 1p first person plural 2p second person plural 3p third person plural I first conjugation (ar ) II second conjugation (er ) III third conjugation (ir ) AEM Analogical and Exemplar Modeling AL Analogical Modeling of Language AM Analogical Model ant anterior arc archaic AUX auxiliary function bk back c century C consonant ca circa Cat Catalan chapter CL Classical Latin COND conditional tense cont cor coronal dial dialectal form dist distributed EM Exemplar Modeling Fr French FUT G glide IDENTIO Faithfulness constraint involving identity between Input and Output IDENTOO Faithfulness constraint involved identity between Output and Output IMPF aspect IMPTV imperative IND indicative mood INF /infinitival marker xiv

IOF InputOutput Faithfulness (or Correspondence) It Italian JSp JudeoSpanish Lg language MArag Modern Aragonese MCat Modern Catalan MEPtg Modern European Portuguese MFr Modern French MIt Modern Italian MLeon Modern Leonese MOc Modern Occitan MOcPrG Modern Occitan, Provençal and Gascon dialects MProv Modern Provençal MPtg Modern Portuguese MRom Modern Romanian MSp Modern Spanish nas nasal NE Modern English OCamp Old Campidanese OCat OFr OIt Old Italian OOF OutputOutput Faithfulness (or Correspondence) OProv Old Provençal OSp OT Optimality Theory P person PART past PRET (or ) PRF Perfect PRS present tense Ptg Portuguese Rom Romanian SBJ (all forms unless otherwise specified) sonorant Sp Spanish ton tonic TV theme vowel UR underlying representation V vowel vel velar voc vocalic VL Vulgar (or Late) Latin X xv

CHAPTER 1: ANALOGY , THEORY AND AEM

1.1—ANALOGY AND SPANISH VELAR -INSERTION . Analogy, or the use of existing information to properly frame new or unknown input, is a welldocumented and longdiscussed tradition in psychological analysis. As such, it is also a crucial aspect of cognitive linguistic processing. Unfortunately, through analogy has been given a bad reputation, thought to apply only after other, more defined linguistic mechanisms make their way through the , leaving behind an organizational mess. Analogy is typically seen as unrestricted, ambiguous and lacking clear definition.1

Thus, it falls outside the main of many theoretical models, being treated in passing

through special formal devices or mere descriptive analysis (i.e. informally).

What follows is an attempt to define the analogical process as being central to the

cognitive structuring of linguistic information. The example of Spanish velarinsertion,

i.e. the presence of a nonetymological velar obstruent /g/ in the 1s PRS IND and all

forms of the PRS SBJ (1s/SBJ) in a lexicallyrestricted group of verbs, sets the stage for

the incorporation of analogy into the theoretical model of Analogical and Exemplar

Modeling (AEM), illustrating how analogy starts and stops and proceeds and retracts

over time. At the heart of this investigation is the examination of the origin of this velar

element and, more importantly, a thorough analysis of its subsequent diachronic

extension during which time some forms alternate between velar and nonvelar. The goal 1 is to best describe and identify both when and how this set of verbs came to be, and when and how it grew, i.e. to explain its localized productivity and its seemingly decreasing productivity over time.

Although analogy has long been the most widely accepted solution to the problem of velarinsertion in Spanish, the analogical developments of previous analyses have always been vague with respect to the precise composition of the Analogical Model

(AM). The current chapter focuses on these formal and theoretical problems. The following chapters attempt to properly define the AM through ‘multiple causation’, including lexical, phonological, morphological, and paradigmatic factors (cf. Sapir 1921,

Joseph 1982, 1983). In the end, it becomes clear that a linguistic model that is immanently realistic and psychologicallybased requires analogy to be a dominant force in the linguistic development.

The major insight that comes from this analysis is that the development of the velarinsert class involves higher order analogical processing of morphological information, taking the form of the inflectional paradigm. The velarinsert data demonstrate the involvement of inflectional patterns in the extension of the velarinsert to new verbs, not to mention their characterization according to said patterns. According to

AEM, this is a consequence of the cognitive organization of linguistic information, here paradigmatic allomorphy, and is the crucial piece of the puzzle that completes the analysis of the Spanish process.

2

1.1.1—ANALOGY IN LINGUSTIC HISTORY . In light of the fact that analogy is an important part of cognition, not only in the processing of linguistic information, but also visual, auditory, and other sensory stimuli, it is necessary to reexamine the analogical process, giving it new life and greater focus in linguistic theory. of the goals of this proposal then, is to detail the formal method by which analogy functions.

Almost every historical linguistic analysis, volume of conference proceedings, textbook, etc., discusses some form of analogy. Even before linguistics developed as a field of study, treatment of language and linguistic practices appealed to analogy. 2 It is beyond the scope of the current investigation, however, to thoroughly document the

history of analogy in linguistic theory. 3 Therefore, the following discussion focuses from the outset on two oftcited, highly similar types of analogy: fourpart or proportional analogy and leveling. The term ‘fourpart’ analogy refers to use of an established formal relationship between two forms to create an identical or similar relationship in another pair of forms, of which the primary target for change may be unknown or innovative.

More concretely, proportional analogy extends a morphological relationship. An example is the use of the NE past participle marker –en dialectally i.e. got :gotten :: shot :X,

where X= shotten .4 Leveling, meanwhile, refers to increasing similarity, or decreasing

idiosyncrasies, normally to the point of identity, between a target form and a model that

share an inflectional relationship (see an example below). 5 It removes (generally

unnecessary) morphophonemic variation from paradigms, tending to establish a semiotic

relationship (one form to one meaning).

3

As indicated previously, traditional analogy is assigned the ‘cleanup spot’ after regular sound change has taken place (i.e. Sturtevant’s Paradox). 6 Accordingly, the process of analogy has been reinterpreted teleologically, implying language change through the regularization of less regular patterns, especially those that are not based solely on phonetic conditioning.7 Such regularization is exemplified in the leveling of

allomorphy in the inflectional forms of a verb (e.g. OSp. 1s yergo , 2s yerzes >> MSp. 1s yergo , 2s yergues ) or in the loss or lexicalization of a plural marker in the nominal system

(e.g. NE pl brothers , brethren ). Analogy, as defined in this way, is a diachronic process

independent of linguistic structure and is often believed to imply that change will occur

when similarity is established.

King (1969) supports the preceding arguments, suggesting that analogy cannot be

central to the development of grammar because regularization may be construed as

simplification. It is generally evident, however, that the simplification of one aspect of

language often involves the complication of another, e.g. the reduction of vowel quantity

in Latin that had a direct effect in the changes among conjugational classes as discussed

in chapter 2. Analogical changes like Spanish velarinsertion and the palatalization

effects in verbs like ceñir (see below) may complicate grammar, thereby creating

nonoptimal situations both internally and externally (cf. ‘morphologization’, Joseph &

Janda 1988; see §1.3 for a definition of grammar in AEM). This is exemplified by the

traditionally accepted linguistic principles that certain forms may resist change or that

there are at least some types of language change which are gradual (see discussion of

these principles below).

4

In contrast, another view of analogy is presented here, namely one where the associations or ‘analogies’ between one form and another (or several others) may result in change or in no change whatsoever. In essence, analogy is the process of establishing relationships of similarity (cf. Anttila 2005). The ‘analogical event’, in turn, is language change resulting from such a relationship. Hence, the proposed view of analogy has two parts:

First, the perception of similarity that underlies the operation of analogy amounts to a judgment of partial sameness...this judgment is unproblematic, since the similarity must only be true for the individual and it only exists as a visible effect if it is enacted in the processing sense of analogy. There does not exist an independent, external standard of correction to which this judgment must conform. It may be more or less evident or felicitous from an observer's perspective, but what is its appropriateness in the linguistic individual's mind. Under the second aspect of analogy, the effect of the similarity judgment is an event that assimilates the target to the model, partially or even . Categorization of the target under the model class amounts to the same kind of , since it attributes to the target a classificatory feature belonging to the model. Vice versa, an assimilation by itself also constitutes a local minigeneralization, whereby the assimilated material takes on the role of the class label. In brief, analogy, in the second sense, is a process that marks items of partial or full local identity by assimilation or classification (categorization). (Wanner 2006b:121)

Accordingly, analogy, at its core, is a “perceived similarity [for individual

speakers] in a linguistic dimension” (Wanner 2006c). Although not generally the case,

such similarity requires as little as a single model and target for change to occur. In this

instance, if the target becomes more similar to its model, then the analogical change is

locally regular. At the same time, the ‘potential’ for change, as opposed to absolute,

deterministic change, may be predicted from the elements or variables common to both

model and target.

5

Before pursuing this line of reasoning further, a handful of arguments that support it over the traditional view of analogy should be discussed. The primary difficulty with the aforementioned types of proportional analogy and leveling is "their relative arbitrariness" (Wanner 2006b:45; cf. Lloyd 1987). Nearly any pair of forms and nearly any perceived relationship between them may serve as the potential model for change.

The process is not constrained in any real way, thus being overly powerful and highly imprecise.

Based on their normally diachronic perspective, these analogies are merely descriptive or observational. They are practically devoid of explanatory power, considering that the proportional equation, for example, tends to list only one model for a target in spite of the potential involvement of countless other models of the same type (cf.

Lloyd 1987). These problems mean that numerous associative insights are overlooked.

Labov (1994), for example, suggests that there are there many factors involved in the process of change (cf. Lenfest 1978). Such factors are informative in defining the analogical process and so need to be included in any analysis. Among the many factors explored here which the basic proportional equation does not acknowledge are frequency effects and associative strength, two very relevant indicators of the potential for change to occur (cf. ch. 4, KraskaSzlenk 2007). In another manner of speaking, traditional formulations do not properly define any aspect of the actual analogical process.

Aside from being overly powerful and lacking insight, traditional analogy is

considered too rigid. Some analogical processes do not fit the available formalizations,

e.g. the identical distribution of allomorphy across the inflectional paradigms of multiple

6 verbs (or, more generally, any model consisting of more than two elements). When analogy is correctly specified and becomes the focus of linguistic processing, however, the preceding issues do not arise, they are rather a "pseudodilemma" (Wanner

2006b:45).

All this relates the proposed view of language change and analogy to those of a number of previous investigations, including Lloyd (1987), Malkiel (1969, 1974), Hock

(2005), Wang (1969), Labov (1994). These works do not completely accept the

Neogrammarian division of regular sound change and analogy, instead contending that language change is not obligatorily global, nor deterministic, in spite of the potential for regularity/finality to occur. Hock (2005), for instance, analyzes the relationship between regular sound change and analogical change as a continuum with degrees of systematicity related to morphological and/or semantic loading. This continuum is analogical in the broad sense of pattern extension. The regularity of each point along the continuum depends upon its “domain” of application. The problem that arises between regular sound change and localized analogicallybased change boils down to the socially marked strength of change and prohibits us from making any (specifically linguistic) predictions as to which types of change will be regular (cf. Joseph & Janda 1988). This is considered a serious theoretical problem in some respects though it seems less problematic in the analogicallycentered analysis elaborted herein because regularity is defined locally.

Regardless of the problems associated with the typical view of analogy, it does provide several interesting insights into the actual analogical process. First, it constructs an inferential relationship between new and previously acquired information. 8 This is an

7 important consequence of analogy, that is, it builds upon previous knowledge. Simply put, “one uses known cases to understand new or unknown cases; there is no mystery”

(Anttila 2005:437). Second, the pair of model forms used in the proportional equation is chosen because of some established relationship between them. This relationship, while glossing over much of the relevant information, does serve as a starting point for the identification of the associative features that hold between the model forms and the target pair respectively. Finally, the model forms are really representative of larger groupings which bear the same relationship. Hence, it is possible and, in fact, necessary to identify the Analogical Model (AM) more completely than is typically done. This allows greater precision in determining the crucial features, or variables, that define the AM, as they may be relevant in establishing an associative relationship with the target. Such variables are also helpful in identifying potential directions for change, thereby adding a slight degree of predictive power to the analogical process.

On the basis of the overly powerful function of analogy in earlier analyses, some investigations have attempted to restrict its domain of application (e.g. Kiparsky 1974,

Lahiri 2000). This is equivalent to properly defining the analogical process as proposed here, since providing the correct level of detail and specification obviously limits its application moreso than the arbitrary, potentially omnipotent force that analogy has been in the past. 9

To properly define the process then, it is necessary to answer the following questions (cf. Wanner 2006a). First, what are the factors which make a speaker favor one set of forms over another as an AM? Next, how general or specific must these factors

8 be? Finally, what does the proportional equation represent, i.e. how many forms or factors are needed for analogy to occur? These questions are important given that "it should always be possible to show that the resemblances noted bear relevantly on the point to be established, whereas the differences are irrelevant. In many cases it is difficult to be sure of this distinction, and arguments from analogy are therefore precarious unless supported by considerations that can be established independently"

(Lahiri 2000:3). The remaining sections attempt to shed light on these and other

questions.

Because analogy is not truly deterministic nor seemingly systematic empirically,

it often comes across as unscientific. 10 It is difficult to incorporate into linguistic theory because it does not appear to be regular or predictable and may complicate the grammar.

This explains why it has so often been viewed as an extratheoretical ‘last resort’.

However, "[i]t is clear...that if the phenomenon of uniformity based on the association of

in the mental exists, then it cannot be dismissed on the grounds that the

contributing factors are too complicated" (Downing et al. 2005:5).

Therefore, the current investigation suggests that analogy is the cognitive faculty

at the center of linguistic processing (cf. Anttila 2005, KraskaSzlenk 2007). It functions

in a usagebased model across forms and meanings as well as across cognitive and

classificatory domains. However, the process is affected by acquisition, production, and perception (Kiparksy 1974). This relates to the fact that it is reallife speakers who use

language, implying a level of insecurity and imperfection that needs to be incorporated

into any linguistic model that is believed to represent actual language use.

9

The preceding discussion sets the stage for identifying the treatment of analogy, especially proportional analogy and leveling, in three of the most widely utilized types of contemporary linguistic theory. A rulebased analysis is shown to account for very few types of analogy in any explanatory way. The use of constraints shows a greater capacity for handling analogical associations by highlighting the relevance of the relationship between surface forms. In the end, however, constraintbased theories are psychologically unrealistic and require similar specialized or ad hoc devices as those used by a rulecentered approach (see below). This leads to the examination of two related usagebased analyses which are combined to provide the basis for the theoretical model used in the current investigation. The direct associative relationships between surface forms that are inherent in such models in conjunction with the level of cognitive realism they provide, suggest that they are more elegant and better suited for an analogicallycentered view of linguistic processing.

1.2—ANALOGY IN MODERN LINGUISTIC THEORY . The treatment of analogy in contemporary theory often carries over from linguistic tradition. In many ways, the process of analogy remains ambiguous and tends to fall into the theoretical margins of linguistic analysis. The following sections demonstrate how proportional analogy and leveling are generally handled in rule and constraintbased theories. In the first case, what could be considered analogical is not treated as such and is resolved through the use of technical or specialized devices. In the latter, innovative constraints and ranking hierarchy determine the optimal output. The more specific treatment of Spanish

10 velarinsertion in both rulebased and constraintbased models is treated in the next chapter.

1.2.1—RULE -BASED APPROACHES . Synchronic rulebased approaches utilize extremely different methods to deal with proportional analogy and leveling. Proportional analogy resulting in exceptionality has to be resolved through formal methods which ultimately obscure or eliminate much of the associative relationship between surface representations. Identifying such nonsystematic allomorphic patterns typically involves specially formulated rules and/or lexical marking (cf. King 1969, Kiparsky 1974).

Essentially, there are two options: (1) to account for only the broadest, exceptionless regularities by rule, marking all others, frequently through manipulation of their underlying representation, or (2) to trigger special (minor) rules through lexical marking, including morphological encoding, to account for the largest sets of exceptions to the broad rule system. 11

If the rules only account for the completely systematic forms, the number of exceptions is great and countless potential associations between forms, irregular and regular, are lost. Alternatively, if the system permits minor rule application on exceptional subsets, the rule system itself becomes far more complex since it requires lexical marking and more than likely a greater number of minor rules to account for a handful of forms than are needed to deal with the exceptionless sets. The consequence of any attempt at maximal generalization through rules in a dualroute system involves of lexical information and an undesirable processing load for the linguistic

11 individual. 12 A more elegant system would incorporate all forms so that exceptionless

and exceptional sets and subsets could be linked externally and internally to create potential generalizations (analogies) which may be manipulated by the speaker for

linguistic processing.

Unlike the complications involved in proportional analogy, leveling reduces the

strain on the system through reanalysis or simpflication of the grammar. This is achieved

through rule loss, reordering or reformulation while the changes in the lexicon are brought about through loss of lexical marking or reanalysis of the abstract, underlying

representation.

The idea that analogy involves identifying the proper underlying representation(s)

to account for the attested surface forms (cf. similar arguments applying to Optimality

Theory, Bachrach & Nevins 2008), is most appropriate for the largescale diachronic

change because it attributes an extremely static picture to language. It also calls into

question the ability of such a system to handle synchronic variation, which may be

explained by way of variable rules and/or the ‘generational gaps’ that occur

synchronically (cf. Skousen 1989 and Kiparsky 1974 respectively). Once more, there is a

heavy cognitive burden on the speaker:

[L]exicallydetermined variation cannot be represented in a variable rule, it must be represented lexically, especially if it affects all or most of the words of a particular phonological shape. Trying to build such variation into a variable rule would require repeating the contents of the lexicon in the rule. (Bybee 2000:71)

Where the aforementioned methods correctly identify surface representations,

analogy is not believed to be involved. Therefore, analogical relationships, i.e. the

12 similarity between surface forms remain outside of the domain of the theoretical model.

Thus, from an acquisitional standpoint, irregularity is initially attained through lexical specification (Kiparsky 1974). This means that the grammar evolves from lexicalization of individual items to generalization of a rule or set of rules followed by the item by item restructuring/relearning of irregularities. Because irregularity of this sort may simply be a sort of subregularity in one direction or another, i.e. more idiosyncratic or more uniform, countless amounts of information and generalization are rendered useless as they cannot be easily or elegantly incorporated into the grammar. Analogy (and borrowing) "…tend to become terminological receptacles devoid of explanatory power— catchalls for irregularities in the operation of ‘regular sound laws.’ This has too often been the demeaning fate of analogy in historical work" (King 1969:127). 13

Another highly related complication for the rulebased approach is the localized

domain of analogical change, meaning that such changes need not apply systematically to

all forms that fulfill the analogical criteria (cf. Joseph & Janda 1988). This suggests the potential for gradualness and resistance to change and makes it necessary to work around

the simultaneous, numerous, often obvious, nonphonological relationships between

forms (cf. Hock 1991). 14 Therefore, additional levels of grammar are incorporated into the rule system (e.g. Lexical Phonology, Distributed Morphology), requiring new types of conditioning environments and features. These include, “(1) natural subsets of the lexicon (, Verbs, ), (2) specific grammatical morphemes and combinations of these morphemes ("first person plural"), and (3) at most a few idiosyncratic lexical items” (King 1969:120). The outcome of such theoretical

13 modifications equate to a weakening of the restriction on universal or systematic application of rules, especially when the conditioning environment cannot be formulated in purely phonological terms.

An important distinction surfacing from a division of labor between phonology and morphology is that the former may be more generalizable (or rulelike) while the latter is less so, being more locally generalized and without global extension. Unlike most of their phonological counterparts, the hardandfast rules needed to generalize morphological patterns are complicated and riddled with exceptions, making a rulebased solution of this sort less elegant and, therefore, undesirable (cf. Carstairs 1990).

Morphology is organized differently from phonology, in part due to the shared phonological shape, semantic, and syntactic functions between forms. This is a primary

motivation for marking exceptionality to phonological rules. 15

On the basis of the limitations of a rulebased system, different approaches to the

formalization of grammar have been suggested:

One might propose a different relation…namely that analogy is the central force and is reflected in such things as rule reordering and rule loss…[changes in the rule system] would then be mere descriptions of what has happened rather than the prior events...that analogy is basic and the other things follow from it—requires us to be very specific about what analogy is and about the rationale for its occurrence. (King 1969:131)

Although this type of theoretical method has been met with some resistance (even by

King himself), it has recently seen a resurgence arising primarily from a more detailed and scientific view of ‘what analogy is’ and why it occurs (see below).

Before outlining alternative views of language change and the emergence of grammar, a final note is required with regard to rulebased approaches. In spite of all of 14 the techniques by which ‘exceptionality’ is resolved, explained or avoided, analogy continues to be an essential part of the linguistic process. Therefore, a system which takes advantage of analogy and discards the sometimes ad hoc nature of rules would be more elegant and logical, especially when viewed in relation to other types of cognition such as the associations of visual or auditory input to previously experienced information. 16

1.2.2—CONSTRAINT -BASED THEORIES . Rulebased approaches do not utilize relationships between surface forms in the development of grammar. Instead, relationships between forms are often abstract and underlying in order to properly trigger ruleapplication. Constraintbased accounts, meanwhile, avoid rules in favor of a hierarchically arranged series of constraints on input to the grammar. These constraints most often promote faithfulness between input and a set of output candidates or identify markedness of these candidates with respect to some (ideally universal) linguistic characteristic. The optimal output is the form that violates the fewest and/or lowestranking constraints among all candidates. Constraintbased approaches formalize the relationship between surface forms. The most common means of doing so is the formalization of OutputOutput Correspondence (or OutputOutput Faithfulness) constraints (OOF), including more refined versions such as Paradigm Uniformity

(Burzio 2005, Kenstowicz 2005, Steriade 2000) and Paradigmatic Contrast (Rebrus and

Törkenczy 2005). The result is an exponential increase in potential associations between

15 linguistic elements. Crucially, these types of constraints also take an important step in assigning analogy, and the inflectional paradigm, a formal, theoretical function.

Both leveling and proportional analogy, the two processes relevant to the present analysis, can be accounted for in constraintbased systems by implying (at least partial) identity/faithfulness across surface representations and specific constraint rankings (cf.

KraskaSzlenk 2007). Leveling is defined as a ranking which favors OOF over the constraint or set of constraints that explain alternations within the inflectional paradigm, e.g. Markedness or InputOutput Faithfulness (IOF) constraints. 17 Proportional analogy that promotes allomorphy is the reverse: Markedness and IOF outrank OOF. In reality, these two ranking permutations are combined in language, i.e. there are cases of both leveling and allomorphy. Thus, OOF for one subclass (lexical, morphological, etc.) outranks the alternation constraints (i.e. leveling), which outrank the OOF constraints applying to another subclass (i.e. allomorphy). Unfortunately, there are additional factors which determine the productivity and systematicity of such constraints over time, including the type frequency of the affected subclasses and the token frequency of their members (cf. §4.3). Diachronically then, changes in these subcategories will result in hierarchical reranking of the constraints.

Crucial theoretical consequences are implied by such constraints. First, there are relationships between surface forms that extend beyond phonetic similarity.

Morphological relationships emerge from surfacetosurface comparisons (cf. Myers

1999 for possible limitations). This is especially evident in the inflectional paradigm, where forms are related by a shared meaning and most often by a shared phonetic shape.

16

More importantly, OOF diminishes the relevance of positing an underlying representation, leading to arguments against underlying representations altogether:

[I]t is worthwhile to observe that outputoutput correspondence highly reduces the need for inputoutput constraints (I:O) and approaches a one level model of an OT grammar…Thus, the extra cost of extended, omnipresent O:O correspondence can be compensated by the absence of the respective I:O constraints. (KraskaSzlenk 2007:16)

Other notable theoretical improvements arise from OOF. First, the inflectional paradigm is assigned theoretical status. This is a salient aspect of the current investigation as the analogical process of velarinsertion in Spanish is confined to the inflectional paradigms of certain subclasses of verbs (cf. §2.1.2). Also, these constraints help to explain synchronic and diachronic variation via ranking permutations, including the possibility that some constraints are unranked relative to each other.

Depending on the approach, however, the relevant constraints vary in the forms they take and the degree of specificity with which they relate surface forms. Some analyses make use of extremely specific details in OOF including phonological features.

Others are general and ambiguous. There is no consensus as to which properties may be included in a correspondence relation and which may not. In large part, this is determined by the ‘faithfulness’ of constraintbased analyses to the traditional, universallyapplicable formulization of constraints.

Abstract, highly generalizable representations have often been considered

‘superior’ for their universality. This theoretical objective notwithstanding, universality can also be criticized as an obstacle that is unrealistic if the role of actual speakers is relevant. In order to process the number of universal constraints and seemingly infinite

17 number of potential outputs or optimal candidates, the processing load is far too heavy.

Additionally, it requires a tremendous increase in formal complexity (KraskaSzlenk

2007). Therefore, recent works have reduced or compromised the importance of the hypothesized universality of constraints and constraint rankings. Myers (1999), for example, formalizes fourpart analogy by incorporating languagespecific, and more importantly, specific information into the equation. 18 Consequently, constraints and rankings begin to resemble rules in the problems they encounter and in the way they are constructed. The following section proposes a theoretical model that is more elegant in that the emergence of organizational structure combines linguistic experience with the basic tools for cognitive processing of input.

1.3—USAGE -BASED THEORIES . The earlier arguments against an analogical approach notwithstanding, if it is constrained appropriately through a high degree of specification, analogy can serve as the basis for linguistic organization and change. Recent usagebased approaches, like Exemplar Modeling and Analogical Modeling of Language, have formulated an innovative version of analogy which is elegant and highly restricted so as to serve as the primary cognitive tool for categorizing language, in conjunction with its role in many other psychological processes (cf. Wanner 2006b).

1.3.1—GENERAL IMPLICATIONS OF USAGE -BASED ANALOGICAL APPROACHES TO

LANGUAGE . Usagebased theory focuses on the function of experience in the emergence

of grammar, contending that analogy, the means of associating experience and input, is

18 not goaloriented. Language is then a dynamic product of the speaker’s environment, shifting ‘on the fly’ (in realtime) as new information is acquired. This ‘shifting’ is analogical change, a weaklyextended and highlydefined process: “the essence of the analogical event is its full embedding in the specifics of a case, thus defying any notion of significant generality” (Wanner 2005:3). The strong implications and lack of elegance typical of the previously described models are consequently avoided. There are several additional reasons for preferring a model which highlights analogy and usage as opposed to rules or constraints.

First, analogy can be used to establish associative relationships of similarity between surface forms without appealing to any underlying representation or abstraction.

This is analogy in the first sense described earlier. New input is categorized according to previously acquired information through some degree of similarity, calculated by identity and difference (cf. Wanner 2006b, Anttila 2005). In this way, the analogical relationship is appropriately contextualized synchronically. The result may be generalization of some property or properties of one form (or set of forms) across an entire set or subset of the language, i.e. grammar. This constitutes the analogical event, or analogy in the second sense.

Such relationships are indicative of the attraction between forms on multiple identifiable levels simultaneously (i.e. the ‘Ganong effect’, Burzio 2005). 19 The

implication is that subcategories are easiest to identify when the broader categories are

known. As a consequence, the properties of the associative relationship between forms

are varied and may be derived from any and all linguistically relevant (and irrelevant)

19 information. 20 In part, this is because lexical input is parsed into smaller analyzable units to determine similarity to lexically stored information. An example of such parsing is provided by Linares et al. (2006), which demonstrates that elicited production of inflected nonce verb forms is based on phonetic similarity with an already existing model. This is an essential aspect of the model proposed in the current investigation as indicated by the analysis of the AM for velarinsertion in chapter 3.

Analogy of this sort is a highly efficient cognitive process, so the properties or

variables connecting input and model are not linear, but rather occur simultaneously at all

relevant levels (Johnson 1997). This allows activated patterns/associations to be

strengthened when they coincide in multiple exemplars. It also means that frequency is a

factor in establishing associations since increased propagation of a pattern is considered

to result in increased associative strength and salience/activation (cf. §4.3, Bybee 1985,

1996, 2007, Hall 2006, Martín Vegas 2007, Pierrehumbert 2001, Wanner 2006b).

Another consequence of acquiring structure or grammar through usage is a

reconceptualization of markedness:

[W]hen a child is learning the forms of his or her language, with every new in input, there is an attempt to relate it to something already stored. So whether plan or planning or hop or hopping comes in first doesn't especially matter, what matters is that given that one is already stored and one is new coming in, relationships of similarity, both phonological and semantic, are sought and when those relationships are found the lexicon takes on this kind of structure. (Bybee 1996:291)

The idea that one form is derived from the other only emerges in the sense that forms

differ in phonetic shape and meaning and that they are acquired in realtime. It is not possible, then, to suggest that one form is basic (or underlying) and all other related

20 forms are built upon it. Additional arguments against traditional markedness are treated in §4.4.

A second motivation for usagebased analogy is that it maximizes the above

associations as much as possible within a relatively local domain (i.e. the ‘homogeneous

supracontext’, below), though this does not imply that such associations are relevant.

Instead, it merely suggests that they are available to the speaker:

The exemplar model outlined here is an indirect model because categorization takes place by reference to items in memory that retain speaker information. That is, the frame of reference…is inherent in the set of exemplars, and the similarity calculation…limits the comparison to items in memory that are sufficiently close to the tobecategorized item. (Johnson 1997:149)

The localized analogical relationship is more efficient than generalizing across broader categories (Whittlesea 1987). This is because “speakers work with what is

locally salient” (Joseph 2004). 21 In other words, ‘restricting’ the associative domain of categorization reduces the processing load and restricts the analogical process to those forms which are relevant to input as opposed to simply comparing it to every possible form available. Consequently, morphological models are learned quickly through their repeated activation and use, while at the same time accounting for overgeneralization in acquisition (Federici & Pirrelli 1997). The retraction of overgeneralization is handled through the emergent, localized generalizations established over time and through competing analogies affecting the distribution of items in the lexicon (Wanner 2006a, cf.

Lloyd 1987).

The local domain of analogy is wellestablished in the literature (cf. Joseph 2009) and is related to various concepts including proximity (e.g. Burzio 2005, Skousen 1995), 21

Gradient Attraction (e.g. Burzio 2005), neighborhood effect (e.g. Pierrehumbert 2001), nearest neighbor (e.g. Federici & Pirrelli 1997), Perceptual Magnet Effect (Kuhl &

Iverson 1995), and selective attention (e.g. Nosofsky 1986, 1988, Nosofsky & Zaki

2002). It also implies that more generalized patterns are not stored but instead are emergent from the local process:

[]o sort of rule or prototypical representation needs to be generalized from the data and stored as a unit or entity separate from the data. Instead, generalizations exist within the stored lexical items themselves. Accordingly, linguistic processing is a matter of lexical access, and analogy to existing patterns found among the lexical items. (Eddington 2002:3)

In turn, this idea finds support in local generalizations and constellations (e.g. Joseph

1997a), family resemblances (Bybee &Moder 1983, Myers 1999), schema (e.g.

CroccoGaleas 1989, Bybee & Slobin 1982, Bybee 1996, Martín Vegas 2007, Myers

1999), frames (e.g. Minsky 1975), and the notions behind Concrete Minimalism (e.g.

Wanner 2006a, 2006b, Culicover & Nowak 2003).

It was indicated earlier that the necessary degree of similarity in localized analogy is highly variable and subjective. This openness and ‘flexibility’ is cognitively advantageous because highly variable input is handled without a heavily complex structure (Federici & Pirrelli 1997). Consequently, complete identity is unnecessary for input to be included in the same category as its AM. Rather, it is only crucial that the salient and relevant aspects of the associative relationship conform (cf. Burzio 2005).

This may result in broadening of a category through its local extension to new contexts.

The velarinsert data examined in chapter 3 illustrate this type of change in categorical structure. 22

Another benefit of the analogical process is that it accounts for variation and change in several ways due to its nonteleological nature. Variation is often attributed to the dynamic nature of linguistic categorization evident in the constant shifting of the prominent or salient identity relationships activated in the minds of speakers (Nosofsky

1986, 1988, Nosofsky & Zaki 2002). The natural drift, i.e. range of variation in the

distribution of tokens of an exemplar, also permits some differences in output

(Pierrehumbert 2001).

Similarly, input may be attracted to competing AM on the basis of distinct

commonalities (cf. Lloyd 1987, Wanner 2006a). According to Arbib & Hill (1988),

variation and change involve confidence factors which can be interpreted as the relative

associative strength between linguistic elements. This establishes a continuum of change

inherent to the current theory of AEM. This continuum and associative strength of

connection are precisely what allow variation or categorical change. For variation to

occur, a form may have two competing connections of relatively equal strength, thereby permitting variation. Alternatively, these connections may be reanalyzed based on new

input or changes in strength (possibly through imperfect memory, linguistic experience

over time, frequency and other factors affecting the cognitive process), resulting in

reanalysis.

In instances of both variation and change, the choice between one output and

another may be determined by randomselection (see below). This is an important aspect

of usagebased models:

The selection is random because of the kind of variability which is displayed in productions. If the production model always selected the 23

single best exemplar (by any measure), then the production goal would be invariant. In fact, however, the outcomes vary with variables at nonphonological levels (such as speech rate, , and speaking conditions). The aggregate effect of such variation as viewed from within the phonological model is random variation over the exemplar cloud…The mathematical nature of random sampling does of course entail that the location selected is more likely to be in a densely populated part of the exemplar cloud than in a sparse part. (Pierrehumbert 2002:9)

Randomselection, then is probabilistic and, interestingly, it implies no real distinction between competence and performance (cf. Ristad 1993). The analogical process is a

cognitive means of categorizing information, thus providing the foundation for the

selection of output. Support for this perspective comes from the fact that the

outputprobability of speakers has been shown to be the same as the inputprobability,

“with some variation for imperfect memory” (Skousen 1995:227, cf. Skousen 1989).

In spite of the imperfections associated with memory, usagebased analogy promotes lexicallyspecific linguistic (and extralinguistic) information in the cognitive

organization of language, creating a multidimensional associative space (cf. Joseph

1997a, Arbib & Hill 1988, Stemberger 1994, Federici & Pirrelli 1997, Pierrehumbert

2001, Bybee 2007, Ristad 1993, Eddington 2002, Goldinger 1998). This aspect of

usagebased theories is often criticized for being unrealistic since it is thought that the brain is only capable of storing so much information, i.e. the ‘headfillingup problem’

(cf. Johnson 1997). While this is certainly true, it does not mean that specific linguistic

experience cannot be captured. Goldinger (1998), Johnson (1997, 2005), Bybee (2002,

2007), and Myers (1999), among others, provide evidence of the retention of large

amounts of highly specific details from linguistic experience in memory. Additionally,

Tomasello (2001) suggests that acquisition is (at least) initially imitationbased, restricted 24 to localized domains and not easily generalizable. Such arguments lend themselves to an usagebased model that is crucially framed squarely within the mind of realistic speakers having imperfect memories.

A rulebased approach fares no better with respect to processing load and the

‘headfillingup problem’. Not only are speakers obliged to record information about particular patterns which trigger rule application, but they are also store vast amounts of

complex, often idiosyncratic information in their mental lexicon (Arbib & Hill 1988).

While the two approaches are similar in this respect, the rulebased model requires the

additional cognitive burden of storing generalized information through abstraction from

input. A usagebased analogical approach, conversely, handles the most regular and

irregular aspects of language without the need to separate them theoretically or psychologically. It also provides insight into the cognitive organization of language as speakers learn specific linguistic information through input and use it to properly contextualize new input.

This analogical means of organizing language implies the localized, direct emergence of generalizations and patterns without appealing to any formal heuristic (cf.

Joseph 1997a, Federici & Pirrelli 1997). These emergent, local patterns are used by the speaker in realtime, provided they are relatively stable in his/her grammar. Another way of contextualizing this process comes from Concrete Minimalism (Wanner 2006a; cf.

Culicover & Nowak 2003). Concrete Minimalism contends that as analogical events add up, a high number of extremely similar/identical items (with regard to a particular context) emerge, resulting in the development of higherorder organizational units,

25

"functioning in their limited, but concrete regularity as a type of rule…[and thus,] lexically anchored patterns" (Wanner 2006a:39). These ‘rules’ are robustly determined through induction, i.e. the enumeration of shared elements that are inferred to belong to the whole. They do not account for as much data as possible, but rather provide a cognitive solution for processing loads (Wanner 2006a).

Joseph (1997a) paints a poignant picture of the function of such generalizations:

[Generalizations] are as general as speakers allow them to be, and that can be very ungeneral or quite broadly general. There are rules, and there are regularities in language, but when one examines where these rules come from, it is often from the cumulative effect of particularized extensions from one lexical item to another. Since this ruleformation process is an ongoing one, a synchronic glimpse of a language is always going to capture the language with at least some incomplete generalizations. Thus if we as linguists are attempting to mirror speakers’ knowledge of their language through our , we should be prepared to have lessthanfullygeneral generalizations, and also subregularities that are defined on a very localized basis; in short, we should expect to find, and thus to have in our grammars, both local generalizations and constellations … (13)

Such incomplete generalizations and localized subregularities define the synchronic stages that are relevant in the diachronic process of Spanish velarinsertion examined in the following chapters.

In sum, the preceding points argue in favor of a usagebased theory of linguistic processing. By appealing to analogy as the primary instrument of categorization, it is possible to see how highly specific associations between forms are locally regular and dynamic in nature. This accounts for the emergence of restricted generalizations along with the variation and change that characterize language synchronically and diachronically. The following sections outline the usagebased analogical model relevant

26 to the subsequent investigation. This ‘immanent’ theory of Analogical Exemplar

Modeling (AEM) has its foundations in Exemplar Modeling (Johnson 1997,

Pierrehumbert 2001, Hall & Boomershine 2006, Boomershine 2005, Bybee 1996, 2007) as well as several aspects of the related Analogical Modeling of Language (Skousen

1992, 1995, 2002, Wanner 2006a, 2006b).

1.3.2—EXEMPLAR MODELING . Exemplar Modeling (EM) derives grammar from the associations made between stored tokens, or exemplars, that are accumulated through experience (Fondow 2007). It originates from a psychological model of similarity and classification or perception and categorization, i.e. an analogy between input and an experienced model (cf. ‘efficient witness’ per Ristad 1993). Each lexical item has a set of exemplars which are intrinsically linked by identity (an extreme type of similarity) of form, meaning and function. These exemplars, in turn, are linked to others, creating associations of similarity between lexical items across a multidimensional cognitive space. Such associations may be taken from any aspect of the stored information, e.g. , phonology, morphology, , , environmental details, social

specifications, etc. (cf. Johnson 1997). The accumulation of associations between

exemplars helps to categorize and organize the stored information into clusters or

‘exemplar clouds’. The associative relationship between forms is related to the ‘salience’

or ‘activation’ of a given exemplar or set of exemplars in the mental lexicon:

Given an item to be categorized, its auditory properties are compared with each exemplar's auditory properties, and the similarity between the item and each exemplar determines the activation level of the exemplar. If the match is good, the activation level of the exemplar is high. The sum of 27

activations over all of the exemplars of a category is taken as evidence that the unknown sound should be categorized as an instance of that category. (Johnson 1997:147; cf. Burzio 2005, CroccoGaleas 1989)

By the same token, the activation of stored information characterizes the notion of a

‘gangeffect’, implying that exemplars with the same behavior in a specific context are more likely to serve as a model for input (cf. Skousen 1995, Stemberger 1994).

Essentially, it is an example of strength in numbers that demonstrates the relevance of frequency effects in language (cf. §4.3).

Crucially, morphological structure emerges from the associative links between exemplars (cf. Wanner 2006a, 2006b). Dense exemplar clouds are those which share large amounts of information on all salient fronts. Exemplars within these clouds, in turn, are connected to those of other clouds according to established associations or

‘nodes’. If two exemplars are nearly identical in that they share more nodes than not, but are distinguished from each other by other nodes which, in turn, are linked across clouds because they share (at least) phonetic and/or semantic content with other exemplars, the consequence is morphological structure (cf. Stemberger 1994, Federici & Pirrelli 1997).

This is easily observed in inflectional paradigms where the same stem is shared intraparadigmatically and the inflectional features are shared interparadigmatically. In this way, the stem and inflectional markers result from cognitive organization, i.e. analogy.

As an example, in the present indicative of the MSp. verb cantar ‘sing’, the relationship between cells of the paradigm and corresponding semantic content identifies the stem cant and the inflectional markers o, as , a, amos , áis , an , of which 5 of 6

28 share the vowel a. On the basis of interparadigmatic associations with other present

indicative paradigms, e.g. hablar , comer , vivir , these (now a, e, i) can be

separated from the following content, i.e. s, mos , (i)s , n. The internal and external

relationships in conjunction with the semantic content determine the paradigmatic

structure of cantar , to the exclusion of other forms sharing a lesser number or different

set of features, e.g. the past participle cantado (nonfinite, nonpresent, no person/number), the imperfect cantaba (nonpresent, no 1sg o), the nominal form

canción (nonverbal, no person/number, tense, aspect, mood, no stemfinal [t], no theme vowel, shared linear elements of differing content, etc.).

On the basis of the preceding example, a crucial aspect of EM, and obviously

AEM, is the emergence of the inflectional paradigm as a cognitive unit of linguistic organization (cf. similar functioning of the paradigm in Penny 2002, Campbell 1974,

Rebrus & Törkenczy 2005, Aski 1995, Maiden 1992, 1993, 2001, 2003, Pirrelli 2000,

Pirrelli & Battista 2000a, 2000b, Vincent 1988, Harris 1973, Malkiel 1967, Menéndez

Pidal 1980, Joseph & Sims 2006, Joseph 2009). The paradigm is a higherorder exemplar cloud with very tightly knit, strongly associated forms (i.e. high density). It is a type of categorization that can then be linked to other, similar categories or clouds. 22

EM also has the advantage of incorporating the potential for change into the model based on differences in the input and usage of speakers, i.e. linguistic drift (cf.

Hall 2006, Pierrehumbert 2001). If a generation/speaker does not remember or use a particular form or set of forms or remembers it differently than another generation/speaker, there is potential for change. Essentially, there is a difference in the

29 edges of the exemplar clouds between these groups/individuals. The instantiation of gradual, local, thus traditionally incomplete change merely occurs in a lowlevel, highly defined, smallsized exemplar cloud, thereby redefining its regularity and propagation

(Wanner 2006b). For these reasons, it is an example of lexical diffusion (Wang 1969; cf.

Joseph 1997a, KraskaSzlenk 2007, Bybee 2002).

Relatedly, as the connections between exemplars are defined, a certain, lowlevel predictive power arises. According to Bybee (1996) and Pierrehumbert (2001), one of the problems with rulebased models is that they do not make predictions about which patterns are productive and which are not. Because regularity equates to productivity, regular rules apply by default, including in cases where the inflected form is missing, forgotten or unknown (Bybee 1996; cf. Martín Vegas 2007). While rules do make partial use of productivity, e.g. by way of ordering, they do not explain it. EM, in contrast, uses realtime, speakerdetermined, salient associations to categorize input (cf. Skousen 2002).

Logically then, these same devices are used to deduce potential directions for language change, though in no way enforcing such changes.

To summarize, EM is a surfacebased theory that depends crucially on associative relationships between elements in the mental lexicon. These associations create linguistic structure, i.e. grammar, through categorization of input according to the variable(s) shared with previously stored information. The detail involved in identifying and analyzing these variables is an essential part of what sets this type of analogy apart from traditional types. It is highly defined and able to be examined scientifically, combining the fine detail of individual exemplars with broader linguistic generalization and

30 categorization. Overall, this involves weak, local patterning since the process is nonglobal. As a consequence, there is increased accuracy (i.e. there are no exceptions since there are no systematic rules), and the dynamic nature of language is explained

(Federici & Pirrelli 1997; cf. Wanner 2006a). Thus, EM is an elegant, psychologically realistic, and immanent means of processing linguistic information, involving multiple levels of analysis which finds support in other cognitive processes that make use of the same general analogical method for categorization: “[T]he general cognitive capabilities of the brain, which allow it to categorize and sort for identity, similarity, and difference, go to work on the language events a person encounters, categorizing and entering in memory the experiences” (Bybee 2007:711).

1.3.3—ANALOGICAL MODELING OF LANGUAGE IN ANALOGICAL EXEMPLAR

MODELING . The theoretical model proposed in the current investigation is founded upon

EM as described in the preceding section. However, additional theoretical properties are required to correctly define and restrict the function of analogy. These elements are derived from the Analogical Modeling of Language (AL) (Skousen 1992, 1995, 2002,

Eddington 2002, Wanner 2006a, 2006b). The combinatory theory that emerges and is promoted in the current investigation, Analogical Exemplar Modeling (AEM), sufficiently deconstructs the analogical process so as to give it a solid, scientific, psychological foundation that is not only welldefined but also elegant. AEM is more specifically based on the theoretical devices described in Wanner (2006b), which include,

"...a reduced cognitive specificity for language, limited to the immediate needs of the

31 linguistic individual (immanence), the guarantee of learnable categorization in an untutored environment (Concrete Minimalism), and the feasibility of giving analogical projections some verifiable content (Analogical Modeling)" (116). However, AEM refines the conceptualization of the AM and the crucial variables that define it.

AL is an exemplarbased probabilistic theory which analyzes the potential for analogical change (cf. Fondow 2007). It was devised to counter the traditional use of analogy which, problematically, lacked explicitness and definition, since any linguistic element could potentially serve as an AM (Skousen 2002). In AL, an analogical set or

‘model’ (AM), representing some level of exemplar cloud, is determined according to the relevant characteristics or ‘variables’ shared between the model and a target form. Once these variables have been identified, they are used to determine the potential for and possible directions of change in the target. Essentially, the more variables that are shared between AM and target, the more likely it is that the target will obtain the outcome provided by the model (a ‘gangeffect’, see above). In spite of the relationship between exemplars, analogical influence in this model is merely probabilistic because change cannot be guaranteed (cf. Pierrehumbert 2001, Bybee 2007, Baayen 2003, Bod et al.

2003, Jurafsky 2003). Thus, such a theory “limits the likelihood of change through the inertia stemming from accumulated experience that guarantees a functional continuation in form and use of language” (Wanner 2006b:262).

This inertia is a type of stability in language that crucially distinguishes the current view of analogy from the traditional definition. Another means of stability is obtained by the existence of competing analogical pressures which may take hold in a

32 local context preventing the systematic extension of change. Realistically, the activation of, or resistance to change are the same process, since analogy may apply to forms thus preventing change in the same way that it results in change in other instances. This

implication arises from the previous discussion that: (1) analogy is unpredictably

triggered and may not occur systematically in identical contexts; (2) analogy is locally

accessible, affecting only the target form(s); (3) it is only weakly extended beyond these

local targets, consequently, it is nonteleological (and frequently nonsystematic); (4) the

individual associations of similarity occur regardless of the effect they may have on the

rest of the system.

AL avoids the global, overapplied declarative rule system as well as the

complexities and potential underapplication of a universal set of constraints. It is a robust

system able to cope with missing or malformed information (cf. Ohala 2005). 23 In other words, fuzzy boundaries and leakage are anticipated and unproblematic (Skousen 2002).

This robustness is apparent in the amount of (often redundant) detail incorporated in the mental lexicon:

Redundancy affects computational efficiency in the two domains of processing and storage. The cost of redundancy in stored items is not high, since human cognitive capactiy has enormous quantities of memory space, including highly efficient retrieval procedures. Since memorized items including redundant information have many more crossclassifiying connections with other similar items than bare representations, access during processing may proceed from more than one angle, thus spreading the chances of possible identification over two or more clues, rather than a single, minimal, required trait. If the single nonredundant required identificational feature is absent from the message received, the strictly formal analysis should not be able to access this item, and there will be a missing link with more or less grave consequences for the overall processing of the utterance. Formal accuracy depends on highly specific information, while a more approximate processing can operate with 33

lessthancomplete information and valid lateral connections to reach the same result. Redundancy becomes a guarantor of efficiency under nonoptimal conditions of communication. (Wanner 2006b:102)

At the same time, the AM is open to reanalysis, to continuous shifting in order to best categorize the exemplars pertaining to it and the variables that define it.

Such openness in the AM gives the appearance of generalization in the directions of both systematic rules and idiosyncratic processes. AEM takes advantage of this weakened distinction by creating a languagespecific continuum of generalization which permits localized phenomena and apparent regularity (and subregularity) (Wanner 2006a, 2006b).

Another crucial part of AL is that it does not always rely on the nearest neighbor to a target (cntr. traditional EM). Instead, the AM is required to fall in the ‘homogeneous supracontext’, the contextual space in which all exemplars are assigned an outcome according to the relevant variables. The goal is to predict an outcome where none is available, due to innovation (new words, borrowing, etc.), reanalysis (accomodation, generational drift, etc.), or loss (imperfect memory). According to Skousen (2002), the

‘variables’ or associative features of the stored exemplars that serve as the AM comprise the ‘given context’ in which a prediction is made with respect to input. The

‘supracontext’ is defined as the group of all possible combinations of variables, including the absence of all such variables (equating to 2 n supracontexts where n is the number of variables). ‘Homogeneity’ applies to the set of supracontexts in which the outcome is the same or there are no intervening (more specific) supracontexts that contain any realized exemplars (because they would be the homogeneous supracontext or the threshold of

34 analysis). This implies the inclusion of more specific supracontexts within more general ones and the exclusion of ‘heterogeneous’ datasets.

While AL has an even greater statistical basis than indicated here, frequency and the formula for determining categorization and output through randomselection are the only statistical aspects of AL required in AEM. 24 These two calculations are essential to the highly defined probabilistic nature of analogy and the salience and activation of an

AM in the mind of the speaker. In other words, frequency and probability are used to determine the likelihood with which each potential outcome may occur in a given context. This is indicative of subjective realtime decisionmaking by the linguistic individual (cf. Jurafsky 2003, Pierrehumbert 2001). It also provides an explanation for synchronic and diachronic variation in language (e.g. between velar and nonvelar verb forms found in Old Spanish, see ch. 2) as well as the relative stability of language.

Variation occurs in AEM when the supracontext is homogeneous but not deterministic, i.e. when more than one outcome occurs and the subcontexts either show the same outcomes or are empty. This may take place at different stages of linguistic experience of an individual or a group of individuals. When it does occur, AEM highlights ‘random selection’ (over the alternative ‘selection by plurality’), where “the probability of choosing a particular homogeneous supracontext is the square of the frequency of that supracontext” (Skousen 2003:6; cf. §4.3.1). Stability, meanwhile, is due to the vast amounts of information stored in memory that characterize usagebased approaches:

If this process yields good results for complex analogical events in adult language practice, it must at the same time and a fortiori also serve to reaffirm the given shape of existing forms in the speaker's usage. Each form used in practice is in principle predicted and controlled by such 35

analogy, firming up the given shape, or bending it in direction of the forces displayed in the motivating data set. (Wanner 2006a:23)

The incorporation of probability into AEM is also indicative of an ‘immanent’ view of language (Wanner 2006a, 2006b, see below). The probability for change is determined through natural statistics involving the token frequency of the forms in the

AM. This represents the subjectiveness of learning and experience for individuals, thereby explaining the differences in language across speakers and speech communities as well as the effects of imperfect memory, attention factors, limits on cognitive processing loads, etc. (Wanner 2006a). So, while a target may have all the variables associated with a particular AM, it is not required to activate analogical change (as long as no change is a potential outcome).

Consider, for example, the data in Appendix K. These Modern are susceptible to the analogical process of velarinsertion described in the following chapters on the basis of the variables they share with the AM. However, they are not part of the velarinsert class. This is because of the nonteleological, probabilistic nature of

AEM. It also illustrates the localized nature of analogy and the relevance of the linguistic individual in language change. These potential analogical targets are examined in more detail in chapter 4.

Provided that the variables relevant to the AM necessary to determine the outcome of a particular context are determined locally, a crucial part of properly defining the analogical process is to identify these variables. In this respect, AEM refines the method of AL, taking a broader approach more in line with the processing of information that characterizes other usagebased models. AEM follows the lead of Wanner (2006a), 36 which proposes to incorporate linguistic and, potentially, nonlinguistic variables in the composition of the AM in addition to those defined in purely phonological terms. The

AM proposed for the current investigation, as detailed in chapter 3, provides evidence in favor of this hypothesis.

A final word of caution is needed regarding AEM and regular sound change. The stance taken here is not incompatible with the traditional view of regular sound change, in the ‘softer’ sense that it is a type of localized analogy that has been extended to all applicable targets. Where sound change occurs systematically across all target forms having the relevant conditioning environment, it is based on an analogical relationship between model and target that is exclusively phonological and which applies to a sufficiently high number of exemplars that no resistance is met as the result of competing models (cf. Hock 1991). The ‘strong’ view of regular sound change suggests a wholesale, nonlocalized extension of a pattern (e.g King 1969). Although this extreme version is a foundation of linguistic investigation, the regularity hypothesis is complicated by those elements which do not fall within its domain, i.e. the irregular, subregular, suppletive, borrowed, etc. These elements provide evidence against regularity as a principle or law governing language, though it is useful in more general, broader analyses. The 'soft' approach taken here, based upon analogy, is preferable since it incorporates these 'marginal' factors into the linguistic equation.

Analogy is a constant process, applying when needed as determined by the speaker and creating a dynamic grammar which evolves both internally and externally

(Wanner 2006a). Regular, exceptionless, rulelike change is merely one (extreme) end of

37 the continuum of analogical change (cf. Hock 2005). It is very common, however, to find changes that are not complete, idiosyncratic or otherwise marred by exceptions.

Completion of analogical change as described herein is the end of change through its application to all local potential target forms which are not affected by a competing AM or by a strong, isolated (or unrelated) lexical record. In this sense, the immanent (‘soft’) version of completion of change differs from the teleological, global completion proposed in other theories.

Similarly then, this suggests that the analogical usagebased model does not contradict a rulebased approach:

[O]ne way of interpreting the notion of ‘rule’ is that it is a summation, a kind of shorthand so to speak, for a series of analogies; in a sense, analogies define the patterns, the regularities, that we tend to interpret as ‘rules’. Generalizations, in such a view, become really a matter of retrospectively summing up a set of analogical extensions from a model to a novel ‘target’; after enough such extensions, there is the appearance of rulegoverned behavior, but even the novel formations could just be ongoing examples of the analogical extension process. (Joseph 1997a:9; cf. Arbib & Hill 1988, Stemberger 1994)

It is important to note that the rulebased conceptualization of grammar, when not understood as representative of the actual or ‘immanent’ cognitive process, may be seen as emergent from one in which analogy is central.

1.3.4—IMMANENCE . Malkiel (1967) indicates that speakers have a “semiconscious” involvement in the direction of change. In other words, speakers’ realtime decisions provide the foundation for acquisition, processing and production. Language is the interaction of input (experience and environment), innate ability (analogical associations,

38 cognitive and physiological ability to use language) and the anticipation of communication between speakers (social awareness).

AEM therefore assumes that the speaker has a central or ‘immanent’ role in language change (Wanner 2006a, 2006b; cf. Myers 1999). Immanence involves the internal, and therefore subjective, part played by the speaker in the development of language within the context of their linguistic environment. This social, communicative environment provides the language learner with imperfect, incomplete, nonsystematic and varied input over time. Thus, these same characteristics are expected in and across the linguistic reality or grammar of speakers in time (e.g. imperfect memory, idiosyncratic differences in the acquisition of lexical items, time effects, etc.). 25

In simpler terms, the process of analogy is completely subjective, determined, usually subconsciously, by the speaker in their linguistic environment. It is governed by human processing loads, memory and experience and is controlled by the principles of human communication (e.g. , sociolinguistic factors, etc.). The variation and differences found in the velarinsert class, both synchronically and diachronically, as well as within and across speakers exemplify this immanent view.

1.4—CONCLUSIONS AND OVERVIEW OF THE FOLLOWING INVESTIGATION . In sum,

AEM is a refined version of EM that incorporates the concepts of immanence, the

homogeneous supracontext and probabilistic, randomselection from AL. The combined

efforts of EM and AL in the guise of AEM is necessary to properly formalize the

analogical process. Consequently, the theoretical implications treated above that are

39 assumed under EM and AL are also assumed in AEM. The emergence of structure through local analogical relationships crucially includes the function of the paradigm as a cognitive unit of linguistic organization. These associative relationships are multilayered, suggesting that the analogical process involves the combination of numerous variables, i.e. ‘multiple causation’ per Malkiel (1967, cf. Joseph 1982, 1983).

Variation is incorporated into the system both synchronically and diachronically through probability and the influence of competing AM. The dynamic, immanent nature of the analogical process is also essential, since it accounts for changes in the categorization of elements in the mental lexicon. At the same time, the potential for no change is also inherent. All of these points are essential to the developments affecting Spanish velarinsertion as analyzed in the current investigation.

The next chapter provides an indepth description of previous examinations of velarinsertion, highlighting the most relevant and crucial internal and external changes involved in the emergence of the Spanish subclass. It is concluded that while the origins of the velarinsert have been thoroughly investigated, no real conclusion has been reached regarding the primary AM. At the same time, the analogical extension of the pattern has been almost completely ignored, especially with respect to the most recent stages affecting this verb class.

Chapter 3 proposes a detailed analysis of the multiple ‘causes’ or variables associated with the AM in Spanish, focusing on the relevance of the paradigmatic structure. Afterwards, individual synchronic stages are examined as they highlight the velar/nonvelar variation of numerous forms as well as the analogical extension and

40 receding of the velar to some lateral, rhotic, and yodfinal verb stems. The localized nature of the analogical process requires such a finegrained, multistage approach. An example of the immanent role of the speaker in such linguistic developments is also provided. Finally, the discussion turns to dialectal variation that characterizes the modern situation.

Chapter 4 treats various theoretical and specific issues. The complexities associated with the analogical processing of linguistic information, the relevance of paradigmatic structure in language change, frequency effects, markedness and the potential for continued change are all explored as they pertain to AEM and the analogical process of velarinsertion in Spanish.

41

CHAPTER 2: THE HISTORY OF VELAR -INSERTION

2.1—THE VELAR -INSERT IN SPANISH . Throughout the recorded history of the Spanish

language, the inflected paradigms of a particular subset of verbs has been characterized by the presence of a voiced velar /g/ in the first person singular present indicative (1s

PRS IND) and in all forms of the present subjunctive (PRS SBJ). The developments

affecting this verbal subclass are the focus of this investigation. The following briefly

describes the Spanish verbal system and details the particular group of verbs under

analysis. Subsequent sections examine more thoroughly those processes that play a part

in the emergence and extension of this velar element.

2.1.1—THE SPANISH VERBAL SYSTEM . The Spanish verbal system is characterized by three regular conjugational classes organized according to the theme vowel found in a number of their inflected forms. Besides its appearance in such finite forms, the theme vowel (TV) is also found in the infinitive (INF) which is used as the default lexical entry for individual verbs in dictionaries, grammars and pedagogical literature. The three conjugational classes are –ar (I), er (II) and –ir (III), where –a, e, and –i are TV and

represents INF. Examples of the present tenses, indicative and subjunctive moods, of these regular conjugations are found in Appendix A. The term regular indicates here systematic use of the inflectional endings associated with a particular conjugational class 42 and, most importantly, the absence of allomorphy in inflected forms. 26 Irregular verbs, meanwhile, are marked by the appearance of paradigmatic allomorphy in stem, , or both. 27 Crucially, irregularity is “understood as a nonbiunique relationship between meaning and form, [that] is an autonomous abstract property of morphological paradigms which influences the implementation not only of phonetic changes, but also of purely morphological innovations” (Maiden 1992:289290). Regardless of (ir)regularity, the basic structure of an inflected verb form, takes the following shape: STEM + (TV) +

(TENSE/MOOD) + PERSON/NUMBER (N.B. those inflectional markers in parentheses do not appear in all finite forms).

The verbal system is also organized according to other criteria related, in part, to conjugational class. First and foremost, the –ar conjugation is the only generally productive class in the and it is defined as being the most regular:

The first conjugation is historically distinguished from the other two by the relative invariance of the verbal root. In Italo and IberoRomance, there are no consonantal alternations in this conjugation, and none of the idiosyncratic root ablaut and consonantal alternation between the preterite and other tenses, which, in ItaloRomance in particular, characterizes most second, and many third, conjugation verbs (Maiden 1992:291).

The second and third conjugations are characterized by increased irregularity with respect to the former; in fact, such irregularities create a broad subcategorization of the verbal system (i.e. ‘paradigmatic rootvariability’). There is a high degree of overlap between the –er and –ir conjugations due, among other things, to identical frontness (and realization) of their TV and SBJ marker in certain inflected forms. Such similarities distinguish these conjugations from the first and are the result of historical consequences.

The salience of such shared variables between conjugations II and III, to the exclusion of 43 conjugation I, are of extreme importance in defining the AM proposed in this investigation.

2.1.2—THE SPANISH VELAR -INSERT SUBCLASS . Modern Spanish has an ‘irregular’ class

of velarinsert verbs (also known as ‘–go verbs’ in pedagogical literature) characterized,

as indicated above, by a nonetymological voiced velar obstruent /g/ in 1s/SBJ forms (cf.

‘Lpattern’ distribution, per Maiden 2001). 28 With respect to the subcategorization of the verbal system described in §2.1.1, it should be noted that all such verbs belong to the –er and –ir conjugations. The Modern Spanish verbs under analysis are listed in Appendix

B.1 and their complete PRS IND and PRS SBJ paradigms are found in Appendix C. In addition, each of these verbs may be marked by prefixation, thereby multiplying the number of members in this verbal subcategory. The prefixed verbs are included as

Appendix B.2. The distribution of the velar across the 1s/SBJ is exemplified by the verb poner “put, place” in Figure 2.1:

MSp Class INF Gloss IND SBJ # P #

2 poner put, place pongo ponga 1 1s pones pongas 2 2s pone ponga 3 3s ponemos pongamos 4 1p ponéis pongáis 5 2p ponen pongan 6 3p Figure 2.1—Distribution of the velarinsert, as seen in poner “put, place”. 29

44

The small set of second and third conjugation verbs that are included in

Appendices B and C are a mere fraction of these conjugational classes as a whole. In addition, the velarinsert class is deemed ‘irregular’ because membership in the class is not predictable by any easily identifiable criteria. As seen in §4.5.4, there are, in fact, a number of verbs which appear to be potential targets for velarinsertion and yet remain unaffected by it.

In opposition to the appearance of a nonetymological velar element in the verbs above, there are at least three general etymological sources for Spanish /g/, affecting both verbs and nonverbs: (1) Learned or semilearned forms which contained /g/, e.g. agricultura “agriculture”, extinguir “extinguish”; (2) , generally in intervocalic contexts, a process characterized phonologically by the simplification of geminates (i.e.

/kk/ > /k/), the voicing of voiceless (i.e. /k/ > /g/), e.g. CL PACĀRE > MSp. pagar “pay”; and (3) the nonsystematic elimination of earlier voiced obstruents (i.e. /g/

> Ø), e.g. CL A(U)GURIUM > MSp. agüero “omen”, cntr. CL DIGITUM > MSp. dedo

“finger”. 30

The importance of the nonetymological nature of the velar obstruent in the class of verbs under investigation should not be taken lightly. The broadest descriptions of the velarinsert class include verbs which derive the velar from a of etymological and nonetymological sources, including those mentioned above (cf. a much broader use of the term in Wilkinson 19781983). The verbs in Appendix B.1 historically did not have a velar obstruent in any of their Latin forms, nor can it be derived by phonological rule.

This situation is made clear in Figure 2.2, which contrasts the PRS paradigms of Classical

45

Latin TENĒRE “have, possess” (CL conjugation II) with same verb in Modern Spanish

(i.e. tener ).

Lg INF Gloss IND SBJ # P #

CL TENĒRE have, possess TENEŌ TENEAM 1 1s TENĒS TENEĀS 2 2s TENET TENEAT 3 3s TENĒMUS TENEĀMUS 4 1p TENĒTIS TENEĀTIS 5 2p TENENT TENEANT 6 3p

MSp tener have, possess tengo tenga 1 1s tienes tengas 2 2s tiene tenga 3 3s tenemos tengamos 4 1p tenéis tengáis 5 2p tienen tengan 6 3p Figure 2.2—CL TENĒRE and MSp. tener .

From this example, it can be seen that the forms that have undergone

velarinsertion in Spanish are exactly those Classical Latin forms where a front theme

vowel, –E, Ē, I, Ī, immediately precedes a non, A or –Ō, the latter pair functioning as morphological markers of SBJ and 1s respectively. In spite of this fairly detailed description of the apparent conditioning environment, the velar element does not appear in all verbs which contain the above morphophonological criteria, thus explaining the ‘irregularity’ described previously.

Another motivation for the ‘irregular’ treatment of velarinsert verbs involves the numerous diachronic changes affecting the class. In the earliest available Old Spanish texts, only tener “have, possess”, venir “come” and poner “put, place” showed

46 velarinsertion. 31 It is believed that fairly soon afterwards, this set came to include other

verbs with rootfinal /n/, e.g. the now defunct OSp. remanir “remain, stay”, remanga .32

As this class increased in membership, velarinsertion became analogically productive,

spreading, over an extensive period of time, to stemfinal /l/, /ɾ/, /j/ and /s/. Some of the

verbs affected by the process of velarinsertion entered the class completely, while others

continued to show synchronic variation between velar and nonvelar forms. This productivity, however, was shortlived, as the forms which were “on the fence” so to

speak, i.e. those which showed an allomorphic alternation between velar and nonvelar

verb stems, tended in large part to disassociate themselves from this lexical class (exc. salir “leave”, below). Most variation, e.g. soler “be accustomed to”, 1s suelo ~ suelgo; ferir (MSp. herir ) “wound, injure”, 1s fiero ~ fiergo , was leveled in favor of the

nonvelar allomorph in the , cf. MSp. suelo , hiero .

The diachronic shifts in class membership allow categorization of the velarinsert

verbs into three primary groups: (1) those which have maintained membership since

entering the class (as seen in the first section of Appendix B.1); (2) a subgroup of the previous set, comprised of only remanir , which has generally fallen out of use in the modern language (Appendix D); and (3) a set affected by velarinsertion only in synchronic variation with nonvelar forms (Appendix E). Some of the verbs in the third group also attest the velar in limited modern dialectal domains (cf. the second half of

Appendix B.1). Included in (1) and enumerated in Appendix B.2 are verbs derived by prefixation, e.g. detener “detain”, prevenir “prepare; foresee”. Although not discussed in detail, this does not suggest that their appearance is irrelevant (cf. discussion of token and

47 type frequency, §4.3). In general, these verbs are subsumed under their respective etymons because they follow an identical path of development regarding velarinsertion.

With respect to the verbs showing multistage and/or modern variation in (3), the appearance of the velar forms may be explained in one of two ways, either as a consequence of continuously limited variation over time in very localized speech areas

(more likely to apply to many of the verbs in the second section of Appendix B.1), or as an indicator of innovation in the velarinsert category, i.e. verbs serving as potential (or renewed) targets for the velarinsertion process.33

In spite of the motivations for classifying the velarinsert set of verbs as

‘irregular’, there are notable shared morphophonological variables which are pertinent to

the distribution of the velar and consequently, the AM for velarinsertion. First, it has been noted that all of the verbs described above undergo velarinsertion between a

rootfinal alveolar sonorant /n, l, ɾ/, sibilant /s/ or glide /j/ and a nonthematic

inflectional nonfront vowel /a, o/, e.g. 1s pon go, 2s SBJ pon gas . In the affected paradigmatic cells, the (front) theme vowel /e, i/ which denotes conjugation class (II or

III), is also absent. This contrasts with its presence in the Classical Latin forms and the

remainder of the Spanish PRS IND, e.g. 2s IND pon . Such similarities are readdressed below, alongside a handful of other factors. A complete description of the AM is provided in §3.2.

2.1.3—OVERVIEW OF CHAPTER 2. The remainder of this chapter examines the existing literature on the velarinsert class. First, a brief discussion of the situation across the

48

Romance languages is provided to better outline those languages which share some degree of velarinsertion and to distinguish them from those which do not. Then, the changes that mark the transition from Latin to Romance are treated, including the development of yod and its subsequent effects in the verbal paradigm. Particular attention is paid to the processes that are crucially related to velarinsertion, including those affecting the set of Classical Latin –NGERE verbs. Next, a brief excursus is made

with respect to the diachronic evolution of Modern Spanish verbs like decir “say” and

hacer “do, make” which are traditionally considered important AM for velarinsertion.

In §2.4, support is provided for appealing to analogy as the primary explanation for the

emergence of the velarinsert verb class. Afterwards, the examination turns to the

hypothesized AM that provide the foundation for velarinsertion. The most salient and

wellaccepted of these theories are subsequently analyzed in greater depth to highlight the

major differences and similarities across the literature and to identify their contributions,

as well as possible shortcomings, with respect to a complete diachronic analysis of the process. Brief treatment of both the Catalan and Italian situations is included, since the

literature concerning these languages overlaps extensively with Spanish. Finally, the

conclusions drawn from previous analyses of the velarinsert verbal subclass are

summarized as a means of introducing the analysis found in the following chapter.

It is determined that previous analyses of velarinsertion only explore the process

in broad strokes without providing the necessary amount of synchronic detail or a proper

definition of the AM. It is argued here that the AM involves a dynamic set of variables

diachronically which is relatively stable at each relevant synchronic stage. Of these

49 variables, it is the emergence of the inflectional paradigm as means of linguistic organization that contributes most heavily to the current investigation. The points are presented more thoroughly in the following chapter.

34 2.2—VELAR -INSERTION ACROSS ROMANCE . Given the fact that the origin of the

velarinsert element in Spanish is not transparently derived from any particular

languagespecific sound change, it is important to consider the situation across the

Romance languages. Appendix F shows the panRomance conjugational patterns

corresponding to those of the primary velarinsert verbs in Spanish. Several important points can be extracted from the distribution of the velar in these languages.

First, it is relevant that the velar does not appear in the geographic periphery, i.e. the eastern and westernmost languages, Romanian and Portuguese respectively. Nor is it found in certain (see below) or in most (N.B. In Old

French forms of the sort vieng , ng represent a palatal nasal). 35

The velar is most common and wellestablished in Spanish (also Leonese and

Aragonese), Catalan (along with Occitan, Provençal and Gascon) and Italian. These

languages together represent a geographically contiguous region. This lends support for

the hypothesis that the development of this verb class occurred prior to their autonomy.

At the same time, there are partial similarities in the traditionally hypothesized AM for

each language. Nonetheless, the existence of notable differences in the paradigmatic

distribution and lexical extension of the velar as well as in the forms and variables

50 comprising the AM suggests the independent development of the velar in these languages. 36 These issues are discussed below.

The relationship between Spanish and the other Romance languages also provides insight into the potential directions for language change and the cognitive organizational function of the paradigm. At the same time, a consistent pattern of paradigmatic organization emerges within and across the PRS tense of the Romance languages regardless of whether or not they are affected by velarinsertion. The paradigm is analyzed more broadly in chapter 4.

2.2.1—PORTUGUESE . Appendix G lists the Portuguese verbs which correspond to the

Standard Modern Spanish velarinsert verbs. Portuguese serves as an important point of contrast against the analogical developments affecting Spanish, Italian and Catalan because it is not affected by a process of velarinsertion. The primary factors for this are the ‘yod effects’, or sound changes initiated by yod (§2.3.2 below), and the lack of the necessary relationship between an existing AM and set of target forms. In Portuguese, the ‘targets’ for velarinsertion were instead primarily affected by the regular process of palatalization by yod, e.g. CL TENEŌ > MEPtg. tenho [teɲu]. There is also no analogical association between forms like ter , vir and the traditionally hypothesized AM for velarinsertion such as dizer , fazer , cingir , conhecer , etc. The changes affecting these groups of verbs did not overlap in such a way as to promote the velarinsert subclass, therefore the Portuguese developments do not lend direct insight into the appearance of the velar element in Spanish:

51

[T]he only we can be sure of is that [the Spanish] were exposed to temptation to substitute /ŋg/ for /ɲ/ which the Portuguese were not exposed to. In the case of Spanish, too, we may observe that by adopting tengo , tener joined the class of verbs which differentiates the first person from the other persons by use of a velar, this being virtually the only differentiating device now used in Spanish. (Wilkinson 1978:29)

Relatedly, the extension of the velarinsert to any lateralfinal verb stems is prohibited by a regular sound change that, in most of these verbs, eliminates or replaces

the lateral segment with yod, e.g. CL SALĪRE > MEPtg. sair , 1s saio . The later extensions

of the velarinsert class to verb stems ending in /l, ɾ, s, j/ in the aforementioned languages

are supported by the presence of an already established AM. In the Portuguese case

however, no precedence was set for the velar in such forms and so it did not arise as

characteristic of its verbal system.

The regular sound changes affecting Portuguese verbs are typical of the language’s conservative character with respect to its Latin origins. In this way, it contrasts with Spanish which shows analogical influence in the emergence of the velarinsert subclass. Portuguese, then, is representative of a series of changes that resemble those that might have affected this Spanish verbal subclass if velarinsertion had not. This is a basis for the discussion of potential directions of change in §4.5.

Another important feature analyzed in chapter 4 that concerns the relationship between Spanish and Portuguese, and the Romance languages in general, is the emergence of similarities and patterns across their PRS paradigms. As can be seen by the highlighted portions of the PRS paradigm of the Portuguese verbs found in Appendix G, the alternation in the shape of the verb stem takes a pattern that is identical to the one found in the velarinsert class of Spanish. However, it is important to note that the 52 emergence of this pattern in each language is the result of very different developments.

In the case of Spanish, it is due to the analogical process investigated herein, while for

Portuguese, it is primarily a consequence of regular sound change affecting individual forms of the PRS paradigm.

2.2.2—CATALAN . The velarinsertion found in Catalan is the due to the convergence of factors within a highly complex AM, one which has a number of crucially languagespecific components from both within and outside of the localized inflectional paradigm. The Catalan verbs corresponding to the Standard Modern Spanish velarinserts are enumerated in Appendix H. On the mere basis of the PRS paradigms provided, it would seem that the Catalan situation is identical to the Spanish one. In fact,

they share, along with Italian, several primary variables in their AM. However, the

1s/SBJ pattern (with systematic wordfinal devoicing producing 1s /k/) is found in an

overwhelming number of both standard and dialectal Catalan verbs, especially in

comparison to the very reduced set found in Modern Spanish. A short list of examples is provided in Figure 2.3 (cf. §4.2.3).

53

MCat INF Gloss IND SBJ # P #

creure believe crec cregui 1 1s creus creguis 2 2s creu cregui 3 3s creiem creguem 4 1p creieu cregueu 5 2p creuen creguin 6 3p

estar be estic estigui 1 1s estàs estiguis 2 2s està estigui 3 3s estem estiguem 4 1p esteu estigueu 5 2p estan estiguin 6 3p

moldre grind molc molgui 1 1s mols molguis 2 2s mol molgui 3 3s molem molguem 4 1p moleu molgueu 5 2p molen molguin 6 3p

vendre sell venc vengui 1 1s vens venguis 2 2s ven vengui 3 3s venem venguem 4 1p veneu vengueu 5 2p venen venguin 6 3p Figure 2.3—Examples of Extension of VelarInsertion in MCat.

The preceding samples demonstrate that velarinsertion in Catalan has extended well beyond the domain that restricts the Spanish and Italian developments. Not only is the velar found in the same verbs as in Spanish, including both the Old Spanish forms that have fallen out of modern usage as well as the modern dialectal variants, but it has also reached the most frequent of verbs including both forms of “be”, ésser , 1s sóc , and

54 estar , 1s estic . In nonstandardized dialects, its territory is even greater and is very telling with respect to the potential targets for velarinsertion in Spanish (cf. §4.5.2).

In spite of providing information regarding possible directions for change, the

Catalan situation differs from the Spanish one in certain very crucial ways. It has gained momentum in part thanks to the additional variables incorporated into its AM, not existing in Spanish, which support the appearance of the velar element. The stems of the

PRET IND, IMPF SBJ and some PART of many of the verbs showing PRS tense velarinsertion are also marked by a /g/ that is derived (initially) from regular sound changes. It is thought to have spread across tense to the PRS SBJ (from the PRET IND to the IMPF SBJ), then across mood to the PRS IND (Wilkinson 1978). In addition to this internal influence, there appears to have been a certain tipping point after which the analogical process of velarinsertion became associated more broadly with entire conjugational classes in Catalan, e.g. verbs whose INF ends in –ure (caure “fall”) or –dre

(vendre “sell”) (see ch. 4; cf. Wilkinson 1982, Costanzo 2006, 2007, forthcoming).

Accordingly, the analogical extension of the velarinsert in Catalan proves useful in examination of the Spanish subclass. While the objectives of this investigation do not permit an independent and highly detailed diachronic and dialectal analysis of the Catalan situation, it is mentioned where relevant. It is readdressed in the section on the potential directions of analogical change found in chapter 4.

2.2.3—ITALIAN . Aside from those forms found in Appendix F, Standard Italian and

Spanish show a number of similarities with regard to the appearance and distribution of

55 the velarinsert (cf. Appendix I). For the most part, these similiarities are coincidental, thus it is believed that the velarinsertion processes in Spanish and Italian are unique (cf.

Fanciullo 1998, Malkiel 1974). Just as in the Catalan case, the Italian process has its own locallydefined AM. The influence of the set of CL stems ending in –NG , RG , and,

most importantly, LG is much greater than in Spanish due to greater type frequency in conjunction with the languagespecific sound changes which affected these sets of verbs.

Therefore, the extension of the velar element differs between Spanish and Italian from the outset, in both the verbs affected and the appearance of the velar in one quarter of the

PRS paradigm (due to the frontness of the inflectional vowels marking particular cells, e.g. MIt. 3p IND –ono , and the distribution of palatal consonants given the effect of front vowels, e.g. MIt. 1p/2p SBJ –iamo /iate , see below). The AM for Spanish, meanwhile, is thought by many to have its roots in yod effects and the voicing of intervocalic obstruents

(i.e. Western Romance Lenition) which did not Italian, compare MSp. 1s digo , hago and MIt. 1s dico, faccio < CL DĪCŌ , FACIŌ . The synchronic and diachronic

differences in class membership are also markedly distinct. Figure 2.4 provides examples

of the dialectal variation between velar and palatalfinal stems having a final alveolar

nasal or lateral which is also characteristic dialectally of tenere and venire . This contrasts

with Modern Spanish which shows no such synchronic variation across dialects.

56

MIt INF Gloss IND # P #

rimanere remain rimango / rimagno 1 1s

valere be worth valgo / vaglio 1 1s salire go up salgo / saglio 1 1s dolere hurt, ache dolgo / doglio 1 1s

volere want voglio / volgo 1 1s 1 1s morire die muoio / morgo 1 1s Figure 2.4—Velar/PalatalFinal Dialectal Variation in MIt.

Old Spanish does, however, show localized examples of a similar palatal/velar variation that is related, to some degree, to a process of palatalization of the stemfinal consonant before front inflectional vowels shared with Italian (see §2.3.2). Palatalization and yod effects are among the most important aspects common to these languages in relation to velarinsertion since they provide the basis for the morphological reanalysis of novel paradigmatic allomorphy to the extent that it comes to have psychological relevance in the form of a ‘morphome’ (Maiden 2001, following Aronoff 1994). It should be noted that these processes highlight the marked contrast of the first conjugation against the second and third in both Italian and Spanish. Finally, as a consequence of their susceptibility to palatalization, the extension of the velarinsert across sonorantfinal stems, though to differing degrees and in spite of its languagespecific origins, is notable as an indicator of the phonological grounding of the analogical process.

It is of no surprise that velarinsertion in Italian has been analyzed and discussed in a number of investigations. Since the Italian developments closely resemble those of

57

Spanish, several such investigations are mentioned in the remainder of this chapter, especially as they pertain to the function of the paradigm in AEM.

2.2.4—THE LINGUISTIC BOUNDARIES OF VELAR -INSERTION . That the development of the velarinsert class is not characteristic of a shared stage of development of the

Romance languages is exemplified by the contrasting data in Spanish and Italian, on the one hand, and Portuguese and (Old) French on the other. The former pair shows the velar alternating in the PRS paradigm with an alveolar nasalfinal root in tener /tenere , venir /venire , poner /porre from the beginning of their respective histories, a fact which complicates the identification of a possible lead verb or of a simultaneous change

(Malkiel 1974). Portuguese and Old French, meanwhile, display an alternation in the

PRS paradigm between a palatal and nonpalatal nasal in ter /tenir , vir /venir and Ptg. pôr.37

The similarities observable between Spanish and Italian contribute to the idea that they might have shared some synchronic development to the exclusion of Portuguese and

French, particulary in the preceding case of tener , venir , poner and equivalents.

Consider, for example, their basic phonological shape, their possible function as auxiliaries, their high token frequency, not to mention the appearance of /g/ in an identical phonological environment (Malkiel 1974). In spite of such shared characteristics, however, it is not logical to suppose any earlier synchronic stage at which

Italian and Spanish had split off from Portuguese and French:

What speakers of Spanish and Italian happened to share at a crucial juncture was (a) their common disinclination to complicate or obstruct 58

their conjugational edifices with a new morphophonemic feature involving a thin contrast: /ɲ/ ~ /n/; (b) their eagerness to set apart certain members of the paradigm, e.g. the 1 st sg. pres. ind. and the subj. (“deep morphology”); (c) and their readiness to use the velar insert as a handy substitute (“surface morphology”). (Malkiel 1974:323)

What distinguishes Spanish and Italian from French and Portuguese is the phonological and analogical paths taken by each language. The palatalizing effect of yod

and the analogical influence of certain lexical items on others determined the direction

taken by Spanish speakers versus their Portuguese counterparts. Specifically, the

analogical influence of a verb like CL DICERE – DICŌ, DICIS on CL FACERE – FACIŌ, FACIS provides an initial point of departure, given MSp. digo , dices , hago , haces and MPtg.

digo , dizes , faço , fazes . Next, consider the palatalizing effects of prenuclear /g/ before a

front vowel. In Spanish, the is converted to yod, which subsequently palatalizes the preceding consonant and is lost, e.g. CL 2s TANGIS > * tan[j]es > tañes

(see §2.3.2). Meanwhile, the Portuguese case involves only palatalization of the velar,

with no effect on the preceding consonant, e.g. CL 2s TANGIS > tan[ʒ]es . At the same time, verbs like CL TENĒRE and VENĪRE were undergoing a separate set of processes by which there arose a morphomic contrast between a stemfinal palatal nasal and mere of the stem vowel. Due to the fact that these forms were diverging from the path of overlapping allomorphic structure in their paradigms, the establishment of an identical AM for Ptg. ter , vir , etc. never took shape.

Based on the preceding information, it is possible to generally exclude all but

Spanish, Italian, and Catalan (and the Occitan/Provençal/Gascon group) from the set of

Romance languages affected by velarinsertion. Therefore, the following sections

59 highlight the diachronic changes resulting in the development of the velarinsert in

Spanish, with support from related changes in Italian and Catalan. Nonetheless, the phonological and morphological changes leading each Romance language to linguistic independence are also relevant. By comparing and contrasting such changes, a picture is painted not only of some of the characteristics common to this language group, but also of the different directions in which such changes have led them. This sets the stage for an updated, more thorough analysis of the Spanish velarinsert verbal subclass in chapter 3.

2.3—(L ATE ) LATIN DEVELOPMENTS LEADING TO VELAR -INSERTION IN SPANISH . Prior to examining the previous analyses of Spanish velarinsertion, it is important to understand the diachronic phonological changes that have occurred between Latin and

Spanish so as to better identify the of factors that have resulted in this process taking hold. What follows is a brief overview of the major changes which play a crucial role in the evolution of the Romance verbal system in Spanish.

2.3.1—CONJUGATIONAL CLASS REDUCTION VIA SHIFTS IN THE CL VOWEL SYSTEM .

Late Latin in the vowel system of early Classical Latin led to a downsizing in the categorization of verbs by conjugational class as well as changes in the syllabic structure of many forms. Appendix J displays the present tense of the four traditional

Classical Latin conjugational patterns, including a subclass of the third conjugation (IIIi) marked by a short high front vowel in the 1s/3p PRS IND and all of the PRS SBJ (beside those of the IMPF (and FUT) IND). The categorization of verbs by the length of their

60 inflectional marker or theme vowel became severely limited as a distinction of vowel quantity (long vs. short) became one of quality (tense vs. lax), the latter then being lost or changed in Spanish (e.g. diphthongization of lax mid vowels in stressed position, lowering of lax high vowels, etc.). One of the most important examples of this change in the vocalic system for the Spanish verb stems affected by velarinsertion is the development of the Classical Latin front vowels Ē, Ĭ and Ĕ. In stressed position, the originally contrastive vowels Ē and Ĭ systematically emerge as a midfront vowel /e/, while Ĕ regularly diphthongizes in the same context, providing Spanish /je/. In atonic position, meanwhile, all three of these vowels produce /e/, thereby eliminating any distinction which had been available to previous generations of speakers. 38 The effects of such changes on the PRS IND forms of verbs like tener and venir are discussed below.

The primary consquence of these changes in the vowel system was a tremendous overlap of inflectional markers across conjugational classes. For example, the 2s, 3s, 1p,

2p, and 3p PRS IND forms of conjugations II and III ended up having identical inflection. 39 The same explanation adds to the understanding of the relationship between

the second and third conjugations of Modern Spanish since, early on, the inflection for

the SBJ forms of Classical Latin conjugations II, IIIi and IV merged completely. The

consequences of this merger are particularly crucial to the development of the velarinsert

class. Because the original conjugational structure was becoming much less transparent

in the transition to , and later HispanoRomance, one of the major changes in

the verbal system involved the general fusion of conjugations II and III. This merged

class took the basic shape of conjugation II, meaning that numerous verbs were added to

61 what is now the –er class. Nonetheless, some verbs, including a few of very high token frequency, instead moved in the opposite direction, changing conjugational class completely. Examples are particularly salient when comparing verbs across Romance languages, e.g. MPtg. dizer vs. MSp. decir (< CL DĪCERE ).

In contrast, a very common change across Romance is a largescale shift of verbs

from the CL IIIi subclass to the fourth conjugation (MSp. ir conjugation). The latter

was the second most numerous category in Latin and it shared with the former the presence of a front inflectional vowel in the 1s/SBJ, e.g. huir < FUGĔRE , 1s FUGIŌ , parir <

PARĔRE , 1s PARIŌ . This front vowel, alongside the one found in the second conjugation, then became the trigger for countless phonological changes that have become well known characteristics of the individual Romance languages. Not only do these changes affect the entire phonological system of Spanish, but they are also influential in its morphology and provide the foundation upon which the process of velarinsertion establishes itself.

2.3.2—YOD EFFECTS AND PALATALIZATION . Along with the changes in Classical Latin conjugations II, IIIi and IV that were a consequence of reductions in the vowel system, there was another change affecting the syllabic structure of their forms that can be pinpointed as central to the origin of the development of the velarinsert class. In the

1s/SBJ forms of the aforementioned conjugations, there existed a vocalic hiatus between the (unstressed) front theme vowel and a following nonfront inflectional vowel, as highlighted in Appendix J. This vocalic hiatus was lost in Late Latin, the primary hypothesis being that the front theme vowel systematically became a palatal glide, or

62 yod, i.e. V ([+voc, bk, ton]) > /j/ ([voc, +high]) / __ V.40 The development of this antihiatic, nonsyllabic element meant that the phonological shape of the affected forms was reduced by one . Hence, a pattern emerges in the present paradigm which differentiates the 1s/SBJ from the remainder of the present on the basis of the appearance of yod (cf. Tekavčić 1980b). This pattern is primary to the analysis of the paradigm

‘template’ (cf. Aski 1995, Maiden 1993, Vincent 1988). The emergent yod triggered a general process of palatalization affecting certain neighboring consonants. This is the first set of palatalization effects relevant to velarinsertion in Spanish. The result is the creation of a palatalized series of that helps to define the transition from Late

Latin to the Romance languages (Penny 2002; cf. Maiden 1992, Tekavčić 1980a,

Wongopasi 1993). In spite of providing the foundation for innovative paradigmatic organization, however, the initial effects of this yod in HispanoRomance are a matter of debate in the literature. The following section treats these yod effects (and resulting palatalization) through the presentation of the diachronic developments as seen in the literature. Afterwards, a second palatalization process triggered by front vowels is similarly explored since it plays a part in the emergence of the palatal series of consonants and, more imporantly, it affects the most commonly hypothesized AM for velarinsertion.

2.3.2.1—PALATALIZATION BY YOD IN THE VERBAL PARADIGM . The Latin developments involving yod are limited almost exclusively to wordinternal position (Tekavčić 1980a).

Examples of the creation of yod (as e is confused with i, both of which become yod in

63 contexts where hiatus is lost) are provided from Late Latin inscriptions, some of which display . 41 This yod regularly palatalizes any immediately preceding consonant except labials, /ɾ/ or /s/. Evidence in support of this analysis is found in the stemfinal palatal consonants in Modern Portuguese, Old French, and dialectal Modern

Italian, as shown in Appendix F. The yod effects relevant at this stage in the examination of Spanish velarinsertion involve those consonants which are susceptible to palatalization in stemfinal position within the velarinsert class, namely the alveolar nasal /n/ and alveolar lateral /l/. Again, the remaining stemfinal consonants in this group, i.e. /ɾ/ and /s/, historically have no palatalized counterpart in Spanish.

Outside of the verbal paradigm, /n/ and /l/ show palatalizing effects before yod, creating, in Old Spanish, a palatal nasal /ɲ/, e.g. SENIŌRE > señor “sir”, PĪNEA > piña

“pine cone” and a pre /ʒ/ (MSp. /x/), e.g. ĀLIU > ajo “garlic”, FOLIA > foja “leaf” respectively (Penny 2002). Within the verbal paradigm, however, these same palatalized outcomes are not found. This is especially apparent in the case of the nasal segment, as there are no attested instances of a palatal nasal in any forms of the earliest velarinsert verbs tener , venir , remanir or poner . In the slightly more complicated lateral case, the evidence for palatalization is quite limited and problematic. Two possible interpretations of the preceding information are presented below while a more detailed discussion of lateralfinal verbs and related orthographic problems is found in §3.4.

In some cases, especially those verbs with a nonpalatalizable consonant, the yod was lost without showing major effects, though, at times, raising the stem vowel prior to loss, e.g. CL DORMIŌ , DORMIAM > MIt. dormo , dorma ; MSp. duermo , duerma ; MPtg.

64 dormo , durma (Tekavčić 1980b). The potential metaphonic effect of yod provides an acceptable means of accounting for the stressed monophthong /e/ in the 1s/SBJ forms of verbs like tener and venir , since the stem vowel is expected to undergo diphthongization just like the other stressed stem vowels in the PRS IND, cf. 2s tienes , vienes . It is also used as an argument against palatalization by yod, a hypothesis considered below.

In spite of the usefulness of yod in explaining such alternations within the present paradigm, the lack of evidence of its effect on the stemfinal (palatalizable) consonant remains. This is demonstrated by the earliest Spanish documents showing latinate forms and/or yod. For example, Wilkinson (1978) attests 1p SBJ poniamus in a Latinate

Castilian document from 844, after which only forms containing the velar /g/ are attested.

Given the somewhat surprising lack of evidence of palatalization by yod in tener , venir and the like, the effect of yod in the verbal system represents a major faultline in the previous analyses of velarinsertion. Elson (1988) notes:

Assuming the presence of this [yod] segment in the first person singular indicative and throughout the subjunctive of such verbs (e.g., 1sg.id. sentio and 1sg.sb. sentia , etc. from VL sentire ), we expect a number of alternations (e.g., s/t in sentir ) which, however, are not attested. This has led some to argue that yod was eliminated in these verbs before the period in which it palatalized preceding consonants. Others apparently assume the alternations in question, claiming that they were eliminated before the period of earliest attestation. (398)

Among those who contend that palatalization of /n/ by yod occurred systematically are: Bourciez (1967), BoydBowman (1980), Bustos Gisbert (1989), Cano

Aguilar (1999), Elson (1988), Fanciullo (1998), Gassner (1897), Hall (1983),

Klausenburger (1984), Lloyd (1987), Malkiel (1958, 1974, 1989), MeyerLübke (1895),

Otero (1971), Penny (2002), Pérez Saldanya (1995a), Rohlfs (1968), Tekavčić (1980a, 65

1980b), and Wilkinson (1978, 1979). Their view suggests prehistorical synchronic variation between a palatalized form and an innovative velarinsert one, where the latter serves as analogical replacement for the former, with possible depalatalization of the nasal due to a phonotactic constraint imposed by the insertion of a following velar consonant. 42 In the broader domain of this hypothesis, these investigations nonetheless differ with respect to the origin of the velar element.

In large part, the preceding perspective is supported by evidence from other

Romance languages like Portuguese and dialectal Italian (as in Appendix F), which show the regular palatalizing effect of yod. Alongside this external data, some investigations attest palatalized forms in a handful of the velarinsert verbs in early Old Spanish, e.g. 3s

SBJ valla from valer , though the phonetic realization of such forms may be called into question (cf. orthographic problem discussed in §3.4). Also, as seen above, the fact that forms like tengo and vengo are unaffected by diphthongization suggests the earlier presence of yod, implying possible palatalization (cf. Wongopasi 1993).

On the other side of the issue are those investigations which do not propose palatalization in verbs like tener and venir because the glide which triggered palatalization disappeared (or turned into /g/) early on or because their stems were

analogically or phonetically restructured prior to palatalizing. 43 These include Alvar &

Pottier (1983), Cano Aguilar (1999), Diez (18741876), Elcock (1960), Ford (1966),

García de Diego (1951), Iordan & ManoliuManea (1972), Lathrop (1986), Lenfest

(1978, 1993), Menéndez Pidal (1980), Mondéjar Cumpián (1995), Resnick (1981),

Spaulding (1971), Urrutia Cardenas & Alvarez Alvarez (1983). Like those promoting the

66 former hypothesis, many of these analyses follow different lines of reasoning as to the emergence of the velarinsert, a matter which is explored in detail below.

MSp OSp Source CL Source

debo devo *debo DEBEŌ hago fago *faco FACIŌ miento miento *mento MENTIOR mido mido *mito MĒTIOR paro paro *paro PARIŌ parto parto *parto PARTIŌ siento siento *sento SENTIŌ temo temo *timo TIMEŌ visto visto *visto VESTIŌ

hiervo *fiervo FERVŌ FERVEŌ huelo huelo OLŌ OLEŌ Figure 2.5—Sp. verb forms not showing stemfinal palatalization by yod.

Points in favor of the nonpalatalization view are the absence of attested palatalized forms for tengo , vengo , pongo , remango , etc. in Old Spanish and the lack of

yod effects in various other verbs, as demonstrated in Figure 2.5. Two of these forms,

namely those corresponding to Modern Spanish 1s hiervo and huelo , demonstrate, in fact,

that the loss of yod was already attested in the Latin verb system (i.e. conjugations II, IV,

and IIIi), likely in the earliest centuries AD, with little to no effect on verb forms.

Likewise, “…FACUNT and FAUNT [from CL FACIUNT ] are already found in Latin

inscriptions…” (Wilkinson 1979:29). Similar support is provided from the Modern

Italian 1s forms sento, dormo, parto, servo , which show no palatalization effect from an

earlier yod. Penny (2002), among others, attributes this loss to analogy with the large

67 number of other inflected forms not having yod across the conjugation, i.e.

‘intraparadigmatic’ analogy. This situation is highly plausible since the paradigmatic and analogical pressures are such that regular sound change is often overtaken or overridden by these forces (cf. §4.2.1). Initially, this analogical relationship resulted in synchronic variation between forms with and without yod. At a later stage, these verbs could be reanalyzed as belonging to a different conjugational class, something which explains the absence of any yod effect in their Spanish derivatives (Menéndez Pidal

1980). Also, the contrast between the Portuguese and Spanish data supports a division in the distribution of palatalization by yod (cf. Elcock 1960). Relatedly, the split in Italian between palatalization and velarinsertion may also be linked to this phenomenon. Once

more, the lack of diphthongization in the stem vowel of forms like 1s tengo is attributed

to the raising effect of the yod which was subsequently lost. This is corroborated by

several facts concerning neighboring Catalan. First, Modern Catalan 1s tinc shows

raising of the stem vowel by yod. Second, this language permits wordfinal palatal

consonants which means that there is nothing preventing the yod from both raising the

stem vowel and palatalizing the (originally) stemfinal nasal. In the end, the divide in the

literature regarding the effect of yod demonstrates that “the explanation of the velarinsert

is in some way connected to the loss or retention of yod in the of different

regions” (Lenfest 1978:899).

Returning to the earlier discussion of yod effects in the Old Spanish verbal

system, the sequence /lj/ typically resulted in the creation of a palatal lateral which

changed to a voiceless prepalatal fricative /ʒ/ in preliterary Castilian (examples above).

68

Since this palatalization process did not occur systematically, more needs to be said about forms like Classical Latin 1s SALIŌ, VALEŌ as they would have been subject to changes resulting in unattested * sajo , * vajo . There are two possible conclusions with respect to this issue. First, it is presumable that the yod in these forms was lost prior to palatalization, as suggested in several highly recognized analyses. For example, Lenfest

(1978) contends that were the yod not lost early on, there would have been much more competition and alternation, e.g. * faço ~ fago , * yaço ~ yago , * teño ~ tengo , * veño ~ vengo (cf. Italian tegno ~ tengo , vegno ~vengo ). Other similar alternations were, in fact, not resolved prior to the written historical record in the and actually lasted for long periods of time. Thus, if archaic or formal contexts do not show * faço or

*teño , it is likely that they never existed in Old Spanish. This is supported by the

rampant (nonverbal) palatalization of /n, l/ in the same contexts outside the verbal paradigm (see above). The last point is also indicative of the second possibility, that such

forms were resistant to palatalization on the basis of paradigmatic influence, again, a

theory which has been proposed on more than one occasion (see, for instance, Penny

2002). The early appearance in Spanish of forms without yod does not necessarily

support one of these possibilities over the other since it can be accounted for by both

analyses, arguing that these forms are the direct development of the loss of yod, or that

they promote intraparadigmatic influence of the remaining verb stems which do not

show yod effects for phonological reasons. Realistically, it is more than likely that the

absence of yod and palatalization effects in the vast majority of verbs is due to a

combination of these arguments. That is to say, the loss of yod may have been supported

69 by the fact that the forms in question are part of a paradigmatic structure in which the other inflected forms lack this palatalizing element. This is a wellaccepted aspect of paradigmatic resistance to sound change as described in §4.2.

With respect to the lack of palatalization in the nasal plus yod sequence, one additional point must be considered. The appearance of the velar element postnasally is a prehistoric phenomenon while the insertion of the velar in the lateralfinal case seems to have occurred in the earliest stages of the written historical record of Spanish. Hence, the palatalization of the laterals may have been prevented by the earlier appearance of the nasal plus velar sequence in a related set of verbs and possible AM. Therefore, the palatalization effects of yod are most relevant to the primary set of velarinsert verbs, i.e. those having a stemfinal nasal /n/. As might be concluded from the longlasting debate in the literature, the lack of salient evidence for one analysis over the other has led to less extreme hypotheses. Malkiel (1974), for example, reconstructs /tenjo/ or /teɲo/, /venjo/ or /veɲo/ for the Late Latin predecessors of 1s tengo , vengo . All of this must be taken

into consideration in the identification of the original source, or AM, for the process of

velarinsertion.

A briefly mentioned effect of yod that is crucial for the extension of the

velarinsert involves the possible palatalization of a preceding occlusive consonant. For

the most part, the antihiatic yod is believed to have been lost in Spanish when found

only in part of the verbal paradigm, generally without palatalizing an immediately preceding occlusive, e.g. CL 1s SENTIŌ > siento , FACIŌ > OSp. fago (MSp. hago ) (cntr.

MPtg. faço , MIt. faccio , etc.). The yod in these verbs was presumably lost through the

70 analogical restructuring of these forms according to intraparadigmatic influence.

However, the sequences /bj, dj, gj/ were susceptible to yod effects, as demonstrated by instances of the orthographic representation of fricative/palatalized consonants in Latin, e.g. ZABOLUS for DIABOLUS (Tekavčić 1980a). The above obstruent plus yod sequences resulted in the emergence of a palatal fricative /ʝ/, or palatoalveolar /dʒ/, that is subsequently lost after a front vowel, e.g. CL 1s VIDEŌ > OSp. veyo > MSp. veo , RĪDEŌ >

OSp. riyo > MSp. río , AUDIŌ > OSp. oyo (>> MSp. oigo ), FUGIŌ > fuyo , 1s SBJ SEDEAM

44 > sea , HABEAM > haya . This palatalizing effect of yod is believed to be simultaneous to

that affecting nasals and laterals. The innovative was also extended

analogically to some originally class III Latin verbs which did not derive it

etymologically, e.g. 1s CADO > OSp. cayo (>> MSp. caigo ), RADO > rayo (>> MSp.

raigo ), RODO > royo (>> MSp. roigo ), VADAM > vaya , TRADO > OSp. traya (>> MSp. traiga ). 45 In some instances, it gained ground intraparadigmatically, affecting highly

reduced forms with an unacceptable hiatus involving a nonfront vowel as the first

element in hiatus, e.g. 2s * oes >> oyes (Penny 2002). 46 As should be noted from the

modern forms above, the velarinsert is found following the offgliding palatal segment

in the 1s/SBJ. Its development in these instances is unanimously attributed to analogy

with the already established set of velarinsert verbs, i.e. tener , venir , salir , valer , etc. (cf.

Menéndez Pidal 1980, Penny 2002). The forms in question attained the velarinsert at a later stage, as discussed in various sections of the following chapter, meaning that they are not central to the analysis of the origin of the velar element but rather are

71 of its extension through the verbal subsystem of the lexicon. Nonetheless, they are affected initially by the appearance of yod.

The effect of yod on what is to become the set of velarinsert verbs is compounded by related changes that took place in other verbs having a similar phonetic shape and allomorphic distribution of stem alternants. In particular, the palatalizing effect of front vowels on velar consonants is relevant for the Classical Latin set of –

NGERE verbs, i.e. class III verbs having the stemfinal sequence /ng/, as well as verbs having (intervocalic) stemfinal /k/ followed by a front vowel, a group that includes the highly frequent verbs decir (CL DĪCERE ) and hacer (CL FACERE ). Because these two groups of verbs are the most commonly cited AM for velarinsertion, the changes, phonological and otherwise, which are involved in shaping their PRS paradigms are examined below.

2.3.2.2—THE CLASSICAL LATIN –NGERE VERB SET . Velarinsertion is intrinsically tied

to the process of palatalization visible in the forms in Figure 2.6, as suggested in previous

research (e.g. Malkiel 1967, 1974, 1989). The syllableinitial voiced velar /g/ suffered

spirantization and palatalization, developing into a palatalizing element (possibly yod) before a front vowel (remaining /g/ before nonfront vowels, per the highlighted 1s/SBJ

alternation). This frequently led to palatalization of the preceding (palatalizable)

consonant, something which, for present purposes, affects the development of the

47 Classical Latin –NGERE verb set (below). However, palatalization was not the only

outcome of contact between syllableinitial /g/ and an immediately following front vowel.

72

Because so many of the examples are verbal, the expected development is complicated by paradigmatic/morphological pressures, leading to specific changes to the velar element depending upon the characteristics of the preceding consonant. Penny (2002), following

Malkiel (1967, 1974) indicates that /ɾg/ maintained its pronunciation before nonfront vowels, though before front vowels it was susceptible to spirantization/palatalization effects, generally producing Old Spanish /ɾdz/ (MSp. /ɾθ/ or /ɾs/). The sequence /ng/, meanwhile, is more problematic since before front vowels it shows three possible outcomes: (1) loss of /g/; (2) palatalization of the nasal with loss of /g/ (possibly through a change /g/ > /j/ / __ /e, i/ (cf. Ford 1966)); or (3) palatalization effects similar to those for /ɾg/, i.e. OSp. /ndz/ (MSp. /nθ/ or //). Lloyd (1987) prefers a direct phonological development with morphological interference resulting in the synchronic variation. The alveolar affricate /dz/ is seen as the regular phonological outcome, but speakers' awareness of the morphological structure of the verb might have permitted them to analyze the verbal morpheme independently from its syllabic structure thus eliminating any possible contrast between velar plus front vowel and antihiatic yod, and leading to palatalization of the preceding nasal (with loss of the palatalizing segment). This is again illustrative of the resistance to change in the inflectional paradigm (cf. §4.2).

73

Lg INF Gloss IND SBJ # P #

CL TANGERE touch TANGŌ TANGAM 1 1s TANGIS TANGĀS 2 2s TANGIT TANGAT 3 3s TANGIMUS TANGĀMUS 4 1p TANGITIS TANGĀTIS 5 2p 48 TANGUNT TANGANT 6 3p

OSp Tañer pluck (a string) tango tanga 1 1s tañes tangas 2 2s tañe tanga 3 3s tañemos *tangamos 4 1p tañéis *tangades 5 2p tañen *tangan 6 3p

MSp Tañer pluck (a string) taño taña 1 1s tañes tañas 2 2s tañe taña 3 3s tañemos tañamos 4 1p tañéis tañáis 5 2p tañen tañan 6 3p Figure 2.6—Diachronic development of MSp. tañer , from CL –NGERE verb set.

In short, the developments that affect an etymological syllableinitial velar before a front vowel are highly complex, given that there are at least three attested outcomes in

Old Spanish, e.g. fra nz er , fra ñer (< CL FRANGERE ), and qui nientos (< CL QUĪNGENTŌS ).

Countless efforts have been made to identify and motivate one or more of these outcomes as being the phonologically regular one. 49 In many analyses, it is argued that the palatal

nasal is the expected outcome in order to motivate the reconstruction of preliterary forms

of tener and venir having the same phoneme, e.g. 1s * teño . What is truly relevant for present purposes, however, is that there arose a period of synchronic variation such that

one finds the original velar in the 1s/SBJ forms of verbs in alternation with either nz or ,

the latter of which quickly becomes the victor in nearly all verbs derived from this 74 source. 50 It seems that this was the situation by the before the paradigms of

these verbs were analogically influenced towards uniformity of the rootfinal palatal

nasal in later stages of Old Spanish (Menéndez Pidal 1980). Paradigmatic analogical

influence, a major tenet of this investigation, stands out since the available Spanish data

shows that development of the voiced velar after a sonorant and before a front vowel is

not purely phonological given the multiple outcomes seen in the resulting synchronic and

diachronic variation.

Additional examination of the set of –NGERE verbs is provided below as it relates to the function of this verb class as a potential AM for the emergence of the velarinsert category. In particular, morphological and paradigmatic similarities are shown to be among the primary characteristics upon which the analogical relationship is based, in conjunction with the phonological aspects indicated here.

2.3.2.3—THE DECIR, HACER SET . Another AM which requires further analysis is the set

captained by the high frequency verbs decir and hacer . It includes those verbs which, at

a stage prior to velarinsertion, showed a 1s/SBJ distribution of a velarfinal stem

allomorph as a result of very different developments. Among them one finds yacer (e.g.

1s yago ), cocer (e.g. 3s SBJ cuega ) and those verbs derived from the Classical Latin base

–DŪCERE , e.g. aducir (1s adugo ), reducir (1s redugo ), etc., see below). Many of these verbs derive their allomorphy by way of regular sound change (‘lautgesetzlich’), while, in some cases, like that of hacer , there are additional factors in play. It is important to note that a large portion of the verbs included in this group showed synchronic variation in

75 their inflected forms and/or have been reanalyzed or leveled in favor of alternative patterns. 51 This is indicative of the competing analogical factors available to speakers and has most recently been attributed to token frequency of the individual lexical items

(though also possibly to the low type frequency of the alternation, cf. Martín Vegas

2007). The matter of frequency is explored further in §4.3.

The allomorphic distribution and general diachronic patterns of this verb group are exemplified by the verbs decir and hacer in Figure 2.7. These two verbs show that the stemfinal voiceless velar is affected by two separate phonological developments determined by the frontness of the following vowel. Voicing of intervocalic obstruents, a wellknown characteristic of the process of Western Romance Lenition, affects stemfinal

/k/ in the 1s/SBJ when it is followed by a nonfront vowel. 52 On the other hand, the presence of a front vowel in this same environment leads to palatalization, producing an

alveolar affricate /ts/ that is also susceptible to Lenition, producing /dz/. In fact, this is

one of the earliest changes in the transition from Latin to Spanish, occurring as early as

the 1 st c. according to the orthographic confusion found in an example like

INTCITAMENTO for INCITAMENTO (Fradejas Rueda 1997; cf. Ford 1966). The verb cocer

and those from CL –DŪCERE can also be explained in this way, thus providing a fairly frequent set of similarly shaped verbs with an innovative allmorphic distribution in the present tense.

76

Lg INF Gloss IND SBJ # P #

CL DĪCERE say, tell DĪCŌ DĪCAM 1 1s DĪCIS DĪCĀS 2 2s DĪCIT DĪCAT 3 3s DĪCIMUS DĪCĀMUS 4 1p DĪCITIS DĪCĀTIS 5 2p DĪCUNT DĪCANT 6 3p

OSp dezir say, tell digo diga 1 1s dizes digas 2 2s dize diga 3 3s dezimos digamos 4 1p dezides digades 5 2p dizen digan 6 3p

MSp decir say, tell digo diga 1 1s dices digas 2 2s dice diga 3 3s decimos digamos 4 1p decís digáis 5 2p dicen digan 6 3p

CL FACERE do, make FACIŌ FACIAM 1 1s FACIS FACIĀS 2 2s FACIT FACIAT 3 3s FACIMUS FACIĀMUS 4 1p FACITIS FACIĀTIS 5 2p FACIUNT FACIANT 6 3p

OSp fazer do, make fago faga 1 1s fazes fagas 2 2s faze faga 3 3s fazemos fagamos 4 1p fazedes fagades 5 2p fazen fagan 6 3p

MSp hacer do, make hago haga 1 1s haces hagas 2 2s hace haga 3 3s hacemos hagamos 4 1p hacéis hagáis 5 2p hacen hagan 6 3p Figure 2.7—Diachronic development of MSp. decir and hacer .

77

In spite of the straightforward analysis of the forms above, the earlier loss of yod is crucial in order to account for the development of 1s/SBJ forms of hacer , as well as those of yacer and placer , since they would have developed an antihiatic yod in Late

Latin. If the yod is lost, as is likely to have happened given the attested evidence

mentioned in §2.3.2.1, the development of 1s hago is identical to that of 1s digo , i.e. the result of regular sound change affecting a reconstructed source, 1s * FACŌ . Thus,

“Menéndez Pidal’s general statements about the loss of flexional yod seem to have even wider application than he envisioned. They explain not only FACIŌ  hago , JACEŌ 

OSp. yago , but also TENEŌ  tengo , VENIŌ  vengo ” (Lenfest 1993:636). Extending the idea of the early loss of yod accounts for the lack of palatalization of the stemfinal nasal in the preceding examples and for the hypothesis that digo , hago and yago are central to the AM for velarinsertion. If the yod were not lost early on, the shape of the stem in the paradigm of Old Spanish fazer would have been completely regular, i.e. 1s * fazo < * faço

(cf. Ptg. faço ), 2s fazes , etc. 53 While hypothesizing the loss of yod in such forms does

require an initial period of analogical adjustment, the presence of yod necessitates a more

complicated and less natural analogical remodeling of a seemingly regular paradigm.

Mondéjar Cumpián (1995) suggests that the latter can be discounted because (1) speakers

would tend not to reanalyze a leveled verb stem like faz once it is traditionally

established, especially considering its high token frequency (cf. Martín Vegas 2007); (2)

there are a very limited number of available AM for such a change; and (3) such an

explanation rests squarely on the uneasiness of speakers before an undesirable paradigmatic or morphological structure.

78

What emerges from these phonological developments is a pattern of allomorphy that is limited to the verbal paradigm (Martín Vegas 2007). This is due to the fact that the inflectional paradigm provides the only context where these two patterns of sound change are found in alternation. The morphological consequence of this localized domain for phonological change is an increased salience of the pattern of allomorphy as a unit of cognitive organization, i.e. ‘paradigmatic template’ or ‘morphome’, especially given the high token frequency of verbs like decir and hacer . This concept is identified herein as a crucial factor in the emergence and extension of the velarinsert subclass.

Additional aspects of the aforementioned AM are examined in §2.4.2.

2.3.3—THE RELEVANCE OF YOD EFFECTS AND PALATALIZATION . The palatalizing effect of front vowels (and the yod derived from them) is an essential starting point in the origin of the velarinsert verb class. Although, in many cases, such an effect is indirect, through its involvement in the development of potential AM for velarinsertion, it does prove a meaningful point of contrast in the literature on the subject. In fact, the importance of the antihiatic yod that emerged from a front vowel in hiatus with an immediately following vowel extends well beyond the matter at hand. Nonetheless, the effect of yod on the earliest velarinsert class members is indicative of the subsequent stages required by any analysis.

If palatalization of nasals (and possibly laterals) by yod did occur in verbs like tener and venir , it did so only prior to the written historical record of Spanish. This hypothesis is in line with the wider Romance phenomenon of palatalization. In such a

79 case, however, synchronic variation between palatal and nonpalatal (or velar) stems would have been shortlived, resolved in favor of the velarinsert forms, likely motivated by a number of external, analogical factors.

If the antihiatic yod is lost before palatalizing the sonorants /n, l/, it seems likely that speakers would have been faced with an intraparadigmatic complication. The distribution of stem alternants that would have arisen in a verb like tener , i.e. 1s * teno 2s tienes , 1p tenemos , 1s SBJ tena , etc., was unprecedented and highly implausible since the regular distribution of the /monophthong alternation was well established across all conjugational classes. The resolution to this innovative and unexpected paradigmatic alternation would have been grounded in analogy, either through intraparadigmatic leveling or the use of a highly similar AM. The high token frequency of the individual forms involved itself to the latter response. This AM required a high degree of phonetic, morphological and paradigmatic similarity. The most commonly hypothesized AM are the set of CL –NGERE verbs and verbs like decir and

hacer , both of which developed a 1s/SBJ allomorphic pattern in the present paradigm due primarily to various palatalization processes which characterize the tranisition from Latin to Romance.

These hypothesized analogical sources are both excluded from membership in the the velarinsert class for different reasons. The –NGERE set is excluded because of the palatal feature of the nasal in alternation with the nasal plus velar sequence, the

etymological nature of the velar element and, more importantly, the eventual leveling of

the paradigm in favor of the palatal nasal. Similarly for the decir /hacer set, given the

80 etymological origin of the paradigmatic alternation in its initial stages, as well as the fact that this group was formed at a stage prior to the emergence of the velarinsert class and the intervocalic context in which the velar occurred, these verbs are not interpreted as members of the highly specified velarinsert category as it is defined here.

An earlier point which requires a more transparent explanation concerns the quality of the inflectional vowels as an important factor in the separation between the first conjugation on the one hand and the second and third conjugations on the other (i.e. the contrast of Classical Latin class I against the remaining conjugational classes throughout the transition to Romance). Spanish conjugations II and III are marked by a front theme vowel in the PRS IND, in contrast to the nonfront theme vowel of I. In the PRS SBJ, however, the situation is reversed such that the first conjugation utilizes a front vowel to mark mood while the remaining classes have a nonfront marker. On a phonological basis, then, all conjugations should be susceptible to the palatalizing processes described above. A brief examination of the regular verbs in each conjugational class, however, shows that this is not the case. Spanish, like most other Romance languages, has avoided regular phonetic developments (like palatalization) in the PRS SBJ due to its intraparadigmatic (morphological) relationship with the PRS IND.

The above morphological interference or involvement has been identified in various ways throughout the literature. Menéndez Pidal (1980) sees it as the “influence” of the indicative on the subjunctive. Martín Vegas (2007), following a very common line of reasoning found throughout the history of linguistic analysis, believes that the subjunctive forms are analogically built upon, or derived from, those of the indicative,

81 especially the 1s or 3s, e.g. MSp. 1s/3s SBJ pague (not * pace ) << 1s pago, 3s paga

(INF pagar ), etc.; 1s/3s SBJ cene (not * ceñe ) << 1s ceno , 3s cena (INF cenar ), etc. It

seems more plausible, and more in line with the proposals made in this investigation,

however, that the avoidance of regular phonetic developments in the subjunctive, while based in part on the phonological shape of the forms in the indicative, is due to an AM sharing multiple phonological, morphological and paradigmatic similarities across the lexical category of verb. Essentially, this AM is the verbal paradigm itself since the overwhelming majority of verbs have shown both synchronically and diachronically a completely regular stem across the PRS. This highly stable and extraordinarily frequent

AM is the reason that previous analyses, theoretical and otherwise, have viewed the verbal paradigm as resistant to phonological change and has led to important work in the relationship between phonology and morphology. These types of issues are touched upon again in §4.2.

2.4—THE ANALOGICAL BASIS FOR SPANISH VELAR -INSERTION . Previous analyses

attribute the appearance of the nonetymological velar in the verbs under investigation to

a proportional analogical relationship with various AM like those mentioned above. At

this juncture, it is pertinent to ask why it is necessary to appeal to analogy in the first place. Couldn’t velarinsertion simply be phonological, morphological or both? In spite

of the phonological contexts where velarinsertion appears in the modern language (i.e.

after /n, l, ɾ, s, j/ and before /a, o/), the proposed solution cannot be based on phonology

alone, i.e. *–EA , EŌ  ga , go / [n, l, ɾ, s, j] __, or synchronically, *no , na

82

 ngo , nga . This is demonstrated by the fact that velarinsertion is absent in forms of

the –ar conjugation (i.e. 1s ceno from cenar ‘dine’, not * cengo ). There are a number of

reasons why this is so. The first major insight into this question is taken from Malkiel

(1967), which stresses the importance of comparing forms that did change with those that

did not, particularly those which were borrowed into the language at different stages with

respect to the phenomenon under examination (i.e. before, during, after) to determine its

degree of productivity and level of systematicity. There are numerous verb forms which

did not suffer velarinsertion or did so to varying degrees, including soler , ferir (MSp.

herir ), doler , toser , oler , coser , pulir as well as the entire first conjugation (see Appendix

K). That these forms do not systematically assume the paradigmatic allomorphy

characteristic of the velarinsert verbs is a complication from the perspective of a regular, purely phonological change. Such target forms are not easily contrasted etymologically

from those whose stem allomorphy was restructured. Therefore, the change which

occurred cannot be based solely on the phonological conditioning environment.

Further evidence of the ‘localized’ regularity of the velarinsertion process is

demonstrated by the fact that it is not found outside the verb system and, within this

grammatical category, its distribution is not globally triggered by any

morphophonological conditioning environment (cf. §1.3.1). There are many verbs (and

nonverbs) which have not undergone velarinsertion in spite of containing all the

applicable morphological and phonological characteristics. Outside of the verbal paradigm, the result is (primarily) palatalization, e.g. the nominal forms SENIŌRE > señor

“sir”, cntr. *sengor ; PĪNEA > piña “pine cone”, cntr. * pinga ; ARĀNEA > araña “spider”,

83 cntr. *aranga . Consequently, the morphological boundary, or the distinction of stem from inflection is also insufficient, even in conjunction with the necessary phonological information. Were the phenomenon triggered by the morphophonological structure alone, it would be possible to argue for velarinsertion in nouns, or at least across all target forms in the verb system. An additional morphological factor is that the inflectional vowel immediately following the verb stem be nonthematic. This implies that the susceptible verbs are from conjugations II (er ) and III (ir ) only. In this way, it

is possible to account for the lack of velarinsertion in both nouns (e.g. araña , cana ) and

verbs of the first conjugation (ar , e.g. cenar ).

Given the preceding information, it is necessary to broaden the exploration for such conditioning factors. Malkiel (1967) suggests searching within a previously uncharted domain, the inflectional paradigm:

Diagnostically, it seems wisest to start from the axiomatic assumption that most sound changes can be accounted for in terms of a phonological system’s internal balance and economy. It is only where explanations of this order fail, without undo stretching, to do justice to ascertainable facts that the analyst is well advised to try out a ‘second string’ of possible factors of causation…It seems theoretically defensible and also feasible to attach to this second string the possibility, rare in some languages yet conceivably frequent in others, of inflectional pressure. (49)

In the case of first conjugation verbs, it may be argued that velarinsertion does not occur due, in part, to “intraparadigmatic” influence involving the distribution of nonfront inflectional vowels across all PRS IND forms. Increasing allomorphy in these forms would require complicating the stability of the leveled paradigm in favor of completely innovative allomorphy and the consequential emergence of an otherwise unattested templatic distribution of internal verbal allomorphy (i.e. IND vs. SBJ stem). 84

The paradigmatic template that emerges exclusively from the allomorphic pattern of verb stems in conjugations II and III is then seen as a consequence of the linguistic organization of inflected verb forms. Also, the preceding supports the fact that Spanish – ar conjugation is resistant to allomorphy, preferring isomorphic stems over alternation.

Though surface true, this does not arise from an imposed constraint, but is instead a

consequence of paradigmatic influence based on the distribution of morphophonological

factors. Although the factors involved in the process are becoming clearer, the lack of

systematic insertion of the velar in the verbal subclass remains problematic. Thus, more

details are required to appropriately identify the conditioning factors triggering

velarinsertion.

From a diachronic standpoint, the fact that velarinsertion was productive before

the written historical record of Spanish (10 th c.) until about the 16 th c. is indicative of the gradual nature of analogical change. Meanwhile, it could be anticipated that, like the SBJ of conjugation I verbs, the earliest members of the velarinsert subclass would have undergone palatalization in the 1s/SBJ as the result of regular sound change. The attested data, unfortuantely, do not support such a hypothesis. This raises the question, if no forms can be found with a palatal nasal, why should it be proposed in such cases? 54

Analogy permits a direct transition to the attested stage where these verbs contain a velar element. In fact, even some of the most notable works promoting palatalization by yod contend as much. Malkiel (1974) finds no reason to suppose an intermediate stage in every case: “the alternative leap from /n/ to /ŋg/ seems equally plausible” (342).

85

On the basis of the preceding points, the explanation for the origin and extension of the velarinsert lies in analogy though it has been assigned a seemingly arbitrary, hence overly powerful function in many instances. This is the only point upon which previous research agrees, that velarinsertion became analogical at some point, either

from the outset or soon thereafter. The focus, however, has been primarily on the origin

of the velarinsert, without paying similar attention to its diachronic extension through

the lexicon. At the same time, the details of the original AM are not specific. The

following sections illustrate the lack of depth that has plagued earlier investigations of

Spanish velarinsertion, setting the scene for discussion of the most wellestablished

analyses of velarinsertion.

2.4.1—VELAR -INSERTION AND LINGUISTIC INDEPENDENCE . The most difficult question to answer and the one to which the most time and effort has been dedicated is where and when exactly did the velarinsert originate? A handful of earlier investigations (Elson

1988, Lenfest 1978, Penny 2002, Tekavčić 1980b) have suggested that the initial stages of velarinsertion are likely to have occurred at a point when there was still a relationship between geographicallydistant Romance dialects. This is based on two important points:

(1) the alternation (reconstructed for Spanish) between velar and palatal stems in the

verbs tener (teng ~ * teñ ) and venir (veng ~ *veñ ) and (2) the fact that velarinsertion

occurs or has occurred in a group of geographically connected Romance languages (i.e.

Sp., Cat., OProv., It.), while in the ‘fringe’ languages (i.e. Ptg., Fr., Rom.), the

velarinsert is not found (Lenfest 1978). 55 More to the point, Hall (1983) indicates that

86 the velarinsert is not ProtoRomance but rather ProtoItaloWesternRomance. In doing so, the linguistic division created by the presence or absence of the velar in the verbal paradigm takes on greater definition.

The preceding hypothesis is less likely than originally thought. Fanciullo (1998) provides arguments against it with specific discussion of Spanish and Italian diachronic developments. Similarly, Lenfest (1993, quoted above) argues against unnecessarily proposing shared synchronic changes across Romance languages. The geographical connectedness of the phenomenon is actually a consequence of the similarities in the effects of certain phonological changes, inflectional features and paradigmatic patterns.

Each affected language, however, has its own particular combination of characteristics upon which their individual AM is founded. On the other hand, French and Portuguese did not share enough of these factors in order for a stable velarinsert AM to arise.

Instead, the same verbs in these languages were molded by other sources including the previously described processes of palatalization. In sum, Spanish, Catalan and Italian each had a particular set of features in a wellestablished set of verbs such that they provided the foundation for the extension of a paradigmatic pattern of allomorphy. In each language, however, the exact details of the direction of change were dictated by the variables associated with a languagespecific AM, which, though unique to each language, share many basic similarities (cf. §2.2 and 4.5).

2.4.2—POSSIBLE AM FOR VELAR -INSERTION . As mentioned earlier, the other major

division in the previous literature, beyond the loss of yod, involves the AM for the intial

87 stage of velarinsertion, i.e. the preliterary appearance of the velar in nasalfinal verb

56 stems. The verbs poner , tener , remanir and venir (CL PONERE , TENĒRE , REMANĒRE ,

VENĪRE ) are believed to have been restructured analogically according to: (1) the set of verbs ending in –NGERE in CL, e.g. CINGERE ‘encircle; sheathe(a sword)’, FRANGERE

‘break’, TANGERE ‘touch’ (Alvar & Pottier 1983, Bourciez 1967, Cuervo, in Bello 1964,

Elson 1988, García de Diego 1951, Gassner 1897, Hooper 1979, Klausenburger 1984,

Lloyd 1987, Malkiel 1974, 1989, Menéndez Pidal 1980, Rohlfs 1968 (It.), Tekavčić

1980b (It.), Urrutia Cardenas & Alvarez Alvarez 1983, Wilkinson 1978); (2) the decir , hacer set, including related verbs yacer , placer , cocer , etc. (Baist 1888, Blasco Ferrer

1984 (Cat.), Cano Aguilar 1999, Ford 1966, Martín Vegas 2007); (3) alternative sources like the expanding set of inchoative verbs (Elson 1988, Lenfest 1978, 1993); or (4) the combined influence of the –NGERE and decir , hacer sets (Bustos Gisbert 1989, Maiden

1992, Malkiel 1967, Pérez Saldanya 1995a (Cat.), Spaulding 1971). 57 Each of these possible AM is examined below.

2.4.2.1—THE –NGERE SET , PART II. As a means of summarizing the analogical

inception of velarinsertion in Spanish, Bourciez (1967) proposes the unorthodox proportional equation: 1s * fraño (analogical, cf. Ital. fragno ) : 1s frango (etymological)

:: 1s * teño (etymological from CL TENEŌ ) : X, where X = 1s tengo . Although the use of

stem variants is unusual, the equation is possible because, "analogy takes place when

there is a high degree of configurational similarity among the forms affected; it most

often occurs when the phonetic environment is essentially compatible with the innovative

88 feature" (Lenfest 1978:898). If it is presumed that palatalization did affect tener , venir , etc., then their paradigmatic shape would have overlapped with that described above for the –NGERE set. That is to say, not only do these groups of verbs have a similar phonetic

shape in their inflected forms (i.e. monosyllabic stems having the structure C(C)VN), but

the allomorphic distribution arising from palatalization of some forms in the –NGERE verbal paradigm, serving to isolate the 1s/SBJ forms, would be highly similar in the analogical targets. The difference between the –NGERE and the (eventual) velarinsert groups at this point would have been the stemfinal consonants that alternated in the PRS.

More specifically, there would be alternation of velar/palatal and velar/nonpalatal respectively. The subsequent changes affecting the –NGERE set are the result of reanalysis due to intraparadigmatic influence (across tense and mood). Malkiel (1967) chalks all of this up to the reaction of speakers:

It must be understood that in the medieval paradigm of verbs traceable to prototypes in –NGERE there survived certain forms, directly representative of the sequences –NGŌ or –NGA , in which –ng was temporarily left intact. Thus, the present indicative of the descendant of FRANGERE ran: 1 frango , 23 franze(s) , 4 franzemos or frañemos ; and the subjunctive was franga(s) …It was the speakers’ reaction to multidimensional complexities of this kind which, in the end, provoked widespread leveling, in the course of which the remaining –ng forms were wiped out in favor of –ñ or, occasionally, of –nz , nç : tango > taño ‘I touch,’ unga > unza ‘let me (or him, her) yoke’. (36)

The main problem with the above analysis involves the associative strength of the

–NGERE class, exemplified in Figure 2.8, affecting verbs like tener and venir . First, if the

–NGERE group makes the velar element available for insertion, why were its effects not felt paradigminternally? How could the influence of the velar element be felt outside but not be maintained within its own paradigm? Similar arguments can also be made for 89 tener and venir presuming they were affected by palatalization. In order to answer these questions, it is necessary to appeal to the entire inflectional paradigm and not just the present tense. Tekavčić (1980b) feels that speakers were conscious of the fact that the velar was etymological/original in the –NGERE verbs and thus presumed the same for tener and venir . What is left out of such hypotheses is the opposition in the distribution of alternants in the model and target paradigms. AEM requires as high a degree of phonetic similarity as possible between model and target. If the –NGERE set exclusively comprises the AM, velarinsertion in the 1s/SBJ should either not occur or be complemented by palatalization of the nasal in the remaining IND forms. However, with the addition of the already established templatic structure of the 1s/SBJ division in the paradigm (see below), the velar could be extended from the NGERE set to the velarinserts. Once established the allomorphic pattern is established in the latter, it is lost in the original AM through intraparadigmatic leveling involving the higher token frequency of the nonvelar stem alternant. Both velarinsertion and leveling in favor of the palatal nasal, then, are representative of the analogical influence of the verbal paradigm, external in the former (‘interparadigmatic’), internal in the latter

(‘intraparadigmatic’).

90

CL Gloss MSp.

CINGERE encircle; sheathe (a sword) ceñir FINGERE shape, mold heñir FRANGERE break frañer IUNGERE unite; yoke uñir, uncir PLANGERE lament plañir PUNGERE prick; sting puñir, pungir RINGERE scold reñir STRINGERE draw tight estreñir, constreñir TANGERE touch tañer TINGERE dye teñir Figure 2.8—Examples of the CL –NGERE verb set.

Another part of the problem with respect to the influence of the –NGERE class of verbs is related to token frequency (cf. Ford 1966, Lenfest 1978). How could speakers permit a group of highly frequent verbs like poner , tener , venir , etc. to be changed by a less frequent one? Such an argument is countered by Lloyd (1987) and Wilkinson

(1978), on the basis that during the initial stages of the process (i.e. preliterary Spanish), these verbs were sufficiently numerous in type so as to be available as an AM. Type and token frequency are concepts which are crucial to the analogical process and are therefore explored in more detail in §4.3.

2.4.2.2—THE DECIR , HACER SET , PART II. Questions concerning the influence of the –

NGERE set on the velarinsert class, like those above, have led to the examination of alternative models. Ford (1966), for example, does not see any viable means for explaining a shift from (expected) /ɲ/ to (unexpected) /ng/ and calls into doubt the influence of the NGERE verbs, since they are not as “common and important” as velarinsert verbs like tener (297). Instead, the appearance of the velar is hypothesized

91 by many investigations to have its basis in the decir , hacer set, in conjunction with possible resistance to palatalization of the nasal in the 1s via intraparadigmatic influence, i.e. the lack of palatal nasals in the other forms of the paradigm (i.e. avoidance of phonemic variation in the verb stem (a type of leveling), cf. Spaulding 1971, Lenfest

1978). The use of both paradigminternal and external evidence is notable since they are crucial factors in the overall development of the velar class.

The primary motivation for suggesting the decir , hacer AM lies in token

frequency. A recent study of token frequency in the modern language finds that decir and hacer are respectively the 28 th and 25 th most frequently used words by lemma

(Davies 2006). Poner , tener and venir , being among the first verbs to undergo

velarinsertion, are numbered 77, 18 and 105. Although these figures are relative, given

the method by which they were calculated and the fact that they are based on modern

usage, they still serve to illustrate the extremely high frequency of these verbs.

Related to the matter of frequency is the resistance of the inflected forms of the

verbs decir and hacer to analogical reanalysis. After the establishment of the 1s/SBJ pattern in the present paradigm as a result of the regular phonological changes described previously, their PRS paradigms have remained relatively unchanged (with the exception

of minor phonetic adjustments). In other words, the pattern of allomorphy found in the

PRS has not been affected by any sort of analogical pressure, be it an innovation in paradigmatic structure or leveling in favor of an isomorphic stem.

This is not to say that the verb class as a whole has been unaffected by analogical pressure. Although the preceding information is true of decir and hacer , the remaining

92 verbs have been restructured according to other counteracting patterns. The set of verbs ending in –ducir , as well as placer and yacer , have be restructured on the basis of the

‘inchoative’ set of verbs described below. Their modern 1s/SBJ forms are marked by a

combination of fricative plus (voiceless) velar occlusive stemfinally in contrast to a

simple fricative in the remaining present forms, e.g. 1s plazco , 2s places . Alternatively, the verb cocer has been leveled by intraparadigmatic influence, i.e. OSp. 1s cuego >>

MSp. cuezo (cf. 2s cueces ). An obvious factor in the reanalysis of these verbs is token

frequency, as all of those verbs that have been remodeled are of a much lower frequency

than decir and hacer .

In spite of the high token frequency and similarity in paradigmatic structure between the decir , hacer AM and the velarinserts, there is one major complication.

According to the principle of the ‘homogeneous supracontext’ described in §1.3.3, digo , fago and the like are one step removed from direct analogical influence on the

velarinsert class because the –NGERE set is slightly more similar phonetically.

Therefore, the decir , hacer class should not be the primary AM for either the reconstructed nasalfinal forms, * teno , * veno , or the hypothetical palatals, * teño , * veño .

It is instead necessary to appeal to the phonological shape of the stem and the paradigmatic distribution of /g/ in the –NGERE set and, secondarily, to the 1s/SBJ patterning of an identical stemfinal velar in digo , hago , yago , etc.

2.4.2.3—AN ALTERNATIVE AM FOR VELAR -INSERTION . Besides the two preceding AM,

Lenfest (1978) and Elson (1988), among others, consider the possibility that the Spanish

93 verbs taking the highly productive Latin ‘inchoative’ infix –SC , illustrated in Figure 2.9, are the principle AM for velarinsertion. 58 The inchoative set is comprised of more than

200 verbs, mostly ending in –cer , including the highly frequent conocer “know, be

familiar with” and nacer “born”, as well as verbs having joined the group through analogical reanalysis, e.g. conducir “drive”, lucir “illuminate; shine”, producir

“produce”, placer “please”, yacer “lie (down)”, etc. (Martín Vegas 2007). 59 The inchoative infix was affected by the palatalization of velars before front vowels described in §2.3.2.3, leading to a similar 1s/SBJ pattern of stem allomorphy involving a (voiceless) velar consonant (cf. Fanciullo 1998). In fact, Elson’s analysis is based on the idea that the velar became an independent indexical marker of 1s/SBJ because of its inchoative function and that this independence (with loss of semantic function) was later extended to verbs having the same paradigmatic distribution of stem alternants as a result of phonological change (described above).

Lg INF Gloss IND SBJ # P #

CL COGNŌSCERE recognize, learn COGNŌSCŌ COGNŌSCAM 1 1s COGNŌSCIS COGNŌSCĀS 2 2s COGNŌSCIT COGNŌSCAT 3 3s COGNŌSCIMUS COGNŌSCĀMUS 4 1p COGNŌSCITIS COGNŌSCĀTIS 5 2p COGNŌSCUNT COGNŌSCANT 6 3p

MSp conocer know, be familiar with conozco conozca 1 1s conoces conozcas 2 2s conoce conozca 3 3s conocemos conozcamos 4 1p conocéis conozcáis 5 2p conocen conozcan 6 3p Figure 2.9—Example of the ‘inchoative’ verb set.

94

In a similar fashion, Martín Vegas (2007) suggests reanalysis of the velar as an inflectional element. In doing so, the verb stem remains unchanged, instead taking on an alternate inflectional marker, /C[+dors] V[+bk]/, in the 1s/SBJ (cf. Wilkinson 1980).

This is thought to more easily explain the extension of the velar to numerous verb forms including tener , venir , etc. and is supported by decir and hacer (Martín Vegas 2007).

One potential problem, though, is the analysis of /k/ and /g/ as variant inflectional

markers. Although these two phonemes differ only in voicing, their conditioning

environments are distinct enough so as to call into question their functional identity. It is

rather difficult to establish any salient relationship, from a speaker’s perspective, between

forms like conozco ‘1s know’ and vengo ‘1s come’, particularly when one considers the

availability of other more similar AM like digo or tango . Additionally, there is a pronounced difference in the phonological shape of the inchoative and velarinsert verb

stems. The former is typically characterized by polysyllabic stems, the majority of which

include some sort of derivational prefixation. The latter, excluding derived forms, are

exclusively monosyllabic (having a primarily CVC shape).

To avoid the problem of establishing phonological similarity, Lenfest (1978) uses

the inchoative infix –sc as motivation for the preference for a longer 1s throughout the verbal system. This hypothesis, like those above, points to the 1s/SBJ distribution of the velar in the present paradigm. While it must be granted that the inchoative set shares this allomorphic pattern with the velarinserts, as well as the decir , hacer and –NGERE subclasses, it cannot be considered the primary AM due to the existence of sources that share more relevant variables with the target under consideration. 95

2.4.2.4—COMPARISON OF POSSIBLE AM. While none of the preceding hypotheses is

sufficient on its own to explain the emergence of the velar /g/ in the 1s/SBJ forms of

verbs like tener and venir , it is logical to consider their combined influence. In doing so,

there are a number of important theoretical points to consider. According to traditional

AL analyses, the basis for the establishment of similarity or identity is phonological.

Accordingly, the analogical source for this process would be the set of –NGERE verbs because of the greater number of matching phonological variables in the conditioning environment. Note, however, that the attested 1s/SBJ alternation between velar (1s tango) and nonvelar (2s ta ñes /tanzes ) in the set of –NGERE verbs coincides to a high

degree with the alternation found in the present tense of verbs like OSp. dezir and fazer

(MSp. decir and hacer ), which arrive at an identical paradigmatic distribution of stem

alternation via a completely different series of diachronic developments. The same

arguments may also hold, though to a lesser degree, for the inchoative verb class. This permits a broader analysis where there exists a secondary component to the AM, one

having a morphological, rather than phonological basis. The allomorphic patterning of

1s/SBJ vs. the remaining present tense forms that emerged phonologically in verbs like 1s

tango , 2s tañes /tanzes (and later, through analogy, in velarinserts like 1s tengo , 2s

tienes ) is supported by the already existing paradigmatic distinction shown in the

decir /hacer group (e.g. 1s digo , 2s dices ) and the growing set of inchoatives including 1s

conozco , 2s conoces (cf. ‘Lpattern’ distribution, Maiden 2001). The everincreasing

96 type frequency of this paradigmatic template led to its eventual morphologization. These points are explored further in the subsequent chapters.

2.5—PREVIOUS EXAMINATIONS OF VELAR -INSERTION IN SPANISH . Prior to the 1960’s,

‘analogy’ with available AM, like those in the preceding section, was the basis for most, if not all analyses of velarinsertion. As a result, further explanation fell outside the realm of most theoretical constructs and was subsequently deemed unnecessary. It has been typically represented through broadly diachronic proportional analogical equations.

The problems associated with such highly generalized takes on analogy have been discussed in the preceding chapter. For a more complete (and immanent) picture of the situation, it is crucial to highlight a confluence of morphological, phonological and paradigmatic factors. Although this has been discussed in earlier works, e.g.

Klausenburger (1984), Martín Vegas (2007), which hint at the potential of the analogical function of the verbal paradigm, an examination of the combination and interaction of all the variables that characterize the AM has been lacking.

The next sections present the most salient explorations of velarinsertion in

Spanish according to different theoretical models. Brief discussion of some studies of the

Italian and Catalan developments are included as a point of contrast and to introduce some important concepts relevant to Spanish. Keeping in mind what has been outlined in the preceding sections, the discussion is limited to those theoretical aspects which in some way advance the examination of velarinsertion in the history of Spanish.

97

2.5.1—SYNCHRONIC RULE - AND CONSTRAINT -BASED APPROACHES . In contrast to those investigations described in the following sections, the rulebased approaches for Modern

Spanish do not explain velarinsertion but rather describe it and so do not (necessarily) appeal to analogy. In part, this is due to the fact that such analyses are purely synchronic and do not take into account the diachronic developments leading to the stage being described. The other reason is that they permit underlying representations (UR) as well as the application of rules which need not be surface transparent. In this way, the intraparadigmatic variation and subregularities that define the group of velarinserts can be derived from hypothetical sources by hypothetical rules.

While such lack of transparency allows for the derivation of multiple forms from the same base, e.g. and verb, it requires one or more marked alternatives in order to correctly provide the proper surface representations. Harris (1972) exemplifies some of the principle techniques for deriving the velarinsert in verb forms. A single UR for all inflected (and derivational) forms is proposed, e.g. /val/ as the stem for forms of valer .

The TV is added to this stem, even in 1s/SBJ forms where it is not surfacerealized. This verb is lexically marked as a velarinsert in order to trigger minor rule application which inserts the velar according to the presence of the TV and a following vowel (in this way distinguishing the 1s/SBJ from the remainder of the IND). The TV is later deleted by rule in these forms (254; see also Cressey 1972, Fancuillo 1998, Klausenburger 1984).

The combination of a hypothetical UR and lexical markedness means that rules can apply as broadly as possible and prevents velarinsertion in potential targets sharing the same phonological conditioning environment. In this way too, the aforementioned problem of

98 explaining those forms which deviate from regular phonological developments is avoided altogether by simply proposing a different UR and/or by marking them in another way in the lexicon.

Obviously then, the synchronic rulebased approach does not require an immanent perspective, nor does it have its basis in the similarity between exemplars. Also, the function of the paradigm is irrelevant as is analogy as a whole. Harris (1972) acknowledges that it does not seem possible to answer the question about the psychological reality of such rules. While it is certainly possible to make a case for the involvement of lexical marking or variable rules in velarinsertion, in the end, such analyses imply something more abstract and, thus less elegant, than the immanently acquired surface generalizations required for an analogical usagebased approach like

AEM. Allomorphy need not be reduced to invariance in UR (or ), but instead should be embraced as an “active, abstract structural property of morphological systems” utilized by speakers (Maiden 1992:285).

Although there have been no specific investigations of velarinsertion, constraintbased approaches do provide some possible solutions which, more often than not, resemble their rulebased counterparts. One possible hypothesis involves

OutputOutput Faithfulness (or Correspondence; henceforth OOF), which constructs an identity relationship between forms allowing analogical influence between them (cf.

‘Paradigm Uniformity’, Steriade 2000). This gives priority to localized paradigmatic identity over more generalized or regular patterns in the language and minimizes the relevance of the Input (to a degree, see below). These are ranked with respect to

99

InputOutput Faithfulness (IOF; cf. §1.2.2). Myers (1999) uses these types of

constraints and rankings to propose a general means of determining the outcome in the

analogical target according to the proportional analogical model. As a simplified

example, consider the analogical influence of 1s salgo in determining the presence of the velar in 1s valgo according to the data presented in chapter 3. Given a proportional equation sales :salgo :: vales :X, where X is valgo , there is necessarily an IOF constraint for each form in the equation where they are faithful to the presence or absence of the velar in the Input, i.e. IDENTIO ( sales ; /g/), IDENTIO ( salgo ; /g/), IDENTIO ( vales ;

/g/), IDENTIO ( valgo ; /g/). In order to attain the velarinsert form valgo , the last of these IOF constraints must be ranked lower than a related OOF constraint, IDENTOO

(salgo , valgo ; /g/). This will provide an optimal outcome in which both salgo and valgo share the same outcome with respect to the velar regardless of whether or not the Input for valgo contains the velar. Unfortunately, the OOF constraint above forces an arbitrary analogical change and does not explain the morphological and phonological relationship between the forms in the proportional model. Thus, Myers (1999) proposes

OOF constraint conjunctions between the forms having the same outcome for the feature under analysis. In the current example, the relevant conjunction is IDENTOO ( sales , vales ; /g/) ^ IDENTOO ( salgo , valgo ; /g/). This ‘morpheme structure constraint’ takes the place of the earlier OOF constraint.

Now consider an actually attested situation in which the outcome for 1s valgo does not contain a velar, i.e. valo . In this case, the solution is merely a shift in the constraint rankings such that IDENTIO ( valgo; /g/) is ranked higher than OOF. 60 The

100

assumption is that the Input does not contain a velar and so corresponds to the changes in

the UR in a rulebased approach. If the Input has a velar and the Output does not, a new

OOF or markedness constraint is required which promotes the nonvelar, e.g.

IDENTOO ( vales , valgo ; /g/). Once such a change is established, the Input is believed to be reanalyzed as being nonvelar (see below). A problem that arises with the reranking of constraints proposed here is its motivation, as it seems that such a change is based on the Output (or possibly the Input to a different grammar).

Finally, it is necessary to discuss the synchronic variation between velar and nonvelar forms appearing at an early stage of development. In order to attain both 1s valgo and valo , the above constraints are reranked such that IDENTIO ( valgo ; /g/) is so

low as to not be involved while the OOF constraints IDENTOO ( sales , vales ; /g/) ^

IDENTOO ( salgo , valgo ; /g/) and IDENTOO ( vales , valgo ; /g/) are unranked with

respect to each other. This provides both outcomes as optimal candidates. For the

tableaux of similar examples to each of the cases discussed here, see Myers (1999).

These OOF and morpheme structure constraints permit paradigmatic forms to

incur violations of high ranking constraints that would make nonparadigmatic forms

infelicitous (cf. ‘Paradigm Uniformity’, Steriade 2000, Burzio 2005). Thus, they are

formalized and ranked in such a way as to promote the 1s/SBJ distribution of the velar in

only those forms where it occurs. Also, most other OOF constraints need to be of a

relatively low rank, as this allows for paradigmatic allomorphy (cf. Kenstowicz 2005).

Unfortunately, the examples above makes use of highly similar techniques to

those seen in rulebased analyses. First, the necessary constraints are language specific

101

and apply to specific lexical items and features. While this demonstrates the relationship between the forms in the proportional equation, it violates the universality of constraints proposed in traditional OT, though doing so out of necessity. If such universality were to

hold, the constraints necessary to explain velarinsertion would be limited by a lack of

specificity that may result in potential problems. Second, the OOF constraints are quite

like the lexical marking that triggered minor rule application. Third, changes in

constraint ranking and/or the Input can be manipulated to determine the synchronic

result. If the conjunction constraints are not required, the information they convey is

thought to be expressed in the Input (Myers 1999). This does not truly predict potential

change, instead describing the state of grammar at a particular point in time. Finally, the processing load implied by solutions of this sort is in no way realistic from an immanent point of view, in spite of the numerous means by which such procedures are potentially

restricted (see Myers 1999). As indicated previously however, it is notable that in OT the

there is an established relationship between outputs and that the paradigm is given a

important function in the organization of linguistic information. 61

2.5.2—AN ANALOGICAL PERSPECTIVE ON VELAR -INSERTION . Historical examinations

of the problem, like their synchronic counterparts, have overlooked a crucial aspect of the

development of the velarinsert category. However, instead of ignoring the evolution of

the verb class, they typically present the emergence, extension, and resulting synchronic

variation too broadly. The consequence is an explanation loaded with sweeping

generalizations that often leads to more questions than it answers. Such has often been

102

the case when invoking analogy as the motivation for language change. A boldfaced

view of analogy, one in which the concept is not welldefined, has tended to serve the purpose of actually avoiding real explanation.

This section presents works that have appealed to analogy as a means of

explaining and describing the emergence of the velarinsert class. However, unlike many

of the investigations mentioned in §2.4.2, the explanation in these is sufficiently thorough

so as to avoid the traditional pitfalls associated with the use of the term analogy. Besides

having a more properly conceived notion of the analogical process, each in some way

expands what is known about velarinsertion, be it with respect to the origin of the verb

class, its extension, both diachronically and lexically, or concerning the AM and its

theoretical consequences.

2.5.2.1—TRADITIONAL ANALYSES . There are numerous proponents of the two most accepted theories concerning velarinsertion as exemplified in §2.4.2. Among these, several examinations stand out for the amount of detail and innovation that they bring to the table. Although grounded in traditional theoretical approaches to the problem, these investigations are often cited for the way in which they defend a particular point of view and the supporting arguments they provide with respect to the former.

2.5.2.1.1—SYNCHRONIC COMPETITION BETWEEN VELAR -INSERTION AND

PALATALIZATION . One of the clearest descriptions of the proposed synchronic

competition between a palatal (e.g. * teñ ) and a velar (e.g. teng ) stem alternant is found

103

in Lloyd (1987). His analysis, like many others, suggests that rootfinal /n/ was palatalized by the antihiatic yod described and that the Latin –NG cluster suffered palatalization when followed by a front vowel. This created 1s/SBJ allomorphy in the verb paradigm according to the frontness of the following inflectional vowel. The resulting stem alternation was resolved in Spanish by leveling in favor of the palatalfinal stem. Prior to leveling, however, Lloyd suggests, following Wilkinson (1978), that the –

NGERE set (having a large type frequency) and verbs having a palatalized nasal, like tener

and venir (having a high token frequency), showed “some mutual analogical influence”

such that the palatal nasal came to alternate with the nasal plus velar stemfinally (Lloyd

1987:164165; cf. similar variation in Italian as discussed in Bourciez 1967).

The necessary processes of palatalization are attested prior to those localized

developments affecting individual Romance dialects/languages (emergence of yod by the

2nd c. AD and palatalization by the 4 th c. AD). However, the contrasts between Spanish

and Portuguese and the similarities between Spanish and Italian indicate that

velarinsertion did not affect all the Romance dialects in the same way. It suggests that

either there was a parallel, unrelated evolution of the velarinsert class in Spanish and

Italian (as well as Catalan and Old Provençal; see §2.2 above) or that velarinsertion may

have already begun in Romance, creating variation between palatal and velar stems, i.e.

/t εɲ/ ~ /t εng/, which was resolved independently in localized domains. Lloyd supports

the latter, with individual dialects/languages experiencing varying degrees of alternation between palatal and velarfinal stems. Because there are no attested cases of a palatalfinal stem in tener , venir and the like, the aforementioned competition is believed

104

to have occurred “at a very early stage of Castilian, or perhaps early northcentral

IberoRomance” (Lloyd 1987:294). In fact, a primary characteristic of

HispanoRomance was the early elimination of alternations of this sort (Lloyd 1987). In

the case of Spanish and Italian, then, it is the availability of an existing AM that

determines the selection of the velar stem over the palatal one. 62

In other words, Lloyd (1987) follows the traditional analysis that the verbs tener and venir were influenced by the –NGERE set, based on phonological and paradigmatic

similarity, and that velarinsertion was also supported by the paradigmatic distribution

(and developing morphological/indexical function) of the velar in verbs like decir , hacer , yacer , aducir , etc. Once these verbs had adopted the velar, poner was not far behind because of its unique status as the only other nasalfinal CVCstem in Conjugation II.

The subsequent stages of extension are based on phonetic similarity, i.e. single syllable

CVCstems with a sonorantfinal consonant belonging to conjugations II and III. The nonsystematic appearance of the velarinsert in these conjugations is based on token frequency: “In Castilian, only the very commonest verbs adopted /g/, however, while the majority of verbs followed the road of regularization” (Lloyd 1987:296). Subsequent extension of the velarinsert to yodfinal stems is due to their having the same 1s/SBJ pattern of paradigmatic allomorphy which increased the likelihood that analogical restructuring could occur. Thus, the apparent, unconfirmed chronology proposed coincides with that of this investigation, namely that the appearance of the velarinsert follows a direct path along the Sonority Hierarchy (cf. §3.2).

105

For the most part, the preceding analysis is and straightforward though

there is no detailed discussion of the paradigmatic factors and evershifting AM that

contribute to the extension of the velarinsert. This is due in part to the fact that the

development of the velarinsert is primarily treated phonologically, according to the

stemfinal segment. In fact, a number of the forms included in the proposed AM must be

discounted since they do not reasonably fall within the domain of the homogeneous

supracontext at a particular synchronic stage. Also, the diachronic developments and

synchronic stages related to class membership are only briefly mentioned, making any proposed chronology difficult to establish. Finally, attributing the extension of the velar beyond the initial stages to phonetic similarity and frequency is insufficient in explaining

the appearance of the velarinsert in many mid to lowfrequency forms, e.g. 1s asgo , or

the synchronic variation that characterizes many stages of development.

2.5.2.1.2—‘M ULTIPLE CAUSATION ’ AND A REVISED VIEW OF ANALOGY . The preceding points are more thoroughly examined by Yakov Malkiel in a number of investigations that span more than three decades (including, for example, Malkiel 1967, 1969, 1974,

1982, 1989). Among these, Malkiel (1974) stands out for its analysis of the extension of the velar over time as well as for providing major conceptual insight into the analogical process involved. Earlier works, while notable for their innovation with respect to the matter, argue certain points which are generally discounted or discarded in favor of more plausible and wellestablished hypotheses.

106

Following the development of the antihiatic yod and its palatalizing effect,

Malkiel reconstructs */tenjo/ or */teɲo/, */venjo/ or */veɲo/ as the Late Latin predecessors for these earliest members of the velarinsert class. Depalatalization is then

a simple phonological change triggered by the immediately following velar consonant

(Malkiel 1974). Although the decir , hacer set of verbs is mentioned as an AM (cf.

Malkiel 1967), the primary model for velarinsertion is claimed to be the set of –NGERE

verbs discussed previously. Unlike most other analyses which assert some sort of mutual

analogical influence between model and target (i.e. taño :tango ::* teño :X, where

X= tengo ), Malkiel believes that the –NGERE set is the leading AM because it shares a high degree of both phonological and allomorphic/paradigmatic similarity with the velarinsert targets. In other words, it is not the synchronic variation but rather its resolution in favor of a (salient) 1s/SBJ alternation (supported by the result of regular sound change in verbs like decir ) that provides an explanation for the insertion of /g/ in

verbs like tener, venir and poner . A consequence of such a point of view is a rethinking

of the traditional notion of ‘analogy’:

The classic concept of ‘analogy’ – long ago recognized as vulnerable – consequently invites refinement. It is not enough to contend that the preexistence of, say, OSp. tango ‘I touch’ (with it –ng acquired through bequest) triggered the shift from * teño  tengo TENEŌ – though it undoubtedly served as one of the models imitated in the process of adjustment. This ageold formulation has invariably provoked the objection that OPtg. tango , ‘all circumstances being equal’ (or so it seemed at least), parodoxically, failed to produce the same effect. That objection can now be parried with the remark that the availability of a good model is not, per se, the CAUSE of an analogical shift, though it may pre or codetermine it specific DIRECTION : TENEŌ developed differently in the two sister languages on account of clashing morphophonemic conditions. (Malkiel 1974:345)

107

Although the preceding hypothesis differs only slightly from the traditional view, it correctly places increased emphasis on the paradigmatic distribution of allomorphy.

Nonetheless, the strength of Malkiel’s assertions has been called into question as being too weak because his earliest analyses treated velarinsertion as a “therapeutic device”

(Lenfest 1978). In spite of such criticism, it is possible, especially considering later investigations, to see this analysis as moving in the right direction regarding the function of the paradigm in analogical change. Through phonological developments which resulted in an allomorphic distribution of the velar in the –NGERE class, the 1s/SBJ alternation of stemfinal consonants became the primary identifiable characteristic of the innovative paradigm (equating to the more contemporary notions of ‘morphome’ or

‘paradigmatic template’ discussed below). Once converted into an AM, other verbs, potentially along with other parts of speech, may become subject to its influence. Such interparadigmatic influence, while nothing new (cf. Ford 1966, Spaulding 1971), has only recently begun to gain theoretical relevance (cf. §§2.5.1 and 4.2.2).

On the basis of this wellfounded hypothesis regarding the emergence of the velarinsert, Malkiel tackles the much less commonly discussed issue of its productive diachronic extension and subsequent (near)crystallization. Malkiel’s intuitions concerning the spread of the velar to lateral, rhotic, sibilant and yodfinal stems are quite valid in the broader historical view. One of the primary distinguishing characteristics of the current investigation is the suggestion that the target forms of a previous synchronic stage can then become the primary AM for following generations of speakers (cntr. Lloyd 1987 above). That is to say, once the velarinsert becomes

108

established in the nasalfinal verb stem, these forms act as the primary model for

lateral and rhoticfinal stems, which, in turn, provide additional analogical support for

the extension of the velar after yod. While the cost of such a hypothesis is minimal, it is

extremely valuable. Referring again to the principle of the ‘homogeneous supracontext’,

i.e. that the forms in the model dataset share the highest degree of similarity with target

forms, the previous stage’s AM is replaced by one that is more similar to its target forms.

In the transition from target to model, the nasalfinal verbs are more similar to their

lateralfinal counterparts than any previously used AM. Consider the fact that the decir ,

hacer set lacks the CVC g stem shape and that the –NGERE set would have already leveled in favor of the palatal stem. The lateral group had no such obvious AM since the set of –LGERE verbs (mainly MULGERE , COL (LI )GERE ) which survived into Spanish were

severely limited in both type and token frequency (cf. Rohlfs 1968, Fanciullo 1998). The

Italian situation differs in this respect as there was a more prominent group of lateral plus

velarfinal stems which pertained, according to some analyses (e.g. Fanciullo 1998,

Maiden 1992, 2001, see below), to the AM used in the initial stage of velarinsertion for

this language. With respect to Spanish, though:

[I]t is the triumph of –ng , as in pongo , that in the end paved the way for partial generalization of –lg , as in salgo , and the occasional appearance of –rg , as in ¡ firgamos ! ‘let us strike!’ ( Cid ) closes the cycle, being essentially due to the intrinsic, paradigmatic affinity of n, l, and r. (Malkiel 1974:326)

Nonetheless, the actual synchronic processes, as well as their specific chronological order, still require a much finer degree of analysis than Malkiel provides, as demonstrated by the identification and examination of the velar in individual verbs found in chapter 3.

109

The extension of the velar after yod is seen by Malkiel, following Menéndez

Pidal’s analysis, as a later (postmedieval) development (1974). In the verbs affected, the velar (actually yod plus velar, /jg/) is seen as an analogical replacement for a palatal consonant, e.g. caigo << cayo (containing a “secondary” or analogical palatal on the basis of forms deriving it from the yod effects discussed earlier, e.g. FUGIŌ > fuyo ).

Another noteworthy aspect of Malkiel’s analyses is the emphasis placed on

‘multiple causation’, motivated by the convergence of both phonological and morphological factors.

[A]t first glance they appear to be purely phonological in content, but as soon as the analyst turns his attention to the matter of causation, strictly phonological conditions fail to provide any adequate, truly dependable clue. This qualification does not exclude the possibility that such conditions played a subordinate role…but the single most plausible factor of causation turns out…to have been analogy… the agency that transmitted the impact was a set of crucially important, rather similar verbal paradigms. (Malkiel 1967:62, cf. Joseph 1982, 1983)

This perspective resembles that of Maiden (2001) which stresses the importance of morphology while assigning secondary status to the phonological aspects that condition the analogical change. While it is true that the morphology is a salient factor in both confining and expanding the domain of such developments, it should be noted, for velarinsertion at least, that the morphology is dependent to some degree on the phonological conditioning environment. Multiple causation is pursued even further in the current investigation on the basis that the associations made between forms include all types of linguistic and nonlinguistic information (§1.3.1). This is demonstrated by the number of linguistic variables that define the AM proposed in §3.2.

110

Another major insight into the development of the velarinsert verb class involves the effect of different types of analogical influence that are visible in the changes in category membership over time. Verbs entered and exited this verbal subclass due to the effects of competing analogical forces. In many instances, the velar forms were overturned in favor of the (often original) invariant (or nearly invariant) root that had appeared previously or as a synchronic variant. Malkiel describes this type of situation as follows:

Thus, one (minor) analogical pressure within a verbal paradigm can deflect a few of its members from the expected path, onto which they may, three or four generations later, be pushed again through a stronger pressure exerted from a different direction. How can such adjustments (backspins) be detected where the early record is fragmentary or seems irretrievably lost? (Malkiel 1967:26)

Because analogy is not a systematic process and given the possibility that analogical influence may assert itself from multiple directions, it is not surprising that such ‘return to form’, so to speak, should take place.

If such a state of dissatisfaction with the inherited status quo prevails, several avenues of escape – in search of improvement – are open to speakers. Conceivably the most arresting is the substitution of a new and more effectivelymarked form for one old and wornout: e.g., Sp. tengo for * teño (as preserved in Portuguese, with a different ). (Malkiel 1974:344)

Finally, the preceding discussion presupposes Wiese’s (1928) perspective on the composition of the AM, i.e., the “inclination to charge secondary changes to the pressure not of a single form, however influential, but to entire constellations of such forms…”

(Malkiel 1974:342; cf. Joseph 1997a). It also stresses the importance of detailing, as

111

scientifically as possible, the analogical forces at work in language change, a concept

which closely parallels similar notions which are inherent to AEM.

2.5.2.1.3—REINFORCEMENT OF 1S/SBJ ALLOMORPHY . The PanRomance study of velars, etymological and otherwise, undertaken in Wilkinson (19781983) is notable for the amount of data it offers from across countless dialects. Unfortunately, the sheer quantity of forms obscures any major conceptual breakthroughs regarding velarinsertion.

There are, however, several important points hidden amongst the presentation of verb forms having a stemfinal velar consonant. Wilkinson (1978, 1982) believes that the NGERE set is the main AM for tener , venir , poner , and remanir while it is suggested

that the decir , hacer group is the main AM for verbs like traer , huir , raer , creer and ver ,

in light of the fact that such a hypothesis goes against any notion of the homogeneous

supracontext, since the chronology of the changes affecting these verbs is indicative of

the former target serving as the AM for the latter.

The overall approach coincides in all respects with previous analyses. The

conclusions drawn focus primarily on the effect of the yod in the Romance languages and

the data is drawn heavily from Catalan, due to the high degree of productivity of

velarinsertion in this language (for reasons explored earlier). The study of the Spanish

situation begins with palatalization of the stemfinal nasal resulting in allomorphic

identity in some cells of the –NGERE and velarinsert target sets. From there, the velar is

taken on in order to create a more salient allomorphic pattern, based on previously

existing models and, accordingly, it appears to be assigned an indexical function (cf.

112

morphome, below). It is argued that velarinsertion is not a singular phenomenon

traceable back to a common Romance ancestry but rather is the result of some earlier

stage which provided a form or set of forms which could expand in the numerous

directions found in the later separate developments of the individual Romance languages:

The general conclusion that emerges is that the forms based on Latin stems containing a yod have only been preserved in rare cases, and have otherwise succumbed to the pressure of the forms based on stems without a yod or have been remodelled on the analogy of other verbs containing a velar. The reason for this must be that the velar provides a distinctive mark of person or mood and is common to many classes of verbs…What is striking is that this tendency to introduce a velar should have been strong enough to have appeared independently in so many areas. It can hardly be traced back to the period of Romance unity, but we can perhaps surmise that each language inherited some spark of a common genius that was already inherent in Latin, and that this expressed itself concretely wherever the opportunity offered. (Wilkinson 1983:193194)

The explanation of the emergence of the velar as an indexical marker of 1s/SBJ paradigmatic allomorphy is based on that of Malkiel (1967), which Malkiel himself corrected in subsequent works. According to this hypothesis, increased salience of allomorphic distinction is preferred such that a previously palatal/nonpalatal alternation was replaced by the velar/nonvelar one (cf. the ‘entropic’ function of the verbal paradigm, Maiden 1992; cf. Malkiel 1974). Added support is believed to derive from the avoidance of homophony or ambiguity between the finite forms of some verbs, e.g. tener /teñir , poner /puñir (Wilkinson 1978). Such arguments have since been shown to be less than desirable because there are instances of (phonemic) palatal/nonpalatal alternations and in Spanish, though not exclusively in the verbal system (cf.

Lenfest 1978). 63

113

In spite of these shortcomings, the emergence of the 1s/SBJ morphome as well as the confluence of multiple characteristics in a single AM (here, phonological and paradigmatic similarity) are correctly identified. Similarly, the influence of the –NGERE set (as well as the –LGERE and –RGERE sets) is acknowledged as being preliterary, at a

stage of the language where their frequency was high enough to permit them to serve as

the AM for the high frequency tener , venir , etc. The conclusion then is that

velarinsertion only occurred in languages where the –NG (and LG ) roots followed by front vowels produced palatals (Maiden 2001). This conditioning environment provided a means for analogical reanalysis such that /ng/ could replace /ɲ/.

2.5.2.1.4—VELAR -INSERTION IN LIEU OF PALATALIZATION . Alvar & Pottier (1983) describes the development of the velarinsert class as an external analogical effect modeled on already existing forms having a similar shape. The treatment of the forms affected by this process is organized around the sound preceding the velar element, which corroborates the broad timeline acknowledged throughout the literature, as it moves from nasal to lateral to rhotic to yod. 64 The investigation largely follows that of Menéndez

Pidal (1980), specifically the view that velarinsertion is morphologically motivated in verbs stems according to available AM of similar phonological (and, implicitly, paradigmatic) structure, in this case, the set of –NGERE verbs. In this instance, the relationship between model and target depends on the (hypothetically expected) identical palatalization of stemfinal nasals by way of the previously discussed /g/ plus front vowel and yod effects (cf. §2.3).

114

Additionally, some plausibility is given to the ‘therapeutic’ explanation mentioned in §2.5.2.1.2, which contends that if yod does not palatalize stemfinal nasals, as in tengo and vengo , cntr. * teño , * veño , it is because the result would have created a complex alternation (‘duality’) in the paradigm, e.g. * teñ o, tien es, ten emos , etc.

(Alvar & Pottier 1983). In essence, analogy is believed to prevent normal phonetic outcomes that would be morphologically anomalous (e.g. 1s *veño but 1p venimos ). An

important observation to consider here is that the alternation between palatal and

nonpalatal phonemes is based on the contrast of a single phonological feature. In the

modern language, there is no such alternation within the stem of a single verb and if there

were, it would result in serious morphological problems for the selection of a stem in

inflected forms across tense and mood. Instead, the established paradigmatic alternations

are of greater phonetic salience, e.g. suppletion (e.g. 1s quepo, 2s cabes),

(1s salgo, 2s sales), diphthongization (1s duermo, 1p dormimos ), or a combination

of these (1s tengo, 2s tienes, 1p tenemos ). Therefore, the alternation of palatal and

nonpalatalfinal stems is deemed a “phonetic perturbation” that is not permitted in

Spanish (Alvar & Pottier 1983:221).

The velar, then, initially emerges as a replacement for palatalization (similar to

the ‘indexical marker’ in Elson 1988, cf. Hooper 1979 and Fanciullo 1998 among others), providing the effect of maintaining the templatic structure and stem allomorphy of the

verbal paradigm. The primary difficulty with this analysis lies in the fact that the –NGERE paradigms are eventually leveled in favor of the palatal stem. The proposed solution relies on intraparadigmatic influence (‘morphological analogy’) in the same direction for

115 both the model and target sets, from the majority of forms in the IND toward the 1s/SBJ,

the result in one case being palatal and in the other, nonpalatal/velar. However, such a

hypothesis requires further explanation in that at some point before the leveling of the –

NGERE class, the nasal plus velar sequence came to be seen as roughly equal to the palatal

nasal in stemfinal position when in 1s/SBJ paradigmatic alternation with a nonpalatal

nasal. This allows the nasalplusvelar stem to appear where a palatal nasal is expected in

alternation with /n/. Subsequently, analogical interference from yet another AM, i.e.

forms like bendigo , yago , cuego , etc., led to reinterpretation of the nasalplusvelar stem

as being nasalfinal with the velar /g/ serving as a linking (or ‘indexical’) element between stem and inflection (Alvar & Pottier 1983). This analysis is analogically

extended to lateralfinal stems (including both palatal lateral and lateral plus velar), i.e.

ñ:ng :: ll :lg (Alvar & Pottier 1983). Similarly, though more problematically, it is later

thought to have affected rhotic and glidefinal stems.

While convincing in some respects, the analysis does lack certain crucial

elements. First of all, while it is indicated that there are some verbs in which the velar is

interpreted as a linking element, making them possible AM for velarinsertion, no

specific mention is made whatsoever of the potentially significant analogical influence of

the verbs decir and hacer . Relatedly, the influence of the paradigmatic structure of the

verb is implied though its effects are not always clear nor straightforwardly proposed.

Intraparadigmatic analogy is demonstrated in leveling while interparadigmatic

analogical effects are limited to individual forms, e.g. 1s forms restructured according to

116

other 1s forms, with no regard for the templatic structure that seems to also be relevant in

these instances.

2.5.2.1.5—INTER - AND INTRA -PARADIGMATIC INFLUENCE . Among those investigations which suggest that both the set of –NGERE verbs and decir /hacer were influential from the

outset in velarinsertion is Penny (2002). This combined AM, specifically appealing to

its inter and intraparadigmatic influence and the emergence of a paradigmatic template,

is at the center of the analysis (Penny 2002; cf. Rebrus & Törkenczy 2005).

Interparadigmatic pressure is crucial since the –NGERE and decir /hacer sets showed

1s/SBJ allomorphic variation in opposition to the other forms of the PRS, thereby

creating a templatic structure supported by a good number of high frequency verbs.

Meanwhile, intraparadigmatic influence is essential in the leveling of the allomorphy

found in the –NGERE class (i.e. cingo , ciñes , etc. >> ciño , ciñes , etc.). It also serves a

similar function diachronically as some velarinsert verbs are reanalyzed as having a

single, nonvelar verb stem. In all other respects, Penny’s examination of the issue

resembles that of Malkiel discussed above, though with additional treatment of the

function of frequency in analogical change (cf. §4.3).

2.5.2.1.6—STEM ALLOMORPHY AND PHONOLOGICAL BALANCE . As mentioned earlier, one of the major proponents of the resistance to palatalization, due to early loss of yod, is

Lenfest (1978, 1993). These works propose the reconstructed (possibly underlying) 1s forms *teno , * veno and * pono and provide several arguments for doing so. First, there

117

are a very small number of attested instances of 1s salo and valo , which share a

“canonical form”, i.e. CVCV (Lenfest 1993:634). In Spanish, these verbs show neither palatalization nor the glide resulting from loss of hiatus between adjacent vowels in

Vulgar Latin (cf. MPtg. tenho , venho , ponho ; OIt. tegno ~ tengo , vegno ~ vengo ).

Second, given that the 1s of the verb sentir is siento , not * sienzo , it is believed to derive from a reconstructed 1s *sento , having a raised stem vowel and no yod (cf. CL SĔNTIŌ ,

cf. Menéndez Pidal 1980). The third motivation for this hypothesis involves the verbal paradigm, which is believed to be resistant to innovation such that allomorphic variation between forms like * veño ~ vienes would be avoided.

In addition to the preceding arguments, the increased productivity of the inchoative infix –SC in VL is used to suggest that there came to be a “preference for

longer 1 st pers. stems in those languages in which the velar of the –sc cluster resisted

assimilation” (Lenfest 1978:900, cf. the ‘indexical function’ of the velar in Elson 1988,

Hooper 1979, Fanciullo 1998, Moll 1952, as well as Pérez Saldanya 1995b and the

related indications found in Lloyd 1987, Maiden 1992, Malkiel 1974 and Martín Vegas

2007). This preference for longer stems in 1s forms, overlapping with a shift from

CVCVtype to CXXCV in a large group of verbs, resulted in velarinsertion in * teno and

*veno for “therapeutic” (i.e. paradigmatic and phonological) reasons. 65 In the case of

*pono , salo and valo , the motivation for velarinsertion was instead “cosmetic”, i.e. because their phonetic shape made them potential targets for the process. In the end, the

combination of factors such as the presence of digo , hago , yago (showing stem

allomorphy), vengo (from vengar ), tango , along with the related set of (learned) –NGERE

118

verbs (supporting the shape of the verb stem), and the inchoative infix (promoting a velar

marker of 1s) are all acknowledged as having attributed to the velarinsertion found in

tengo , vengo and pongo (Lenfest 1978, 1993).

The primary difficulty with such arguments is that they select the AM for

velarinsertion from a conceivably unrestricted domain according to the nature of

traditional analogical explanation. At the same time, they disregard the contradictory or

opposing influence that is found across these model verbs (e.g. the 1s/SBJ stem

allomorphy in the decir /hacer set in contrast to the single stem of 1 st conjugation vengar ).

In other words, what is provided is a set of potential AM whose actual role is unclear at best and, in some cases, either clashes with the influence of other verbs in the model, or is

too far removed from the equation to truly be useful since there are other more similar

analogical triggers available. In other words, anything can and does serve as the model

for analogy. Secondarily, the ‘cosmetic’ explanation affecting verbs like poner , salir and

venir seems to contradict the ‘therapeutic’ repair strategy for tener and venir . Given the

1s forms in the preceding paragraph, the ‘cosmetic’ group had achieved paradigmatic balance by leveling to a single verb stem throughout the inflected paradigm (with the possible exception of the FUT and COND). The phonetic balance attained through

velarinsertion in tener and venir does not apply, which requires appealing to the

‘cosmetic’ preference for longer 1s stems, in contradiction to attested forms having both a

velarinsert and diphthong (e.g. suelgo , duelgo , fiergo ) as well as those affected by

leveling.

119

Several other difficulties also arise. First and foremost is the lack of explanation for the appearance of /g/ in the SBJ. Ignoring the obvious morphophonological conditioning environment, there is no salient motivation provided for the partition of the

1s from the remainder of the IND. Supporting evidence is also lacking for the idea that the phonetic substance of a form like vengo (CVCCV) is equal to that of viene (CGVCV).

Finally, as is the case with the overwhelming majority of the previous research on the

subject, any treatment of the extension and loss of productivity of the velarinsert class beyond the initial stages (i.e. tengo , vengo , pongo , salgo , valgo ) is completely absent, although forms like duelgo , suelgo , fiergo , caigo , vaiga , etc. are mentioned briefly.

In spite of these shortcomings, however, Lenfest provides support for the linguistic reality and function of the verbal paradigm. In particular, it is suggested that:

[I]n a period of rapid evolution, if not always, paradigmatic forms are under tension and are subjected to pressures of varying intensity and origin, that they may be shaped and reshaped until they assume a form which conforms to the overall pattern or are otherwise felt to be members of a special subclass. Both regular sound change and analogical sound change can create forms which break the conformity of verbal paradigms; however, it has been demonstrated that a regularizing tendency will usually reassert itself and that speakers will have recourse to a wide variety of devices in their efforts to reestablish balance within the system. (Lenfest 1978:903)

The preceding paragraph summarizes quite succintly the diachronic development of the velarinsert class whose members are each uniquely susceptible to competing analogical forces, each of which has a particular set of defining variables. Thus, it is crucial to appeal to the synchronic stages at which the velarinsert members entered the class. This is the primary topic of discussion in chapter 3.

120

2.5.2.2—CONTEMPORARY ANALYSES OF VELAR -INSERTION ACROSS ROMANCE .

Although hinted at previously, the distribution of the velar element in the paradigm has only more recently been considered a relevant part of the analogical process. The following investigations examine this ‘templatic’ structure and, at the same time, provide additional insight into the development of the velarinsert subclass. In those cases where the primary language of study is Catalan or Italian (in particular, Pérez Saldanya 1995a,

1995b, and Maiden 1992 respectively), the conclusions discussed here specifically apply to Spanish.

2.5.2.2.1—THE VELAR -INSERT AS AN INFLECTIONAL COMPONENT . Martín Vegas (2007)

considers the possibility that the velar element is morphemic rather than representative of

stem allomorphy. The velar creates allomorphy in the inflectional markers for 1s/SBJ

(cf. the ‘infix’ status given the velar in Fanciullo 1998). Taking the example of MSp.

variant 1s yazgo , the stem yaz is unchanged and instead takes on an alternant inflectional

marker /Ko/, where /K/ is a velar consonant that may or may not be marked for .

The goal is to more easily explain the extension of the velar to numerous verb forms

including tener , venir , etc. and is supported not only by decir and hacer but also by the inchoative group (Martín Vegas 2007).

Este reanálisis [of /K/ as part of the inflectional marker for 1s/SBJ] sería el foco de difusión de un modelo de flexión que se caracteriza por tener en la 1a p. del presente de indicativo la desinencia go en el presente de subjuntivo ga , gas ... El esquema tiene relativa fuerza léxica porque afecta inicialmente a verbos muy frecuentes... Algunos verbos menos usados que forman parte de este esquema acaban nivelando la alternancia: p. ej. cingo  ciño . Pero la alternancia se mantiene en verbos tan frecuentes como digo /dices , pongo /pones , salgo /sales ...Con un análisis de 121

este tipo no hablaríamos de una inserción de g en determinados verbos, sino de la extensión de un esquema/modelo de flexión que se caracteriza porque la 1 a p. del pres. ind. y el pres. subj. terminan en go , ga , gas ... (Martín Vegas 2007:168) 66

A potential problem with this analysis is that, from an immanent perspective, it

adds allomorphy to another level of the grammar which was previously systematically

rigid. In essence, it complicates the grammar in two ways: by altering an otherwise

stable part of the system and by not simply making use of the established contexts for

allomorphy.

Additional problems are related to dialectal variation and the heavy burden placed on token frequency. It is claimed that the voicing of the velar element is predictable according to the phonological context in which it occurs. However, this is does not appear to be the case. The 1s variant yazgo mentioned above contradicts the

of 1s conozco . Likewise, the presence of the velar in forms of very low

frequency, e.g. 1s asgo , roigo , and raigo , complicates the function of frequency such that

it cannot be relied upon as a primary factor in the extension of the velar element,

regardless of whether it is independent of stem, inflection or both. Another aspect of this

issue is the question of the productivity of this particular inflectional element. What are

the characteristics which trigger its influence, especially with respect to the traditional set

of inflectional markers? How does the phonological conditioning environment relate to

the paradigmatic structure, the token frequency of the verb, etc.? What factors contribute

to the leveling of the paradigms of verbs like ceñir described above? It seems more

appropriate to focus on the intra and interparadigmatic relationship between stem and

inflection, and their respective phonetic shapes, so as to avoid such theoretical 122

complications. While this does limit, in a sense, the relevance of some of the possible

AM, e.g. the influence of the inchoative set, it serves to more properly define the

analogical process so as to be useful from an cognitive point of view.

2.5.2.2.2—THE ‘INDEXICAL ’ FUNCTION OF THE VELAR. Unlike the unique analysis of

Martín Vegas (2007), Elson (1988) concentrates on the diachronic extension of the velar

element in verbs stems, particularly to modern synchronic contexts. 67 A velar consonant is argued to appear in those forms (1s/SBJ) which are derived from the basic stem, defined as the lexical stem plus thematic vowel.

Looking beyond the complicated derivational focus, the most relevant aspect of the investigation is that the velar element is thought to take on an independent indexical function, a concept which closely corresponds to those described in Alvar & Pottier

(1983), Fanciullo (1998), Hooper (1979), Pérez Saldanya (1995b) and Maiden (2001), among others. The identity of paradigmatic function of the velar element in the set of verbs under analysis provides a means for their categorization as a lexical subclass, i.e. their having a “single paradigmatic unity” (Elson 1988:396).

In spite of sharing such innovative claims with other analyses, very little treatment is actually given to this indexical function and the limited subclass to which it pertains. In some instances, the explanation for the activation or nonactivation of the velar element within a set of similar verbs is based on very vague semantic resemblance

(Elson 1988). In fact, it is suggested that certain verbs are not susceptible to velarinsertion in spite of their actually having attested velar variants. Rather than

123

deconstructing form (and meaning) within the paradigm beyond a level commensurate

with that of the linguistic competence of speakers, the current investigation relies on the

immanent role of speakers to explain linguistic stability and change.

2.5.2.2.3—THE IMPLICATIONS OF VELAR -INSERTION IN CATALAN . The indexical function of the velar (in Catalan) is more elaborately defined in Pérez Saldanya (1995b).

It has two distinct origins which come together to form a larger distributional AM, the first being phonologically derived from the CL PRF marker –UI , while the second is originally etymological in the 1s/SBJ. Together, they are analyzed by speakers as a sort of morphological marker, functioning to set apart a verbal subclass. In the beginning, there were verbs which showed the velar in one or both of these instances, but analogy led to their merger through reanalysis in order to increase stability in the inflectional classes (Pérez Saldanya 1995b). This reanalysis gave the velar a single morphological status which increased the transparency of the inflectional morphological system (Pérez

Saldanya 1995b). or salience (esp. contrasting 1s from 3s and PRS from PRF) was thus increased in the paradigm through the independent status of the velar and the verbal inflection, which could be used separately to construct a formallydefined system of oppositions (Pérez Saldanya 1995b). In other words, the velar took on an indexical function that appears to have attained grammatical/morphological significance.

The numerous extensions of the velar and combinations of its functions into a single subclass of the verbal system are believed to regularize paradigmatic allomorphy

(Pérez Saldanya 1995b). In this way, the presence of the velar in one form implies its

124 presence in all other so marked forms (cf. the extension of a , Maiden

1992, following Tekavčić 1980a, 1980b), thereby resulting in implicatures that not only

create greater stability in the inflectional classes but also provide a more systematic

distribution of allomorphy that facilitates acquisition.

2.5.2.2.4—THE 1S/SBJ ‘MORPHOME ’. Maiden (1992, 2001, 2003) takes a fairly

traditional approach to the inception of velarinsertion in Spanish and Italian, one of early

competition between velar and palatalfinal stems, based on the combined influence of

verbs like Sp. decir , hacer , as well as the –NGERE , RGERE , LGERE verb sets (the latter being crucial for Italian). Nonetheless, the innovative examination of the establishment and extension of the 1s/(3p)/SBJ, or U and Lpatterns (for Italian and Spanish respectively), gives these investigations an elegant and suitably immanent perspective of the processes in both languages. 68

Velarinsertion is characterized by Maiden (1992) as the emergence of novel

allomorphy, that is, not the creation of a completely new paradigmatic template but rather

the creation of allomorphy in an individual lexeme, in this case, due to analogical

influence of an already existing pattern. “Generalization of a velar REPLACES a

historically regular alternant, giving rise to alternation patterns (/ng/ vs. /n/; /lg/ vs. /l/;

/gg/ vs. /d/) for which there is no historical precedent” (e.g. venire , rimanere , salire , valere , vedere , sedere ) (Maiden 1992:298). In contrast to the indexical functions described in previous works, velarinsertion is seen as having no easily identifiable grammatical function since it is not possible to delineate those forms in a paradigm which

125

take it from those that do not in a way that follows typical morphosemantic lines. In

other words, the 1s/SBJ distribution of the velar in Spanish crosses the traditional paradigmatic boundary associated with mood. For this reason, Maiden (2001), following

Aronoff (1994), adopts the term ‘morphome’, which is defined as a unit independent of phonological and morphological conditioning (after its initial development via regular

sound change or analogy) (Maiden 2001). It is an internal paradigmatic structure not able

to be expressed solely in phonological or grammatical terms and has “properties which

are autonomously morphological , inherent in paradigmatic structure but not anchored in, or derivable from, ‘extramorphological’ factors such as grammatical or lexical meaning or phonology” (Maiden 2003:4).

From an immanent point of view, the U/Lpatterns are given psychological relevance for speakers as a means of categorization and generalization (cf. Maiden 2001).

Of major relevance is the fact that there are no mixed alternations, implying that forms are not memorized as isolated units (Maiden 2001). Instead, they may be interpreted as elements organized within a larger organizational structure, the morphome, binding its member forms together through the strong associative links of particular cognitive and linguistic features.

The morphomic distribution of the velar is motivated by the following points: (1)

there is no shared morphosyntactic or semantic function that distinguishes the velar cells

in the paradigm (1s/(3p)/SBJ) from the nonvelar ones (remainder of the IND); (2) it is

not phonologically conditioned because the nonfront vowel does not trigger

velarinsertion in all cases; (3) the appearance of the velar in the paradigm shows

126

diachronic coherence, i.e. if it is found in one of the marked cells, it is found in all of

them, i.e. an “inviolable distributional regularity”; (4) the velar is not inserted by proportional analogy since this would imply the creation of innovative allomorphy and

would contradict the observed diachronic coherence described above (Maiden 2001,

2003). An important consequence of point (3) is that this morphosyntactic categorization

is seen synchronically as an “autonomous morphological entity, inherent to the paradigmatic structure of the verb” (Maiden 2001:45). In other words, the paradigmatic

structure has psychological relevance for the speaker as a unit of linguistic organization,

i.e. the ‘morphome’, which is independent of the morphological content of the individual

cells. With regard to (4), the process is not creating new allomorphic structure, but rather

undergoing a sort of leveling to increase transparency between the phonological form and

the paradigmatic structure, that is, a convergence of effects with a concrete, formal

realization represented by the already established distribution of the velar element

(Maiden 2001; cf. Federici & Pirrelli 1997).

With the foundation of a lexical subclass through the identification of the 1s/SBJ morphome, the velar is extended to these forms through a sort of analogical leveling, or hypercharacterization of paradigminternal (‘morphomic’) structure, which increases transparency between the form and the morphome (Maiden 2001, cf. Lenfest 1978, 1993,

Malkiel 1967, 1974). This process ruptures the established transparency between form and meaning in favor of another, intramorphological type (Maiden 2001).

The paradigmatic template/morphome that comes to be associated with velarinsertion is perhaps the most salient contributing factor of the AM that permits the

127

extension of the velar diachronically. For this reason, it is a concept that is emphasized

throughout the current investigation.

2.6—DIACHRONY , ANALOGY , PARADIGMATIC FUNCTION AND IMMANENCE . While

numerous insights into the process of velarinsertion are found in the preceding analyses,

their major shortcoming is the lack of specificity regarding both the analogical factors

which affect this verbal subclass and its chronological and diachronic development beyond the initial (prehistorical) stage. The general tendency is to explain its inception

as ‘analogical’ and then (often partially) list the forms affected while completely

disregarding the importance of the various stages of synchronic variation and the Modern

Spanish standard and dialectal outcomes. The omission of major periods of variation in

some analyses, while providing a workable amount of information for a general

investigation, has created a very inflexible, too broadly defined analogical process. It

should be noted, however, that such neatly organized analogical effects are not

representative of the actual diachronic situtation when viewed with greater synchronic

detail.

The analogical processes involved in the creation of the velarinsert verbal

subclass can be defined much more formally and specifically within AEM. Of the major points that are missing or downplayed in earlier investigations, the changes in categorical

membership and a concise analysis of the analogical/morphomic function of the paradigm are of particular interest. Velarinsertion is an analogical change and it is the paradigmatic structure (or template), in conjunction with the morphological, lexical and

128 phonological properties of these verbs, that play the decisive role in its development. To

a large extent, the diachronic expansion and apparent closure of this lexical subcategory

are crucially based on the interaction of multiple morphological, phonological and,

especially analogical factors (i.e. “multiple causation”, following Malkiel 1967), which

are crucial to AEM. The initial stage of the Spanish process involves an AM based on

the set of –NGERE verbs that is supported, in a secondary fashion, by the paradigms of related verbs such as decir ‘say’ and hacer ‘do, make’. Afterwards, the AM used to extend velarinsertion included the set of verbs which had previously served as targets.

In both the subsequent expansion and retraction of velarinsertion (via interparadigmatic analogical extension), the AM, composed of forms sharing multiple variables within a

‘homogeneous supracontext’, is constantly reanalyzed and pitted against competing models, each of which has the potential to influence the direction of change of a particular verb. These shifts in analogical influence necessitate a more elaborate examination of the chronology of the process of velarinsertion from beginning to end, highlighting the synchronic stages which show changes in class membership.

The point of departure for the current investigation is taken from Maiden (1992), which indicates that the ‘immanent’ strategies utilized by speakers are founded in analogy:

So far as the Romance data are concerned, I suggest that speakers, faced with conflicting typological evidence in lexical roots, resort to a variety of strategies whose fundamental aim is to anchor as firmly as possible in the grammar the various alternation patterns with which they are confronted. One strategy is to associate allomorphy with another major (and semantically arbitrary) morphological class, namely conjugation…Another strategy…consists in speakers’ seeking to identify

129

common distributional patterns underlying the range of phonologically disparate allomorphy which they encounter. (309)

These patterns are then separated, when possible, by semantic and/or morphophonological prominence (in this case, by phonological salience of already existing “abstract alternation patterns within paradigms” (Maiden 1992:310)). In other words, speakers make use of the most closely associated AM possible in order to resolve such linguistic problems as paradigmatic allomorphy. The following chapter attempts to answer the questions concerning the variables of the AM relevant to the emergence of the

Spanish velarinsert class and the diffusion of the velar element beyond the initial stages.

The process of analogy as a psychological or ‘immanent’ means of language organization and change is also emphasized.

130

CHAPTER 3: A USAGE -BASED APPROACH TO VELAR -INSERTION

3.1—VELAR -INSERTION AS AN ANALOGICAL PROCESS . As indicated in the previous chapter, Spanish velarinsertion is most elegantly and effectively explained as an analogical process. The AM is defined through multiple causation, i.e. the (accidental) confluence of numerous phonological, morphological, lexical and paradigmatic characteristics in a shared group of exemplars resulting in their potential to serve as a model for language categorization and change. The constructs of AEM provide the most concise and ‘immanent’ portrait of this particular AM. In doing so, this theory also accounts for the diachronic shape of the AM through synchronic mutation as well as the synchronic and diachronic variation that characterizes the velarinsert subclass.

3.1.1—IMMANENCE AND ‘MULTIPLE CAUSATION ’ IN AEM. The analogical function of exemplars as an immanent means of cognitive linguistic organization is what separates the current analysis from other theoretical solutions. The fact that Spanish velarinsertion is traditionally viewed as an analogical process means that it is neither easily nor directly included in the typical rule or constraintbased analysis. Nonetheless, there are many theoretical solutions, each of which handles the data in a different way, although in most cases, such explanations account for the variation or subregularity found in the output through idiosyncratic and/or extratheoretical methods. Herein lies the problem with the 131

typical analogical approach. The ideal solution is one which not only accounts for said

variation and subregularity, but which also incorporates it into the theoretical model and proves to be a realistic means for the actual (imperfect) speaker to handle the

complexities of linguistic information.

The nonteleological nature of AEM is exemplified in Figure 3.1 by a comparison of the different outcomes taken by the Romance languages with respect to the forms belonging to the Spanish velarinsert subclass.

Lg INF IND SBJ #

MSp venir vengo venga 1 vienes vengas 2

MCat venir/vindre vinc vingui 1 véns vinguis 2

MIt venire vengo venga 1 vieni venga 2

MEPtg vir venho venha 1 vens venhas 2

MFr venir viens vienne 1 viens viennes 2

MRom a veni vin vin 1 vii vii 2

MProv venì vèni vengui 1 vènes vengues 2

MOc venir veni venga/vengui 1 venes vengas 2 Figure 3.1—Example of various paths for change in Romance.

132

Velarinsertion is only one of the many potential directions for change since the AM that

affects a particular form or set of forms may differ and/or have different consequences

across speakers or groups of speakers. In all these particular cases, however, one can see

emergent patterns and changes which are exclusive to the verbal paradigm.

The number of factors that are essential in the emergence, extension and variation

attributed to Spanish velarinsertion necessitate a more elaborate examination of multiple

causation within the AM than has been previously proposed. This conception of the AM

is realized as exemplars linked through the variables they share that are relevant to the

model. Support for multiple causation in acquisition is found in Nosofsky & Zaki

(2002):

At early stages of category learning involving highly separabledimension stimuli, most observers attend selectively to single dimensions. As learning proceeds, if no single dimension provides adequate information to allow satisfactory performance, then observers spread attention to multiple dimensions of the objects. Thus, at early stages of learning, the exemplars stored in memory may be composed of only singledimension or other lowdimension combinations of information, whereas in later stages of learning, the exemplars are composed of multipledimension combinations of information. In both cases, observers classify objects on the basis of their similarity to these stored exemplars. (938)

The scenario described above directly corresponds to the diachronic changes in the AM

for velarinsertion analyzed below.

3.1.2—SYNCHRONY IN DIACHRONY. In the last decade, there has been increased interest in the elaboration of synchronic stages between diachronic endpoints A and B (Janda &

Joseph 2005, Joseph 2001, 2006, Wanner 2006b). By examining these ‘synchronic slices’, it is possible to more reliably identify the dynamic nature and multiple causation 133

involved in language change. Though many analyses have explored the origin and

development of velarinsertion, none of these has laid out the precise details of the

“synchronic slices” or the resulting diachronic situation in a way that did not exclude or

ignore some of the crucial factors involved, such as the function of the verbal paradigm

and the existence of allomorphic variation.

Greater detail in the diachronic analysis means that each synchronic stage paints a revised picture of the AM. In other words, synchrony in diachrony makes it possible to see the everchanging nature, productivity and systematicity of the AM over time (cf.

§2.4). In this way, the synchronic steps involved in the diachronic change and their importance in the larger historical development become clear. Typically, the lack of synchronic detail in diachronic analyses is due to the initial establishment of an AM.

That is to say, once an AM has been defined and is thought to be influential over the remaining diachronic development of a particular phenomenon (assuming that all affected targets become part of the expanded AM), it may appear unnecessary to examine subsequent stages at all. Given that each synchronic stage actually has its own welldefined AM based on new input, however, it continues to be crucial to dissect the characteristics of the model to identify not only how it has changed over time, but also to provide greater insight into the underlying analogical process. Properly defining the AM helps to explain its potential productivity and diachronic direction. The following sections investigate the stages of velarinsertion that show relevant changes in the AM and consequently its affected targets. The potential targets for velarinsertion that are not influenced by the AM are examined in §4.5.

134

Another aspect of the analogical process that is not often discussed is the

influence of competing AM (cf. Zuraw 2003). In each synchronic stage, there is pressure

exerted on an analogical target from various directions, including from within the particular group (not to mention paradigm) to which it already belongs. It should not be

overlooked that competing linguistic changes are inherently handled in AEM. This

resembles the concept treated in Wang (1969) concerning the nonsystematic diffusion of

change, analogical or otherwise, as something that has long been a problem for traditional

linguistic theory. Competing analogical forces permit generations of speakers (or even

an individual speaker) to constantly restructure their grammar such that a form which is

affected by one particular AM may subsequently be reanalyzed due to the external pressure of another AM. In such instances, language change can be seen as gradual and

lexically diffuse. Additionally, it is not solely the result of (near)identity between a particular AM and its target, but depends upon other factors like frequency and the role of

the speaker and their environment as defined by an immanent view of language (cf.

Joseph 1992, Wanner 2006a, 2006b).

Synchronic deconstruction of diachronic change highlights this immanent perspective and provides the analogical basis of linguistic organization, including the

emergence of structures like the inflectional paradigm (cf. §§2.6 and 4.2). The theory of

immanence contends that a speaker’s memory is inherently imperfect, that input is

continuously susceptible to change, degradation and/or loss due to any number of factors,

including inhibited reception, new input, (un)conscious reforming, and time. Changes in

the categorization of information involve reanalysis according to a different AM. At the

135

same time, accessibility (or activation) of an AM, as defined by associative strength or

salience and frequency, is representative of the time needed to extract from memory a potential AM. Memory recall is supposedly measured by time and so the most accessible

AM is also the most quickly recalled. The speed with which it is accessed is a

consequence of its degree of similarity to the target and the type and token frequencies of

its members (cf. §4.3).

3.1.3—OUTLINE OF CHAPTER . The remainder of this chapter analyzes the diachronic

analogical process of velarinsertion in Spanish. First, a definition is provided of the

characteristics that are essential to the velarinsert AM. Afterwards, the historical background relevant to the diachronic situation is reviewed with particular emphasis on

the emergence, spread and motivation of the analogical process. Next, a brief overview

of the orthographic problems associated with the historical data identifies complications

that have plagued many previous investigations. The diachronic process is then broken

down into a series of synchronic stages in order to provide a greater level of detail

concerning the analogical targets and the changes affecting the AM. The last data set

examined includes dialectal examples of velarinsertion demonstrative of the continuous potential for change that is a consequence of AEM. Finally, the theoretical consequences

of this analysis lay the foundation for the discussion of numerous theoretical and practical

issues in chapter 4.

136

3.2—DEFINING ‘MULTIPLE CAUSATION ’ IN THE AM. As indicated in §2.4, the AM for velarinsertion is complex and cannot be limited to any single linguistic domain. What follows is a more precise examination of the particular features attributed to the AM.

Special emphasis is placed on the phonological, morphological and paradigmatic factors that contribute to the localized regularity and productivity of this model. 69

3.2.1—THE PHONOLOGICAL VARIABLES OF THE AM. The focus on the phonological

factors involved in velarinsertion is one of the main reasons that previous research has

fallen short in properly defining the AM. Velarinsertion occurs precisely on account of

the convergence of multiple causes from various aspects of linguistic structure and

organization, i.e. phonology, morphology and paradigmatic structure, the latter comprised

of lexical, grammatical, phonological, morphological and psychological elements, none

of which is sufficient by itself to motivate analogical change in this case. The function

and internal characteristics of paradigmatic structure are an innovative and crucial

addition to the AM for velarinsertion. Likewise, immanence, associative strength and

the dynamic nature of the cognitive organization of language are considered essential to

the AEM analysis.

In spite of these multiple causes, there are several phonological factors that prove

essential to the synchronic and diachronic views of velarinsertion. The most obvious of

these involve the sounds that are immediately adjacent to the velar’s position, i.e. the preceding and following segments of the phonological environment (cf. ‘temporal proximity’ in Burzio 2005). Amongst all of the necessary characteristics of the relevant

137

AM, these are perhaps the most salient. However, they are not the only relevant phonological variables. The shape of the verb stem is also crucial, especially in the

earliest stages of the process, as seen below.

3.2.1.1—PRECEDING CONDITIONING ENVIRONMENT . The initial stages of velarinsertion demonstrate the generally accepted pattern of development on the basis of the stemfinal, or prevelar segment. The velar extends from nasal to sonorant (lateral and rhotic) then seemingly simultaneously to sibilant and glidefinal verb stems (the latter representing respectively a lower and higher degree of sonority) due to a simplification or broadening of the phonological conditioning environment (cf.

Maiden 2001). After these criteria are established, verbs containing these same stemfinal phonemes that are also marked by the other numerous variables of the synchronic AM, are susceptible to velarinsertion and, consequently, the associated paradigmatic structure/morphome.

The first synchronic slice is preliterary and is characterized by the insertion of the velar after the stemfinal coronal nasal /n/, as found in tener , poner , remanir , and venir . Later, in the earliest recorded historical stage, the verbs affected have a stemfinal coronal sonorant, i.e. /n, l, ɾ/, e.g. salir , herir . Subsequently, (nonsyllabic) yod and sibilantfinal stems are added to the mix, as the velarinsert class adds members like caer , huir , oír , traer , and coser , as well as soler , valer , and toller. It can be seen through this brief description of the first three stages of development that all of the affected stemfinal

138

segments are included. However, there are additional synchronic stages of

velarinsertion that merit detailed examination, which are therefore treated below.

The immediately adjacent stemfinal segment is not the only part of the stem

involved in defining the AM. Early on, velarinsertion affects coronalfinal stems preceded by a (typically nonhigh) vowel, i.e. [+voc, –high] [voc, +cor] __ [+voc, +bk].

The stem vowel is relevant because velarinsertion does not, in most cases, affect any of

the aforementioned alveolar sonorants when immediately preceded by a high vowel. 70

This eliminates numerous verbs from both the model and target sets as explained by the

notion of the ‘homogeneous supracontext’, i.e. the contextual space in which all

exemplars act in a similar fashion with respect to the relevant variables (Skousen 1989).

Additionally, this context excludes other verbs and even forms within the same paradigm,

a consequence of which is the salience of the 1s/SBJ template.

The phonological features that characterize the stemfinal segments in the affected

verbs demonstrate the broadening of the AM over time. The fact that the preliterary

stage shows velarinsertion exclusively after the coronal nasal /n/ means that the

conditioning environment is defined by all of the distinctive features of that phoneme,

where [voc, +son, +cor, +ant, dist, +nas] are relevant for the diachronic process. It is

important to note that the phonemes affected at each stage comprise a natural class in

Spanish, a fact that should not be overlooked in the synchronic or diachronic

development of the velarinsert group. This is especially important for the composition

of the AM because including the specific phonemes as variables, as opposed to the

specifically relevant phonological features proposed here, would exponentially increase

139

the necessary processing time, create a greater number of disagreements between possible

outcomes, and possibly result in false predictions of similarity (cf. Skousen 1989; cntr.

Eddington 2002). The motivation for including such highly detailed variables in the AM

is examined in §1.3.

In the second synchronic stage, the above set of features becomes less specific so as to include stems ending in a coronal liquid /l, ɾ/, requiring loss of specification for the feature [nas]. For inclusion of the sibilant and yodfinal verbs in later stages then, it is again essential to eliminate some of the phonological features that characterized the preceding AM. Specifically, the specification of the features [son] and [ant] is removed.

Diachronically, the changes affecting this natural class crucially rely on the relative positions of the phonemes along the Sonority Hierarchy and the fact that the included segments are the only ones (naturally) permitted syllablefinally in Spanish (see below). 71

The analogical extension of the velar essentially involves the broadening of its phonological context to the extent allowed by the phonotactic constraints of the language.

This resembles a rulelike progression, i.e. the way in which a generative system handles

rule formation, in that it is initially more highly defined, dealing with specific cases, then

giving way to the more broadly and widely applicable rules. The difference lies in the

fact that the rule system typically anticipates completely systematic application

synchronically while the analogical process permits imperfect, gradual change without

complete closure.

140

3.2.1.2—POST -VELAR CONDITIONING ENVIRONMENT . The phonological conditioning environment following the velar position is also of utmost importance. Since these verbs come from the second and third conjugations, many of their inflectional forms are marked in some way by a front vowel, e.g. 1s IMPF salía , 1s PRET salí , 1s IMPF SBJ saliera . This includes the PRS IND forms which are not marked by a velar element, e.g.

2s sales , 3p salen . Thus, the distribution of /g/ is phonologically conditioned by the following nonfront vowel /a, o/ in the paradigm. The appearance of the velar before a nonfront vowel in the 1s/SBJ forms of the verbal paradigm highlights the interplay between phonological and morphological change. It is based on the inflectional distribution of vowels in the verbal system and prevents palatalization effects like those associated with 2 nd and 3 rd conjugation verbs having an etymological velar (cf. §2.3).

This situation has implications for the development of the verbal subclass. First, the

hypothesis that the velar augment appeared in both 1s and SBJ at the same time is much

more plausible. It is not necessary to appeal to a basicderived relationship between these

forms, nor to attempt to argue in favor of the appearance of the velar first in the IND,

then in the SBJ or vice versa (cf. §4.4). Second, once the 1s/SBJ distribution is

established (and supported by other verbs with the same allomorphic distribution of verb

stems, e.g. the inchoative set, decir , hacer , etc.), it becomes available to speakers as an

identifying characteristic of lexical organization. This means that when a speaker learns

a new form having a phonetic shape that could place it into such a group of verbs, it is

automatically a potential target for class membership. Whether it falls into a group or not

depends upon factors such as frequency and which of the conflicting analogical forces,

141

i.e. competing AM, wins out. Finally, this also explains why the 1 st conjugation should be excluded from velarinsertion since the distribution of the nonfront vowels does not match the established paradigmatic template. Were this conjugation class marked by the velar per the same phonological conditioning environment, it would appear in all cells of the paradigm except the PRS SBJ and 1s PRET, thereby disrupting the established pattern of variation and creating a completely innovative and unnatural distribution of paradigmatic allomorphy. This last point is notable because it demonstrates the symbiotic relationship between the phonological and morphological aspects of the analogical situation.

3.2.1.3—ADDITIONAL PHONOLOGICAL FACTORS . Hooper (1979) some of the phonological factors in velarinsertion:

The answer has to do with syllable structure…Vowels, glides, nasals, liquids and /s/ happen to be just those segments that can end in Spanish. Note that with the inserted velar, the stemfinal segment becomes syllablefinal. If the velar were inserted after an obstruent, or a cluster, an unacceptable syllable would result…In this case, then, it seems to be the general syllable structure conditions of the language that determine the morphological class membership. (124)

The fact that these phonemes are found in syllablefinal position when the velar is inserted relates to another important phonological component of the AM, the basic shape of the verb stem. In general, the verbs affected by velarinsertion have a basic

CVC stem shape. 72 With the possible exclusion of the verb remanir , all of the verbs

showing velarinsertion in the first two stages mentioned above have a CVC structure.73

This stem shape accounts for approximately 3/4 of all the velarinsert verbs as identified

142

in this investigation, excluding derived forms. The numbers are even higher if it is

restricted to velars attested prior to the 16 th c. (a stage marked by an increased frequency

of velarinsertion in learned or semilearned forms, see below), in which case nearly 80%

of the verbs share this phonological pattern.

Ultimately, the AM for velarinsertion is defined by numerous phonological

factors including the immediately adjacent segments in the conditioning enviroment as

well as certain features of the stem. Their influence, however, is necessarily intertwined

with certain morphological aspects and it is only through the combined effect of all the

necessary characteristics of the model that they have the associative strength required to

trigger language change.

3.2.2—MORPHOLOGICAL CHARACTERISTICS OF THE AM. The initial distribution of the velar augment is a consequence of phonological developments affecting inflectional (i.e. morphological) elements. It is quickly extended by analogical means to verbs (i.e. lexical items having identical grammatical/syntactic function) of similar phonological shape and morphological categorization (i.e. conjugation class). In spite of this consistent overlap between the phonology and morphology involved in the construction of an AM, the morphological criteria can be examined independently of the phonology. While in some instances these criteria are quite obvious, their relevance should not be understated.

Perhaps the most salient of these observations is that velarinsertion is restricted to the grammatical category of verbs (cf. §2.4, Stemberger 1994). While this indication may appear rather inconsequential on the surface, it has important implications for the

143

overall AM. First and foremost, forms carry inflectional markers. These

markers and their phonetic realization are implicated in the morphological

subcategorization of verbs into conjugational classes, the paradigmatic distribution of the

velarinsert, and the phonological changes affecting particular forms within the paradigm.

Thus, just as the phonology and morphology of the AM overlap, so does the morphology

overlap with syntactic function and paradigmatic distribution.

3.2.3—THE MORPHOMIC FUNCTION OF THE PARADIGM . Perhaps the most crucial point

of contrast between this and previous investigations involving velarinsertion in Spanish

is the function of paradigmatic structure in AEM. While more recent works have

initiated the discussion of its relevance (e.g. Maiden 2001, 2003), they have only touched

the surface of the matter, in some cases overlooking its usefulness, as well as that of

many other of the multiple causes that define the AM. What can be taken from previous

research, however, is the conception of a unit of linguistic organization and

categorization that is independent of the phonological and morphological conditioning

that give rise to it, i.e. the ‘morphome’ (cf. §2.5.2.2.4). Though it functions apart from

the phonological and morphological aspects of the AM, it is important to note that the

morphome is essentially a consequence of the ‘morphologization’ of phonological

information within the restricted domain of the inflectional paradigm (cf. Joseph & Janda

1988). As seen previously, the 1s/SBJ paradigmatic template, or morphome, is the result

of the systematic categorization of the phonetic realization of morphosyntactic properties.

144

Not only did the velar extend the 1s/SBJ pattern to previously invariant verb

stems, but other alternations also came to be associated with this paradigmatic template

(cf. Maiden 2001). In Spanish, for example, it is possible to include both the ‘inchoative’

set as well as verbs with other types of allomorphy, e.g. caber (cf. §4.2). This paradigmatic template is so well established that it has become a marker of the Romance

languages in general, as demonstrated in Appendix F. It is largely the result of phonological changes that created allomorphy which is maintained, to greater or lesser

degrees, across the Romancespeaking territories (Maiden 2001). The establishment of

the 1s/SBJ morphome led to its psychological activation for speakers as a unit of

linguistic organization (cf. Pirrelli 2000). It continues to have psychological relevance to

the speaker thanks to its diachronic convergence and coherence. The immanence of this paradigmatic pattern is supported by the current view of analogy as described in Maiden

(2003):

In effect, it seems that speakers actively seek out, then reinforce and generalize, recurrent patterns of similarity and difference between ‘cells’ of morphological paradigms across lexemes, perhaps as a way of constraining and making as predictable as possible the patterns of deviation between lexical meaning and form in the wordforms of paradigms. That speakers recognize and maximize such patterns is manifest in the fact that they display ‘coherence’ (prominent diachronic resistance to morphological and even phonological changes which might be expected to disrupt the paradigmatic patterns) and may be subject to ‘convergence’ (in effect a form of intramorphological analogical levelling such that the relevant cluster of paradigmatic cells comes to be endowed with certain common phonological characteristics across all lexemes in which it occurs). (Maiden 2003:5)

The relevance of the 1s/SBJ template for allomorphic variation is essential to the

current investigation since it provides the last essential piece of the analogical puzzle that

145

explains velarinsertion in Spanish. As a realistic means of linguistic organzation for the

speaker, it also contributes to the immanent view of language change taken herein.

3.2.4—IMMANENCE AND LINGUISTIC CATEGORIZATION . What the preceding quotation

so succinctly addresses is the psychological reality of linguistic organization. Similarity

and difference between forms or groups of forms results in the emergence of patterns

which may, in turn, become the locally generalized characteristic that defines a partiuclar

lexical subclass. At the same time, the associative strength (including frequency)

garnered from such categorization may protect members of such a group from other types

of change. While the ‘coherence’ and ‘convergence’ of forms within a paradigm tends to be taken for granted, it is important to note the special, psychologically real relationship between them, as well as the relationship between entire paradigms or individual cells (or

groups of cells) across paradigms.

By adding the immanent effect of the speaker in their linguistic environment,

warts and all, into the equation, it is possible to see that language, even at the lowest

level, is in a constant state of flux, continuously being reanalyzed and resorted.

“Speakers can and do come up with analyses that make sense and to a certain extent are

almost forced on them by a certain configuration of facts, and such reanalyses often tell

the linguist just what is, or will be, going on in the language” (Joseph 1992:131). This is

the dynamic aspect of language that is incorporated into AEM and is so crucial to

diachronic and synchronic linguistic analysis.

146

3.3—LOOKING BACK AT VELAR -INSERTION . In order to better understand the diachronic process of velarinsertion by way of the most relevant synchronic stages, it is first fundamental to return briefly to the question of its origin. This lays the groundwork for closer examination of the AM and its extension (and retraction) over time. Although these matters have been discussed thoroughly in the preceding chapter, it is important to shed additional light on three questions essential to every analysis of the process: (1)

Where does the velarinsert come from?; (2) How does it spread?; (3) Why does it spread?

The first question deals with the proposed analogical source, or AM, for velarinsertion in the primary, preliterary stage, affecting tener , venir , poner , and remanir (see below). Although some previous analyses have suggested that the central

AM for these forms is the decir , hacer set, such a hypothesis is not permitted in AEM by way of the principle of the ‘homogeneous supracontext’ which states that the target and model share the highest possible degree of similarity. Because there is another set of more similar verbs, those derived from CL stems ending in –NGERE , the analogical function of the decir , hacer set can only be secondary. The NGERE forms provide the

most similarly available AM for velarinsertion to occur in verbs like tener and venir .

The similarities involved are those described previously: (1) the CVC stem shape; (2) a

nonhigh stem vowel /a, e, o/; (3) a coronal nasal /n/ preceding the velar; (4) a following

nonthematic, nonfront inflectional marker, /a, o/; (5) the lexical categorization of these

forms as verbs of the second and third conjugations; and (6) the 1s/SBJ paradigmatic

shape emergent from the allomorphic variation in the verb stem. Such factors are

147

influential not only through their involvement in the development and expansion of the

velarinsert class, but also by the linguistic material which they exclude, e.g. nonverbal

forms, VĪNEA > viña “vineyard”, ARĀNEA > araña “spider” (cf. §2.3.2.1). They also

validate the need to explore motivation for analogical change beyond the phonological

context to include the incorporation of morphological and paradigmatic information.

While the decir , hacer group shares most of the characteristics of the AM

summarized here (cf. §3.2), it lacks one crucial element, that of the preceding phonological conditioning environment. In this instance, the AM relies on the presence

of a coronal nasal immediately before the point of insertion, a requirement which is

fulfilled by the CL –NGERE class. In spite of the fact that they are not the main AM for

firststage velarinsertion, the decir , hacer set does provide support for the emergence of

the velar as indicative of the 1s/SBJ morphome. AEM relies upon comparison of any and

all lexical forms sharing identity in any respect that may be salient (consciously or

otherwise) to the speaker. In other words, exemplars are linked to each other via

assocations of relative strength made on the basis of any shared characteristic that is in

some way relevant in the mind of the linguistic individual. When exemplars are strongly

linked in this way, they form ‘exemplar clouds’, which may be abstractly envisioned as

distant views of a galaxy or the rather popular conceptualizations of the connections between neurons in the brain (such as those often seen in television commercials, cf.

§1.3.2). In the central part of the exemplar cloud that connects the velarinsert targets

tener , venir , poner and remanir to the verbs derived from CL –NGERE , the decir , hacer

set is peripheral, associated, fairly strongly, to both model and target, by virtue of its

148

identity in all of the aforementioned characteristics with the exception of the stemfinal

consonant. Their paradigmatic shape and high token frequency provides stability for the

allomorphic distribution of the velar. Additionally, after the –NGERE set is reanalyzed

and leveled in favor of a palatalfinal stem (and so is no longer strongly associated with

the velarinserts), the decir , hacer set asserts its stabilizing function for the survival of the velarinsert class and the 1s/SBJ morphome. This is supported by the extension of the velarinsert to verbs characterized by other stemfinal segments.

A related issue which has been scrutinized in previous investigations is whether or not these initial analogical targets ever contained a palatal nasal, even if only in synchronic variation with a nonpalatal. On one hand, it is possible that the analogical influence of the –NGERE set was enough to create a context for velarinsertion in these

targets such that the velar stem may have alternated for a time with the palatal one, i.e.

*teñ ~ teng . This hypothesis is supported by the palatal stems found in Portuguese and

the (somewhat limited and dialectal) palatal/velar alternations found in Old Catalan, Old

Provençal and Italian as seen in Figure 3.2 below. Unlike the set of –NGERE verbs, however, the velarinsert targets were not pressured into leveling their paradigms in favor of the palatal stem. This is explained by the stability of the newly established velarinsert set in conjunction with the analogical force of the paradigmatic template they shared with the decir , hacer set. On the other hand, the fact that these firststage verbs have no attested cases of palatalfinal stems in the recorded history of the language suggests that they have never had a palatal variant. The lack of palatalization effects in these and other verbs, e.g. sentir (cf. §2.5.2.1.6), is indicative of the early loss of yod. The panRomance

149

geographical distribution of the velar and palatal stems in these verbs also lends credence

to the nonpalatal hypothesis. In the end, the presence of a palatalfinal stem variant in

these verbs is actually inconsequential since they would be susceptible to the influence of

the same AM centered around the –NGERE set in either case.

Lg 1s

MEPtg tenho venho

OCat/OProv tenc /tenh, venc/venh venc/venh

OIt tengo /tegno (dial.), vengo/vegno (dial.) vengo/vegno (dial.) Figure 3.2—Evidence in favor of palatalization prior to velarinsertion.

The preceding explanation of the first stages of the velarinsertion process should be considered in conjunction with the need to dissect the diachronic process into its

relevant synchronic stages. This helps to answer the second question indicated above,

concerning how the velar spread. All of the most wellaccepted investigations of

velarinsertion agree that it is an analogical process and therefore have proposed an AM

of one form or another (cf. §2.4.2). However, it is not sufficient to declare velarinsertion

an analogical development and then provide a singular, static model for the entirety of its

diachronic extension. It is true that from the outset, velarinsertion is analogical and it

continues to be so until its associative strength is weakened through an everbroadening

domain of influence. The point, though, is that the forms which comprise the AM are

150

constantly changing, as forms take on the velar, fall out of use, or are reanalyzed

according to competing AM or via phonological change. For these reasons, the following

analysis addresses the process by focusing on each synchronic stage where the AM

differs in some way.

Finally, the matter of why the velarinsert spread is less clear. Its extension involves changes in the composition and frequency of the relevant AM and its members as well as the shape and frequency of potential analogical targets. At the same time, the interaction of competing models contributes to the overall lack of systematicity that characterizes the process at all stages. These points are intricately related to immanence, especially the speaker’s imperfect memory and the resolution of new or reanalyzed input according to preexisting linguistic information and structures.

This contextual background is the starting point for the closer examination of the synchronic stages contributing to its diachronic development. The AM, composed of the

–NGERE set (and, secondarily, verbs like decir , hacer , and yacer ), triggers velarinsertion

in the first forms entering the class. Through the everchanging composition of the AM,

it continues to exert analogical influence in the minds of speakers, producing various

types of synchronic and diachronic change.

3.4—ORTHOGRAPHIC PROBLEMS. In spite of the assertions of Oelschläger (1940) that the orthographic representation of the Spanish language has shown “startling consistency…from its very origins up to the present” (x), the earliest scribal practices encountered numerous problems in dealing with yod effects and palatalization. These

151

matters were often resolved in a variety of ways, some of which created conflict between

the traditional and innovative use of certain , like < g>, which were already

representative of other sounds. Because of this historic variation in the use of this particular letter, so crucial to the issue at hand, some of the earliest attested forms of apparent velarinsertion should not be identified as such. This is due to the fact that they do not appear to represent the velar occlusive /g/, but rather other, often innovative fricative and/or palatal phonemes. Motivation for their exclusion involves the representation of certain nonvelar phonemes as well as the fact that the forms under discussion show a vast chronological separation from the next attested instance of velarinsertion in the same verb if any are attested.

Related to the multiple functions of the is the aforementioned consistency of orthographic practices in the Spanish language. Given that the written language is characterized by a surprisingly systematic representation of sounds, it should be noted that the use of < g> was in a state of flux during the periods in which these socalled cases of velarinsertion are found, primarily the 10 th to 12 th centuries. The variation found is due to the effects of yod and palatalization which resulted in the creation of a whole new set of sounds in the Romance phonological system. For example, the grapheme < g> could, like < j, i>, represent a /ʒ/, even before nonfront vowels, e.g. ECLESIA > eclegia , egrija “church”, this phoneme, early on, also being represented by < gi , ig , gg > (Cano Aguilar 1999). Adding to the problem is a

similarly if not identically realized phoneme, typically represented by the grapheme

(e.g. RADIU > rayo “ray, lightning”, FAGEA > haya “beechwood (tree)”), which, due

152

to its similarity with /ʒ/, could be represented by the same graphemes, i.e. < i, j, g, ig , gi >

(Cano Aguilar 1999). This accounts for several apparent cases of velarinsertion in the

Glosas Silenses and Emilianenses, documents which are widely believed to serve as the starting point of the recorded history of the Spanish language (cf. Lapesa 1980). Thus, the grapheme < g> in the 3s SBJ siegat and 1p SBJ segamus (from ser “be”), attested in the Glosas Silenses and Emilianeses respectively, represents a prepalatal fricative /ʒ/ (cf.

García de Diego 1951, Hanssen 1913, Lapesa 1980, Malkiel 1974). Based on this information, it is possible to contend that these earliest attestations of this grapheme in verb stems with a final nonconsonantal segment are not representative of the occlusive under investigation here, rather they are merely variants of the prepalatal fricative. Not only does this apply to the attested forms above, but also to others including 1p SBJ kaigamus “fall”, as attested in the (cntr. Menéndez Pidal 1980). 74

Support for this argument comes from the fact that these forms are attested in the earliest recorded history of the language after which they are not attested again for a span of centuries, if at all. After the establishment of more regulated and systematic orthographic practices in the language, ca. 13 th c., no such isolated anomolous forms are attested (cf. Cano Aguilar 1999).

The palatalization affecting nasals and laterals, meanwhile, was also realized, on occasion, with < g>. In the 11 th c., there are examples of palatal graphemic variations, in the Glosas and other documents from the same monasteries, such as valge ~ valle

“valley”, 3s SBJ punga ~ puña “poke”, though by the 12 th c., they are primarily

represented by < ll , nn > (Lapesa 1980; cf. Cano Aguilar 1999). Add to this the fact that

153

the palatal nasal can be derived from CL NG (where G could become yod by the effect

of a following front vowel), NG ’L, and GN , among other sources, e.g. NGERE > ñir

(cf. punga , above), UNGULA > uña “fingernail”, SIGNA > seña “sign, signal” (Cano

Aguilar 1999). The relevance of the set of –NGERE verbs becomes all the more apparent in light of this information. This is because there was a time when both the nasalplusvelar /ng/ and the palatal nasal /ɲ/ could have been represented, though less than systematically, by the same graphemic units.

Arguments similar to those made for the representation of the nasal can also be made for the lateral. For valer and salir in particular, there are highly limited attested cases with a double grapheme, e.g. 1s/3s SBJ valla , 3p vallan but the confusion between simple and geminate in early writings calls the pronunciation of such forms into question, as demonstrated by others like 3s IND salle , 3p IND sallen , INF valler , sallir , where palatalization would not be expected phonologically (cf. Menéndez

Pidal 1980). There is no getting around the absence of the velar in numerous early examples of the 1s/SBJ forms of these verbs, e.g. 1s/3s SBJ sala , vala , but this only lends itself to the hypothesis that the forms indicated above may not have been realized with a

(pre)palatal pronunciation. The appearance of geminate grapheme < ll > in forms of valer and salir is likely a hypercorrection, given the preceding information and the absence of any other alternate forms having a prepalatal fricative /ʒ/ which would be expected as the regular, phonetic development of these forms, cf. MPtg. 1s IND saio (cf. §2.3.2.1).

According to the details discussed above, the obvious confusion characteristic of early documents, especially in the numerous uses of the grapheme < g>, is not surprising

154

and tends to complicate the identification of the earliest attested examples of the

velarinsert. This information, in conjunction with etymological sources of the forms

under analysis, discounts several cases of unexpected velarinsertion early on in the

recorded history of the language, cases which have been considered problematic in previous analyses, including 1p SBJ kaigamus and 1p SBJ segamus discussed above.

Such inconsistencies in the primitive stages of the Spanish language suggest that early

and far removed examples like those described in the preceding discussion are not true

velarinsert forms.

3.5—VELAR -INSERTION : A DIACHRONIC ANALYSIS THROUGH SYNCHRONIC

EXAMINATION . The diachronic development of the velarinsert class spans the entire recorded history of the Spanish language. Within this time period, there are various individual stages during which there have been changes affecting class membership and the AM affecting target forms. Additionally, there are several instances of synchronic variation between velar and nonvelar forms in many verbs. What follows is an overview of the synchronic stages that are essential to piecing together the diachronic puzzle.

Along with the preceding matters, treatment of these stages includes, as much as possible, discussion of how and why the velarinsert spread. Within each particular section, the verbs are generally divided according to their stemfinal (prevelar) segment. Whenever pertinent, details concerning the AM are also examined.

The data used in this investigation comes from a number of sources. Attested examples of velars and nonvelars alike are mainly found in the CORDE (Real Academia

155

Española 2005) and CREA databases (Real Academia Española 2007). The former

includes documents from the earliest recorded history of Spanish into the 20 th c. The latter is a database of both oral and written works from the late 20 th c. through the present. In some instances, information has been taken from some of the previous

investigations and historical analyses of Spanish mentioned in chapter 2. The dating of

texts in the CORDE has been compared against the PhiloBiblon database (Faulhaber et

al. 2002). When there is a notable discrepancy in the dating across these databases, the

text has been excluded. An attempt is made to utilize sources that are characterized by a

relatively brief chronological gap (usually less than 3 generations) between manuscript

and copy. The typical problems associated with the aforementioned databases should be

taken into consideration as they relate to the following sections.

Due to the volume of examples in some cases and the available filters, it has not been possible to completely differentiate the use of the SBJ from that of the IMPTV.

However, this is not considered a problem on the basis that the IMPTV is generally

identical in phonological form to the SBJ former and is representative of one of its pragmatic functions, i.e. influence. The inclusion of the IMPTV is especially relevant for counts of frequency since none of the first or last attested examples indicated below are believed to be IMPTV forms. In all cases, the frequency numbers are raw and are only meant to convey very general tendencies in the language.

3.5.1—STAGE 1: PRE -LITERARY DEVELOPMENTS .

Verb: tener , e.g. 1s tengo Earliest attested forms of velarinsert: preliterary, attested beginning of 12 th c. 156

Verb: venir , e.g. 1s vengo Earliest attested forms of velarinsert: preliterary, attested beginning of 12 th c.

Verb: poner , e.g. 1s pongo Earliest attested forms of velarinsert: preliterary, attested beginning of 12 th c.

Verb: remanir , e.g. 1s/3s SBJ remanga Earliest attested forms of velarinsert: preliterary, attested beginning of 12 th c. Last attested forms of velarinsert: mid14 th c.

The preliterary stage of velarinsertion affects the nasalfinal stems of tener ,

venir , poner and remanir . While other verbs are affected by velarinsertion during the

time when the first attested examples for these verbs are recorded (see salir and herir below), they differ in the fact that tener , venir , poner , remanir are marked by the

exclusive appearance of velar 1s/SBJ forms, i.e. they show no synchronic variation in the

recorded history of the langauge. Another point of contrast is that the other verb stems

are not nasalfinal and so are dependent upon the subsequent reformation of the AM on

the basis of the above forms as mentioned earlier. In sum, this nasalfinal set is affected by velarinsertion in an unrecorded stage after which they provide the basis for analogical

extension of the 1s/SBJ velarinsert pattern to nonnasal stems as demonstrated in the

following section.

The AM for velarinsertion in tener , venir , poner and remanir is derived from the

CL verbs ending in –NGERE , e.g. 1s tango , 2s tañes /tanzes (with secondary support from

the decir /hacer set, cf. §§3.2.1.2 and 3.3). The variables that characterize the

homogeneous supracontext and are essential to the emergence of the AM, in addition to

their implied status as second or third (i.e. not first) conjugation verbs, include a basic

CVCstem having a nonhigh stem vowel and a final coronal nasal, with a following 157

nonfront, nonthematic vowel marking the 1s/SBJ (cf. §§3.23.3). Note that the

inclusion of specific morphological, lexical and paradigmatic information makes it

difficult, if not impossible to formulate the conditioning environment for velarinsertion

in any other way. Although the AM and the context for change are now defined, there is

much more to say about 1s tengo , vengo , etc. and their leading role in the creation of the

velarinsert verbal subclass.

Were it not for the strength and number of associations between these verbs,

velarinsertion might not have been so productive in the Spanish language. One of the

most important aspects of this relationship is token frequency. This is obvious in that the

first synchronic slice of the diachronic development affects a set of highly frequent verbs.

Higher token frequency means increased input and equals increased lexical strength, not

only of the individual lexical forms, but also their distributional patterns as a consequence

of linguistic organization in the exemplar cloud. Thus, the establishment of a pattern of paradigmatic allomorphy (or ‘morphome’) in these verbs is salient enough to provide an

active AM that is on par with any other that is potentially available to the linguistic

individual.

Not only is the effect of the verbal paradigm in analogical change made apparent

in such instances, but also its capacity to inhibit phonological change. The regular phonological developments of nasal plus yod (and even lateral plus yod) did not

(generally) affect inflected verb forms. Similarly, the stem vowel in the 1s/SBJ forms of

Sp. tener and venir is unaffected by diphthongization, cf. 1s tengo , vengo ~ 2s tienes ,

vienes . The monophthong found in Spanish is typically attributed to yod effects and

158 phonotactic constraints involving closed syllables (cntr. Old Leonese and Old

NavarroAragonese, e.g. 1s tiengo ), though it is also necessary to appeal to the

interparadigmatic influence of the 1s/SBJ morphome of the AM (cf. Malkiel 1974,

García de Diego 1951). This explains why the effects of frequency and paradigmatic

associative strength tend to be linked to analogical change.

The lack of diphthongization in tener and venir is only one commonality between

these two verbs. In fact, they exhibit a special connection in many Romance languages.

In some dialects of Italian, like French, CL TENĒRE has shifted from the second to the

fourth conjugation, i.e. tenire , under the evident influence of venire (Tekavčić 1980b).

They are also thought to be included among the verbs showing yod effects that provide

the AM for poner , due to its being the only other 2 nd conjugation verb having a

nasalfinal CVC stem (see remanir below, cf. §2.5.2.1.1). This attraction to a highly similar AM (and avoidance of regular sound change) is supported by its inclusion in velarinsertion in Italian and Spanish and its having a palatalnasal in Portuguese

(Tekavčić 1980b; cf. Urrutia Cardenas & Alvarez Alvarez 1983). This is made all the more apparent given the attested OCamp. ponio as evidence of the availability of a form

with yod in Late Latin (Wilkinson 1978).

While remanir appears to be isolated from the effects of the aforementioned AM, it is easy to see the potential for a speaker to analyze it as a prefixed derivative of the verb manir (OSp. maner ) ‘remain’. In this way, it maintains the CVC stemstructure associated with the AM and is suscpetible to its analogical influence. Notwithstanding,

Malkiel (1974) suggests that remanir did not continue on in this group because its

159 phonological shape differentiated it from the other velarinserts at the time. Not only did

this verb lose its status as a velarinsert, it fell almost completely out of use in Spanish

during the 14 th c. It may have continued to be available to speakers, in spite of its being

generally replaced by quedarse , given its appearance (as anticuated and uninflected) in

the most contemporary version of the Diccionario de la Real Academia (Real Academia

Española 2004).

The ramifications of this preliterary stage are felt throughout every subsequent

stage of velarinsertion in Spanish through the changes in the composition of the AM and

the forms included in it. The unique combination of factors that contribute to the

formation of the AM provide the speaker with the ability to create a new lexical subclass built upon an originally phonological development affecting the language. The following

sections examine the changes in this subclass that result from reanalysis of the variables

that define the AM.

3.5.2—STAGE 2: LATERAL (AND RHOTIC ) EXPANSION .

Verb: salir , e.g. 1s salgo Earliest attested forms of velarinsert: middle of 12 th c. Variation: from mid13 th c. to beginning of 14 th c., intratextual, e.g. 1s/3s SBJ sala

Verb: herir , e.g. 1s hiero Earliest attested forms of velarinsert: middle of 12 th c. Variation: from beginning of 13 th c. to end of 16 th c., intratextual, e.g. 3p SBJ fiergan Last attested forms of velarinsert: end of 16 th c.

The first forms to be visibly targeted by the velarinsertion process are salir and

herir . These verbs are not marked by a nasalfinal stem consonant, but rather by a lateral

and rhotic respectively. This means that very early on, the AM is broadened through 160

reanalysis so that the preceding phonological conditioning environment includes the

natural class of coronal liquids and nasals, i.e. [voc, +son, +cor, +ant, dist]. The

motivation for such a change is related, in part, to the appearance of other

lateralplusvelar and rhoticplusvelar verbs in the language and their relationship to the

earlier AM. There is a set of verbs in CL ending in –LGERE and –RGERE that, given the high degree of identity to those of the more frequent –NGERE group, lend themselves to the AM. In other words, presuming that the –NGERE set is the primary AM for the preliterary nasalfinal targets, a natural progression of the AM is its extension to the next

most similar set of targets. Salir and herir share every aspect of the AM exemplified by tener , venir , poner and remanir except for the feature [nasal] of the stemfinal segment.

The (limited) support of the –LGERE and –RGERE sets of verbs and the established

relationship between the –NGERE AM and tener , venir , poner , remanir is sufficient to trigger reanalysis of salir and herir in the mind of the speaker.

The extension of the velarinsert to the lateralfinal roots is interesting because the expected regular sound changes involving palatalization and yod effects would have resulted in an allomorphic split in verbs like valer and salir initially between a palatal and

nonpalatal lateral (Malkiel 1974, cf. toller below). This phonological change would

have created allomorphy in the 1s/SBJ forms of the PRS tense and so supports the

function of the 1s/SBJ morphome charateristic of the AM. Similar arguments concerning

this paradigmatic pattern can be made for many of the targets in the following synchronic

stages. The limited and questionable appearance of palatalized forms of lateralfinal

verbs in Old Spanish in contrast to the widespread appearance of forms not showing yod

161

effects, e.g. 1s salo ~ salgo , argue in favor of the earlier loss of antihiatic yod in Old

Spanish (see §3.4; cf. Elson 1988, Lenfest 1978, 1993, Menéndez Pidal 1980).

As seen in the preceding example, the paradigmatic development of the lateral in

such situations provides no less than two fairly widespread solutions: (1) leveling in

favor of the nonpalatal lateral throughout the entire verbal paradigm, e.g. OSp. salo ,

valo , duelo , suelo , etc.; or (2) velarinsertion, e.g. salgo , valgo , duelgo , suelgo , etc.

(Malkiel 1974). Wilkinson (1978) suggests that valo and salo are the regular

development of the Castilian verbs (through loss of yod, cf. §§2.3.2.1 and 2.3.3) but were

replaced by the introduction of valgo and salgo from the neighboring dialects, a

replacement supported by the existing nasalfinal velarinserts. In addition to the

synchronic variation of both of the above outcomes in a single form, the distribution of

the velarinsert across all lateralfinal stems is also demonstrative of the nonsystematic

nature of analogical change. For example, while salir and valer survive into the standard

modern language as members of the velarinsert class, the variation found in soler and

doler is leveled in favor of a single nonvelar stem throughout their inflectional paradigms. 75 The differences between these lateralfinal pairs of verbs is discussed

further in §§3.5.3 and 3.5.5.

Although salir and herir show the effects of velarinsertion during what could be

considered the same synchronic stage as those of the preceding section, they are

distinguished from the firststage (nasalfinal) velarinserts by attested velar/nonvelar

variation. For salir , this variation is mainly limited to the middle of the 13 th through the beginning of the 14 th c. Within a majority of the documents containing a nonvelar stem

162

for the 1s/SBJ forms, the velar variant is also attested. The remaining nonvelar

examples are heavily outnumbered by the exclusive appearance of velarinsert forms in

contemporaneous texts.

The most productive period of variation for herir closely resembles that of salir ,

occurring between the 13 th and 14 th c., with the notable difference that the velarinsert

variant is lost. The question is why herir does not retain membership in the velarinsert

class. The answer is that velarinsertion tends to have very little effect on rhoticfinal

verbs. Wilkinson (1979) finds it difficult to establish a model for the insertion of /g/ in

the rhoticfinal verbs. The reason for this difficulty involves the characteristics of the

AM and the lack of a sufficiently strong and homogenized set of exemplars marked

stemfinally by a rhoticplusvelar combination. First, /n/ and /l/ share the feature [+cont]

to the exclusion of the flap /ɾ/ in Spanish. 76 While they are all sonorants, and so are distinguished from the obstruent class, the fact that /ɾ/ is one step removed from the strongest and most productive AM for velarinsertion, it is less likely to trigger restructuring. It is also not susceptible to palatalization as are /n/ and /l/, thus, it does not show the same potential for yod effects. An additional factor is the absence of any highly salient 2 nd or 3 rd conjugation verb forms whose stems are characterized by the /ɾg/ cluster.

In fact, the CL verbs ending in –RGERE that managed to survive into the Spanish language undergo a separate and very complicated series of changes (cf. Malkiel 1974).

Of the few avaiable rfinal 2 nd and 3 rd conjugation verbs in Old Spanish, only

herir (and its derived forms) shows any attested velar forms, in most cases, in sychronic

variation with a nonvelar in the same document.77 Taking MSp. 3p SBJ hieran as an

163

example, it undergoes several interesting changes. First, it loses its original yod and then

is susceptible to remodeling on the basis of the velarinsert AM, producing synchronic

variation, i.e. OSp. fiergan ~ fieran .78 The velar variant is ultimately eliminated through intraparadigmatic analogy of the 1s/SBJ forms with the remainder of its inflectional paradigm not having /g/, i.e. fiergan >> hieran (Urrutia Cardenas & Alvarez Alvarez

1983). 79

Although lateral and rhoticfinal verbs stems are ultimately distinct with respect to velarinsertion, this stage demonstrates how speakers redefine the AM through weakening of its feature specifications. The term ‘weakening’ is especially appropriate in this case because velarinsertion is not systematic and shows rampant synchronic variation and attrition. The following stage shows how continued weakening of these specifications define the natural limitations of the AM.

3.5.3—STAGE 3: SONORITY AND THE AM.

Verb: soler , e.g 1s suelo Earliest attested forms of velarinsert: middle of 13 th c. Variation: from mid13 th c. to mid16 th c., e.g. 1s suelgo Modern dialectal usage: Last attested forms of velarinsert: middle of 16 th c.

Verb: valer , e.g. 1s valgo Earliest attested forms of velarinsert: middle of 13 th c. Variation: from mid13 th c. to beginning of 14 th c., intratextual, e.g. 3p SBJ valan

Verb: toller , e.g. 3p tuellen Earliest attested forms of velarinsert: beginning of 13 th c. Variation: from mid13 th c. to end of 15 th c., intratextual, e.g. 3p SBJ tuelgan Last attested forms of velarinsert: end of 15 th c.

Verb: caer , e.g. 1s caigo 164

Earliest attested forms of velarinsert: middle of 13 th c. (cf. §3.4) Variation: from mid13 th c. to beginning of 17 th c., intratextual, e.g. 1s/3s SBJ caya

Verb: huir , e.g. 1s huyo Earliest attested forms of velarinsert: middle of 13 th c. Variation: from mid13 th c. to beginning 17 th c., intratextual, e.g. 1s huigo Modern dialectal usage: , , , Spain Last attested forms of velarinsert: beginning of 18 th c.

Verb: oír , e.g. 1s oigo Earliest attested forms of velarinsert: middle of 13 th c. Variation: from mid13 th c. to mid17 th c., e.g. 1s/3s SBJ oya

Verb: traer , e.g. 1s traigo Earliest attested forms of velarinsert: middle of 13 th c. Variation: from mid13 th c. to beginning of 17 th c., intratextual, e.g. 3p SBJ trayan

Verb: coser , e.g. 1s coso Earliest attested forms of velarinsert: middle of 13 th c. Variation: from mid13 th c. to end of 16 th c., e.g. 1s/3s SBJ cosga Modern dialectal usage: Spain, JudeoSpanish Last attested forms of velarinsert: end of 16 th c.

Once salir establishes itself as a member of the AM, the extension of the

velarinsert to lateralfinal stems like soler and valer is straightforward (cf. Fondow

2007). The identity of the relevant variables in target and model explains the potential for reanalysis in the former. Velarinsertion, in such instances, is not obligatory, but rather all the pieces are in place to promote change. The extremely high frequency and salience of the members of the AM, as well as the relatively high frequency of the target verbs (though lesser than that of the model forms), seems to increase the likelihood that they will, at least partially, participate in the analogical process (cf. §4.3).

From the immanent point of view, it is not surprising that valer maintains the velar into the standard modern language while soler does so only dialectally. At this synchronic stage, the difference primarily involves frequency.80 Because the velarinsert 165

class includes a number of heavily used 2 nd and 3 rd conjugation verbs, any other such

verbs are likely to be compared against them to determine the degree of similarity in

relation to frequency. If a form appears frequently enough in everyday usage, it is less

likely to be restructured unless there is another highly similar and more frequent/salient

form (or set of forms) that can serve as the basis for analogical change. In contrast, if a

form appears less frequently, it is more susceptible to competing influences, especially pressure from the class to which it already belongs (i.e. no change whatsoever) or the

most frequent in both type and token (i.e. for soler , the set showing diphthongization in

the stressed nucleus of the verb stem without velarinsertion).

Frequency is intimately related to direct conflict between two or more competing

AM. When this occurs in a single speaker or group of speakers, it results in the

synchronic variation that is found in verbs like soler and earlier herir , as well as almost

all the verbs in the remaining stages below. While the highfrequency verbs that have been examined up to this point have shifted in favor of the velarinsert, the competing

influence of other models tends to define the diachronic developments affecting the

mid to lowfrequency verbs examined in the following stages.

Although it may be included as a lateralfinal stem alongside soler and valer , the

case of toller is rather exceptional. During the period in which it shows velarinsertion,

the is fairly consistent regarding the geminate –ll in inflected forms. For

this reason, it is safe to assume that the stemfinal phoneme is palatal. There are two possible interpretations of the analogical process affecting toller . First, it could be

representative of a change in the feature [coronal] from variables relevant to the

166

stemfinal phoneme of the AM. The palatal lateral would depalatalize in syllablefinal position (after velarinsertion) due to the phonotactic constraints of the language. The motivation for such an explanation lies in the expansion of the AM along phonological lines (see asir , caer , huir , etc. below). The second hypothesis is that the AM for toller includes the verb cullir ‘grab’ which contains a presumably etymological velar in the

1s/SBJ in alternation with a palatal lateral (cf. Wilkinson 1978). There are several reasons that the former hypothesis is more plausible. The broadening of the phonological factors in the AM is a characteristic of this synchronic stage. It also assumes the second analysis since the distribution of the velar in cullir , in the broader phonological context of the AM, makes it part of the continuously reanalyzed analogical source for velarinsertion. Additional evidence comes from the modern dialectal extension of the velarinsert to verbs like bullir and zambullir (cf. §3.6.2).

As suggested above, the spread of the velarinsert at this stage is a consequence of

the broadening (or weakening) of the relevant phonological features of the stemfinal

segment (cf. §2.3.1.1). Crucially, this change in the AM follows a phonological path that

directly correlates to sonority. In part, the relevance of sonority is a consequence of the phonotactic constraints on the Spanish syllable, nonetheless, it is an important element of

the AM that has not been explored previously. The changes in the relevant phonological

features in the stemfinal segment of the AM are such that they follow the Sonority

Hierarchy in both directions from the coronal sonorants /n, l, ɾ/, which defined it in the

earliest stages, to the coronal glide /j/ and the coronal sibliant /s/ (as well as its possible

shift to noncoronal palatals, as in toller above). The transition to targeting of yodfinal

167

verb stems involves increased sonority and loss of specification for the feature [ant]. In

the direction of decreasing sonority, meanwhile, the phonological context is expanded to

include sibilantfinal coser through removal of the feature specification for [son]. From this stage forward, the features relevant to the immediately preceding phonological conditioning environment are [voc, +cor, dist]. The forms that are subject to velarinsertion dialectally in the modern language exemplify the relevance of these phonological criteria in the composition of the AM.

The development and extension of the velarinsert at this stage represents a process that is unique to Spanish. The verbs huir and oír (as well as reír , haber , etc. below) are characterized in earlier stages of Old Spanish by an antihiatic yod (or coronal

‘semiconsonant’, cf. Malkiel 1974) that emerges through regular sound change. In a

verb like OSp. 1s fuyo , this ‘semiconsonant’ is derived from the phonological change by

which CL FUGIŌ shows palatalization effects in the velar consonant followed by yod

(resulting from its contact with an immediately following vowel) (i.e. /gj/ > /ɟɟ/ > /ɟ/, cf.

§2.3.2.1). Notably, the 1s/SBJ paradigmatic distribution of this development coincided in

many cases with that of the already established velarinserts, primarily because these

were the same cells in the paradigm which contained the necessary vowels in hiatus so as

to trigger the formation of antihiatic yod. The resulting palatal consonant in forms like

1s fuyo takes on its own antihiatic function and helps to establish the AM for the emergence of yod in mainly monosyllabic roots like those of caer and traer (and

concluir , creer , destruir , raer , roer , ir below), providing OSp. 1s cayo , trayo (cf. Alarcos

Llorach 1954, Cano Aguilar 1999, García de Diego 1951, Lloyd 1987, Malkiel 1974,

168

Penny 2002, Wilkinson 1981). This analogical extension, “presented an established pattern of conjugation to follow…Thus we may say that in Spanish a subconscious sense developed that for stems ending in a vowel it was proper to add a y before –o and –a”

(Wilkinson 1981:78). In this way, hiatus between *aa (haya ), * ao (cayo ), * oo (royo ) was generally avoided (cf. rare dialectal 1s roo , rao , etc. and the accepted combinations ae

(cae ), eo ( ), ee (cree ), ea (vea )) (Malkiel 1974). The extension of the antihiatic consonant to these forms is attributed to changes in phonotactic preferences on the part of speakers towards:

[A] new, ‘ideallyshaped’ syllable: C + V…then it becomes clear why a solid block of speakers, at a variety of socioeducational levels, went from here one step farther and rejected cayo , oyo , also hayo , then again subj. haya and vaya beside creya and leya in favor of caigo , oigo , huigo , dial. creiga , etc. Granting that C + V was gradually emerging –within the speech community – as the ‘ideal’ syllable, then it was obviously gratifying to speakers endowed with one spark of initiative to place in this favored position a fullblown consonant, such as /g/, rather than having a mere semiconsonant, such as /j/, perform the crucial rôle. (Malkiel 1974:336)

The transition to velarinsertion in these same forms follows similar lines, as it

(often) affects the aforementioned 1s/SBJ variants having an antihiatic palatal consonant

(cf. the identical paradigmatic distribution of /k/ in the ‘inchoatives’). These yodfinal

verb stems are part of a group having a high degree of identity and similar relative

frequency which explains their susceptibility to velarinsertion at the same synchronic

stage, not to mention that they are affected by velar/nonvelar variation over the same

time period and serve as the primary AM for yodfinal verbs in subsequent synchronic

stages. As before, velarinsertion in these forms involves mutations in the variables that

169

define the evergrowing AM and the high frequency exemplars included in it (cf. Malkiel

1974, Menéndez Pidal 1980).

The synchronic variation in nearly all of the verbs affected by velarinsertion during this stage continues dialectally in the modern language. This is telling of the weakening of the associative strength of the AM as the number of analogical targets increases. As the definition of the AM is widened, it becomes less specific such that competing forces are more likely to interfere due to decreased salience. As a matter of fact, all of the verbs affected by the velarinsertion process from this point forward are plagued by the same conflict. Further treatment of this type of modern variation is found in §3.6.

The effects of velarinsertion on sibilantfinal verbs like coser is also indicative of the conflict between competing AM. Due, in part, to the low token frequency of this verb, it is susceptible to reanalysis (and uncertainty) in the mind of the speaker. This is the last synchronic stage that includes a number of verbs (prescriptively) included in the standard MSp. velarinsert class due to the much higher frequency of their velarinsert variants diachronically. These verbs, in all forms, are also of notably higher token frequency than the majority of those found in the following stages. Coser , however, falls

into the latter category as it has extremely low token frequency and involves such a broadening of the AM that it is difficult for speakers to fully commit to the idea that it

should be marked by a velar in the 1s/SBJ. In most dialects of Spanish, this is

compounded by the appearance of other verbs in the homogeneous supracontext, e.g.

cocer “cook” and/or toser “cough”. Given the complexity of this situation, it is clear that

170

such targets test the permissible (phonological) limits of the AM. Sibilantfinal stems are

examined further in treatment of asir below.

The remaining stages include verbs that are primarily marked by velarinsertion in limited, dialectal contexts. Therefore, it is possible, in a very general sense, to identify this synchronic stage as the last truly productive period of velarinsertion in the standard language, especially considering that at least one verb of each of the stemfinal consonants that characterize the AM has already been affected by the process. There are numerous factors influencing this assertion including frequency, learnedness, decreased associative strength and salience of the broadened AM, increased salience of competing

AM, etc. The first of these factors is examined further in chapter 4, while the others are discussed below.

3.5.4—STAGE 4: CONTRASTING DIACHRONIC RESULTS .

Verb: destruir , e.g. 1s destruyo Earliest attested forms of velarinsert: beginning of 14 th c. Variation: from beg. of 14 th c. to beg. of 17 th c., intratextual, e.g. 1s/3s SBJ destruiga Modern dialectal usage: Unconfirmed Last attested forms of velarinsert: beginning of 18 th c.

Verb: instruir , e.g. 1s instruyo Earliest attested forms of velarinsert: beginning of 14 th c. Variation: beginning of 14 th c., e.g. 3p SBJ instruigan Last attested forms of velarinsert: beginning of 14 th c.

Verb: haber , e.g. 3s/3p hay ; 3p SBJ hayan Earliest attested forms of velarinsert: beginning of 14 th c. Variation: from beginning of 14 th c. to present, intratextual, e.g. 3p SBJ haigan Modern dialectal usage: , , Cuba, Dominican Republic, Mexico, , , , , Spain,

171

Since the verb huir entered the velarinsert subclass in an earlier stage, very little

is generally said about Spanish verbs ending in –uir like destruir and instruir (cf. Urrutia

Cardenas & Alvarez Alvarez 1983). While the analogical relationship between these

targets and their model is straightforward and includes all of the previously examined

variables pertaining to the established AM, these verbs are discussed in conjunction with

those found in a later stage as they offer a unique example of language change.

Relatedly, for the verbs affected in this and subsequent stages, with a few notable

exceptions, the nonvelar is the overwhelmingly dominant variant throughout. In some

instances, velarinsertion of a particular verb is restricted to a single synchronic stage, or

even to a single author or text (see §3.5.6).

In contrast to the other verbs included in this stage, most historical examinations

of the Spanish language devote special attention to the verb haber . This is due, in large part, to its multiple functions, meaning “exist”, “have, possess” and its use as an . Consequently, there are innumerable tokens throughout the recorded history of the language. Although haber is not a standardized member of the velarinsert subclass, it is one of few verbs to show continuously attested variation from the first appearance of the velar through the present. In most other verbs of this sort, velarinsertion is attested for a highly limited period of time, later reappearing in modern, dialectal contexts. Another reason for the popular treatment of haber is that it is also affected by a number of interesting phonological and morphological processes. The palatal stem appears in the

SBJ forms of the earliest Old Spanish texts. As indicated above, the velarinsert is attested several centuries later and, like all the other verbs affected at this point in time, it

172

is modeled on the welldeveloped and productive set of velarinsert verbs, including the

highly similar caer and traer .

The relationship between models like caer and traer and a target like haber is

interesting because the former are prescriptively velarinsert verbs (practically from the

first appearance of the velar) while the latter is not. Also, caer and traer undergo velarinsertion at roughly the same point in time, with haber following relatively later.

Aside from several other arguments for distinguishing haber from its AM for velarinsertion (e.g. paradigmatic alternations, defective paradigms, morphologization of certain patterns of use, etc.), token frequency once more involves itself in the matter.

Haber is among the most frequent of all lexemes (not just verbs) in the language and so is

immanently of greater lexical strength and ‘autonomy’ in the mind of the speaker (cf.

Bybee & Brewer 1980). Such cognitive salience, in association with other exemplars or

AM, means that it may be more resistant to analogical change, especially when there are

other factors, outside of the essential variables of the AM, that are not completely

identical across model and target.

3.5.5—STAGE 5: PRESCRIPTIVE VARIATION.

Verb: doler , e.g. 1s duelo Earliest attested forms of velarinsert: beginning of 15 th c. Variation: from beginning of 15 th c. to mid16 th c., intratextual, e.g. 1s/3s SBJ duelga Modern dialectal usage: Spain Last attested forms of velarinsert: mid16 th c.

Verb: raer , e.g. 1s raigo , rayo Earliest attested forms of velarinsert: beginning of 15 th c. Variation: from beginning of 15 th c. to present

173

Verb: reír , e.g. 1s río Earliest attested forms of velarinsert: middle of 15 th c. Variation: from mid15 th c. to beginning of 16 th c., e.g. 3s SBJ riga Modern dialectal usage: Spain Last attested forms of velarinsert: beginning of 16 th c.

Verb: roer , e.g. 1s roigo , royo , roo Earliest attested forms of velarinsert: middle of 15 th c. Variation: from mid15 th c. to present, intratextual

The appearance of the velar in lateralfinal doler overlaps very slightly with that

of soler , and the support of verbs like valer and salir , along with the AM as a whole,

contribute to its susceptibility to velarinsertion. Like soler , it does not continuously

show velarinsertion diachronically, even though it is attested in modern dialects of the

language (cntr. haber , §3.5.4).

The remaining three verbs in this stage are more closely related because they are

all derived from CL verbs containing an intervocalic /d/, i.e. RADERE > raer , RĪDĒRE > reír , RŌDERE > roer . A stemfinal palatal segment emerges in these verbs as a means of

eliminating an intolerable hiatus in the 1s/SBJ forms, e.g. 1s rao from raer , following an

AM including verbs like huir , oír , caer and traer above (cf. §3.5.3, Cano Aguilar 1999,

García de Diego 1951, Lloyd 1987). As a consequence of this earlier analogical change,

these forms become targets for velarinsertion. In addition, each has at least two attested

synchronic variants. Interestingly, all the variants for raer (e.g. 1s rayo /raigo ) and roer

(e.g. 1s roo /royo /roigo ) are prescriptively accepted in the modern language (Real

Academia Española 2004, Kendris 1996b).

Another notable point about raer and roer is that they are exceptions to the

diachronic dilemma described above because they seem to attest continuous synchronic 174

variation between velar and nonvelar through the present. Unfortunately, the numbers

concerning raer are obstructed by the interference of rayar ‘underline; scratch’ (PRS

IND and IMPF) and N rayo ‘ray; spoke; thunderbolt’, raya ‘line’, and by the low token

frequency of the verb itself. Notwithstanding, during the period between the only two

attested instances of 3p raigan (15th to mid18th c.), there are no examples of the

nonvelar equivalent, as all instances of 3p rayan are IND forms of rayar . Notably,

during that same time period, there are two attested cases of 3p SBJ rayen from this same

first conjugation verb.

With respect to roer , the palatal stem appears in synchronic variation with the

velar in the same mid15 th c. text and in contemporaneous texts for all attested examples until beginning of 17th c., at which time the palatal variant seems to fall out of use. The third variant, with vocalic hiatus, is first attested in the 16th c. and continues to appear fairly consistently, and much more frequently than the other variants, through the 20th c.

The complications associated with velarinsertion at this stage are indicative of the weakening of the AM and the fact that its productivity is becoming heavily restricted due to competing, more salient AM and frequency effects. The prescriptive variation associated with the verbs raer and roer provides evidence in support of these arguments.

In fact, they are also prime examples of the directions of language change associated with vowel and palatalfinal verbs stems in Spanish, something which is even more apparent in the following synchronic stage.

175

3.5.6—STAGE 6: EXTREME VELAR -INSERTION AND IDIOSYNCRATIC ANALOGY .

Verb: moler , e.g. 1s muelo Earliest attested forms of velarinsert: middle of 16 th c. Variation: from mid16 th c. to early 17 th c., intratextual, e.g. 1s muelgo Modern dialectal usage: Spain Last attested forms of velarinsert: early 17 th c.

Verb: creer , e.g. 1s creo Earliest attested forms of velarinsert: beginning of 17 th c. Variation: from beginning to middle of 17 th c., intratextual, e.g. 1s creigo Modern dialectal usage: Colombia Last attested forms of velarinsert: middle of 17 th c.

Verb: atribuir , e.g. 1s atribuyo Earliest attested forms of velarinsert: beginning of 16 th c. Variation: from beg. of 16 th c. to beg. of 17 th c., same author, e.g. 1s atribuigo Last attested forms of velarinsert: beginning of 17 th c.

Verb: concluir , e.g. 1s concluyo Earliest attested forms of velarinsert: beginning of 17 th c. Variation: beginning of 17 th c., intratextual, e.g. 1s concluigo Modern dialectal usage: Spain Last attested forms of velarinsert: beginning of 17 th c.

Verb: contribuir , e.g. 1s contribuyo Earliest attested forms of velarinsert: beginning of 17 th c. Variation: beginning of 17 th c., same author, e.g. 3p SBJ contribuigan Last attested forms of velarinsert: beginning of 17 th c.

Verb: disminuir , e.g. 1s disminuyo Earliest attested forms of velarinsert: beginning of 17 th c. Variation: beginning of 17 th c., same author, e.g. 1s/3s SBJ disminuiga Modern dialectal usage: Unknown (unconfirmed as modern though plausible) Last attested forms of velarinsert: beginning of 17 th c.

Verb: excluir , e.g. 1s excluyo Earliest attested forms of velarinsert: beginning of 17 th c. Variation: beginning of 17 th c., same author, e.g. 1s excluigo Last attested forms of velarinsert: beginning of 17 th c.

Verb: extruir , e.g. 1s extruyo Earliest attested forms of velarinsert: beginning of 17 th c. 176

Variation: beginning of 17 th c., e.g. 1s/3s SBJ estruyga Last attested forms of velarinsert: beginning of 17 th c.

Verb: instituir , e.g. 1s instituyo Earliest attested forms of velarinsert: middle of 16 th c. Variation: from mid16 th c. to beginning of 18 th c., e.g. 1s instituigo Last attested forms of velarinsert: beginning of 18 th c.

Verb: sustituir , e.g. 1s sustituyo Earliest attested forms of velarinsert: end of 16th c. Variation: from end of 16 th c. to beginning of 17 th c., e.g. 1s sustituigo Last attested forms of velarinsert: beginning of 17 th c.

Verb: asir , e.g. 1s asgo Earliest attested forms of velarinsert: beginning of 16 th c. Variation: from beginning of 16 th c. to beginning of 17 th c., e.g. 1s asso Modern dialectal usage: Standard, prescriptive (though generally not used)

Although the verbs included in this stage are more widespread chronologically, their periods of variation overlap sufficiently with each other, as well as with the previously affected verbs so as to warrant grouping them together. In addition, they are

representative of the massive expansion of territory in which the Spanish language is

used, as the data from prior stages is primarily restricted to the . From

this point forward, sociolinguistic factors play a tremendous role in the velarinsertion process, as productivity is limited to dialectal, nonprescriptive usage (even in the case of asir , see below).

Moler has the benefit of a slight chronological overlap between its first attested velarinsert and the last attested instances in soler and doler . This is advantageous from the standpoint of the immanent take on AM since these verbs share an ever so slightly higher degree of homogeneity than the other, already wellestablished instances of velarinsertion, even compared to lateralfinal verbs like valer and salir . However, by 177

this stage, the AM has been broadened to the point where the specific phonological shape

of the verb stem may be considered less relevant.

A similar analysis explains the appearance of the velar in creer . The support of

the other verbs in the AM not having a stemfinal nasal or lateral is extremely important

in this case because creer does not make use of a stemfinal palatal as do so many of the other verbs analyzed herein. The availability of an AM showing a 1s/SBJ distribution of the velarinsert, often in contrast to forms showing hiatus, e.g. 1s traigo , 2s traes , increases the likelihood that speakers may reanalyze creer as a member of this verb class.

Thus, the importance of the inflectional vowel and the paradigmatic distribution of the velar are made all the more apparent.

As noted above, verbs ending in –uir , like destruir and instruir , are demonstrative of a possible direction of change. These mostly polysyllabic verbs are characterized at this synchronic stage by a stemfinal palatal ‘semiconsonant’ in all forms of the PRS except the 1p and 2p IND, e.g 1p IND disminuimos ~ 3p IND disminuyen . As in the case of creer above, the paradigmatic distribution of nonfront inflectional vowels is crucial to the emergence of the velar in these verbs. Huir is obviously at the center of the AM for these verbs, given the continued presence of its velar variant and its high degree of similarity, or homogeneity, with these targets.

The relationship between huir , destruir , atribuir , concluir , contribuir , disminuir ,

excluir and sustituir , is such, in fact, that they all appear with a velarinsert in texts from a single Spanish author whose works date from the beginning of the 17 th c., in several instances, representing the only attested example of velarinsertion in these verbs. The

178

velar and nonvelar forms often alternate indifferently within the same text, seemingly

indicating a permissible degree of synchronic/idiosyncratic variation in the language used by this particular author. The random nature of this variation is displayed in Figure 3.3 below. It is telling of the susceptibility to velarinsertion of palatalfinal stems and, more

specifically, of the more restricted set of verbs with an INF ending in –uir . These verbs share the necessary variables with the AM and are of a mid to lowfrequency suggesting that they may be continuously reshaped as the result of reanalysis. In other words, the associative strength and ‘autonomy’ of these verbs is low due to their frequency and the availability of a highly similar AM in direct competition with the lexical forms that have been used up to this point. For the author in question, this is enough to cast doubt on exactly how these forms are conjugated, creating the (perhaps conscious) intratextual variation found in his writing (cf. Joseph 1997a, 1997b).

INF Velars Non-velars % Velar Date (# docs) Velars Non-velars %Velar

atribuir 4 1 80% 16031607 4 3 57% concluir 5 36 12% 1605 1 0 100% contribuir 1 0 100% 1606 (2) 7 4 64% destruir 9 5 64% 1607 (2) 30 13 70% disminuir 1 7 13% 1609 (2) 5 6 45% excluir 1 1 50% 16091610 (2) 3 1 75% huir 60 0 100% 1610 (3) 6 4 60% influir 0 1 0% 16101612 9 8 53% sustituir 2 0 100% 1613 (4) 18 12 60%

Total 83 51 62% Total 83 51 62% Figure 3.3—Variation of –uir verbs in S. Juan Bautista de la Concepción (17 th c.).

179

This type of variation and the gradual extension of the velarinsert through the lexicon is described in Wang (1969):

According to this view [of gradual lexical diffusion], during the early phase of the change only a small sector of the relevant morphemes is affected. Some of the affected morphemes may change to the Ypronunciation directly. Other morphemes, however, will at first have both the Xpronunciation and the Ypronunciation, fluctuating either randomly or according to some such factor as tempo or style. (14)

Although unintentional, this description of the diachronic situation is wholly applicable to analogical change of the sort explored here (cf. KraskaSzlenk 2007). Velarinsertion is initially restricted to a localized set of verbs and is then extended to others as a synchronic competitor. This results in changes in the composition of the AM such that the entire process begins anew. Figure 3.3, meanwhile, demonstrates the synchronic

‘random fluctuation’ that may affect some analogical targets.

For the verbs ending in –uir , it is also important that they may all be considered learned or semilearned, appearing for the first time at a much later stage than the verbs affected by the first and most productive stages of velarinsertion. Of this subclass of verbs, only huir and destruir are attested before the 13 th c. and nearly all the others are

first attested between the 14 th and 15 th c. 81 It should be noted that the 8 verbs that are at

the core of the velarinsert subclass across all modern dialects of Spanish, i.e. caer , oír , poner , salir , tener , traer , valer , venir , all appear in the earliest recorded history of the language, prior to the first attested forms of the verbs under examination here. Moreover, these ‘original members’ undergo velarinsertion before the first appearance, velar or otherwise, of nearly the entire –uir set. This is yet another point in favor of the

180

synchronic divisions and loss of productivity according to changes in the AM proposed in

this investigation.

Like the –uir set, the first attested appearance of the verb asir is at a very late stage relative to velarinsertion. Unlike the former, however, asir is often a focus of

diachronic analyses because it is still prescriptively considered part of the velarinsert

subclass in the modern language. The inflected forms of this verb, however, are largely

unused and unknown by most modern speakers (cf. Fondow 2007). 82 The extremely low frequency and cloudy etymological history of this verb complicate the matter and have given rise to a number of hypotheses, some more plausible than others, regarding the development of the velarinsert forms. Wilkinson (1981) proposes a fairly widely held theory that the velar is derived analogically from the combined set of inchoatives, e.g. 1s conozco , traduzco and already existing velarinserts, e.g. 1s tengo , valgo , caigo , etc.

(including 1s digo , 2s dices ). 83 However, an AM including forms like 1s conozco in the

AM for 1s asgo violates the homogeneous supracontext assumed under AEM and so prohibits such an analysis. Nonetheless, the modern outcome establishes an associative

relationship between them that is of interest with respect to the salience of the 1s/SBJ paradigmatic template discussed further in §4.2.

This represents the last stage in which velarinsertion affects the language proper.

What remains of velarinsertion is its analogical extension in localized, dialectal contexts that move beyond the prescriptivelyassigned verbs in the standard langauge. The next section explores this more limited, though still productive extension of the velar.

181

3.6—DIALECTAL AM AND POTENTIAL DIRECTIONS OF CHANGE . The dialectal

developments involving velarinsertion are divergent and therefore are more indicative of

the potential directions for analogical change than of the commonly shared characteristics beween dialects. First and foremost among the dialectal examples is the more

widespread extension of the velarinsert in yodfinal stem allomorphs in rural dialects,

e.g. huiga , haiga , vaiga , creiga , leiga , reiga . According to Malkiel (1974), such forms

exemplify the acceptance of earlier attested instances of velarinsertion that lost out to

another variant in the standard language. In addition, they support the psychological

reality of paradigmatic structure and its associative strength against the less cohesive

groupings typical of other syntactic categories. These forms also demonstrate the

changes in the definition of the AM for Spanish velarinsertion. The synchronic variation

and limited productivity that characterizes the later stages of the process is explained by

the constant struggle of competing AM:

Thus one (minor) analogical pressure within a verbal paradigm can deflect a few of its members from the expected path, onto which they may, three or four generations later, be pushed again through a stronger pressure exerted from a different direction. (Malkiel 1967:26)

The preceding quote implies the constant search for connections between one form and

another, or analogy in the first sense as proposed in §1.1. Associations of this sort follow

multiple linguistic and psychological lines and are at the heart of the immanent perspective. As a consequence of the realtime comparison of exemplars, language, even

at the level of the individual, is dynamic and susceptible to reorganization and

reformation. This explains the synchronic variation and diachronic changes in the

velarinsert subclass that continue to show their effect in the modern language. 182

3.6.1—THE QUESTION OF OUTSIDE INFLUENCE . Before discussing modern dialectal

velarinsertion in more detail, it is necessary to acknowledge the likely external

influences affecting the development of the Spanish velarinsert. Dialectal examples, and

even some of the earlier attested instances, could certainly be the result of borrowing or

analogical influence of neighboring languages/dialects. However, due to the

chronological overlap of the appearance of the velarinsert in targets and models

throughout the entire diachronic process in Spanish, it is unnecessary for present purposes to explore the influence of other languages. This argument is corroborated by

Malkiel (1974): “Spanish morphology…has been far less affected by pressures from the

outside than has the Spanish lexicon” (341). In the early stages of the process, there was

most certainly influence from NavarroAragonese and possibly Catalan. Similarly, the

modern examples must be situated geographically against the developments affecting

these languages. Nonetheless, it is possible to establish an analogical relationship

exclusive to Spanish in order to explain the diachronic process of velarinsertion.

3.6.2—STAGE 7: DIALECTAL VELAR -INSERTION IN MODERN SPANISH .

Verb: bullir , e.g. 1s bullo Earliest attested forms of velarinsert: early of 20 th c., e.g. 2s SBJ buigas Modern dialectal usage: Mexico

Verb: dar , e.g. 1s doy Earliest attested forms of velarinsert: 20 th c., e.g. 1s/3s SBJ daiga Modern dialectal usage: Spain

Verb: ir , e.g. 1s voy Earliest attested forms of velarinsert: middle of 19 th c., e.g. 2s SBJ vaigas Modern dialectal usage: Rural Spain, Mexico

183

Verb: parir , e.g. 1s paro Earliest attested forms of velarinsert: unconfirmed, e.g. 1s/3s SBJ parga ? Modern dialectal usage: unconfirmed

Verb: ver , e.g. 1s veo Earliest attested forms of velarinsert: middle of 19 th c., e.g. 2s SBJ veigas Modern dialectal usage: Spain

Verb: zambullir , e.g. 1s zambullo Earliest attested forms of velarinsert: early 20 th c., e.g. 1s/3s SBJ zambuiga Modern dialectal usage: Colombia

These dialectal cases, as well as those of all the verbs affected by velarinsertion in the preceding sections, are demonstrative of the limits of the AM according to the phonological constraints on the shape of the syllable in Spanish. The broadening of the

AM through the loss of phonological specificity in the conditioning environment corresponds to those phonemes which are permitted naturally (i.e. not derived through borrowing or affixation) in syllablefinal position. This set includes vowels, glides, nasals, laterals, rhotics and . Consequently, the potential for continued analogical change is restricted to verbs characterized by these phonemes in stemfinal position, as demonstrated by the fact that each of the cases above involves an analogical target that resembles a previously examined analogical change.

This continued potential is readily apparent in the expansion and growth of such a highly reduced, lexically idiosyncratic allomorphic verb stem to a productive, surprisingly widespread categorical indicator. In the end, velarinsertion comes to be affiliated with a tightly restricted, though fairly systematic grouping of verbs.

Notwithstanding, in MSp. there are also a handful of cases where velarinsertion does not occur where anticipated (e.g. pulir ‘to polish’, abolir ‘to abolish’). These examples

184

demonstrate the loss of productivity, weakening and/or localized nature of the AM, even

in cases where the target meets all necessary characteristics of the model. Such

unaffected targets are also important with respect to the functions of frequency and time.

They are much less frequent than those verbs that are central to the standardized

velarinsert category. Consider, for example, that the extremely low frequency verbs pulir and abolir are not attested prior to the 15 th c., meaning that they are not available for reanalysis during the most productive period of velarinsertion in lateralfinal verb stems

(i.e. prior to the end of the 13 th c.). Forms that are not used or not available at the right period of productivity seem to have a decreased chance that analogy might take hold in

spite of meeting all the other the necessary criteria for change to occur.

Even if they are available, competing analogical pressures may prove

insurmountable for any number of reasons. Presumably, even if an innovative or learned

form were to be affected by velarinsertion, it would typically do so only dialectally or in

synchronic variation with a nonvelar form (cf. moler , doler above). This point holds

true in a general sense for all the verbs that have or could have been affected by

velarinsertion due to their similarity to the AM. The other side of this issue involves

those forms which have been present throughout the history of Spanish but do not

undergo velarinsertion. Verbs like unir ‘unite’ and morir ‘die’, rather frequent and

highly used forms, which are characterized by all the necessary variables to attain the

velar obstruent, instead follow their original evolutionary path. For these verbs, it seems

that the analogical pressure of this model is not sufficient and is outweighed by another.

185

3.6.3—ANALOGICAL CHANGE ACROSS DIALECTS . Velarinsertion has taken different

directions in all affected languages and dialects, as shown in Appendix F. 84 Consider the extreme example of neighboring Catalan, where the velar has been extended exponentially across several high frequency verbal paradigms in many dialects due to additional contributing factors in the AM (cf. §2.2.2 and §4.5.2). While Spanish does show dialectal extension of the velarinsert, it is severely restricted by comparison.

For example, the changes affecting the velarinsert verbs of two dialects, the

Pasiega and Belsetán dialects of Spain, are analyzed in Penny (1969) and Badia i

Margarit (1950) respectively. In the Pasiega , tener has a contracted form [teu] for

1s PRS IND (Penny 1969), while in the Belsetán dialect, it shows the velar systematically

in 1s/SBJ, excluding the 1p/2p SBJ forms which are marked by [j] postnasally, i.e. 1p

SBJ teniamos , 2p SBJ teniaz (id. salir , venir ; Badia i Margarit 1950). In these dialects,

the composition of the AM differs with respect to the shape of the 1s/SBJ morphome in

that it excludes the 1p/2p SBJ forms.

Caer , creer , oír , reír , traer and ver do not make use of the velarinsert (N.B. traer and oír do so in limited areas) in the Pasiega dialect, instead being characterized by a palatal fricative in the 1s/SBJ forms (cf. discussion of the developments prior to

velarinsertion treated in §3.5.3). Additionally, each of these verbs has two separate paradigms, one where the palatal is extended to all forms of the PRS IND and one where

it is found only in the 1s/1p/2p as a means of hiatus (in the other IND forms, it is an

offglide, e.g. caer , 2s /kajs/) (Penny 1969).

186

Another interesting example is taken from Clark (1986) which provides the INF traigar and pongar from native Spanishspeaking children, indicating a shift of these

verbs to the first conjugation (ar ) and the making innovative use of the originally 1s/SBJ

allomorph. From the immanent perspective, the reanalysis of these two verbs involves paradigminternal and external analogical influence and is interesting for several reasons.

It can be argued that children produce such forms because they lack the linguistic

experience of adults. This includes both perception and production and may be indicative

of the salience of the PRS tense forms and, more to the point, on the 1s/SBJ forms. The

first conjugation is overwhelmingly more productive and frequent, in both type and

token, than the –er and –ir conjugations combined. The predominance of the 1s PRS is

logical for the egocentric speech of children while the SBJ appears in contexts of

influence, like commands. The limited amount of input and output, in conjunction with

more limited attention and memory, that children have in comparison to adults means that

the domain from which they are able to extract the necessary AM is limited to those

lexical items that are more cognitively salient at a particular point in time.

In Spanish, velarinsertion is restricted by a number of factors including, though

not limited to frequency, timing, and phonological, morphological and paradigmatic properties of the language. All of these must also be considered in relation to the

realtime salience of a particular AM in the mind of the speaker when production requires

reanalysis of a form due to imperfect memory, inhibited recall time, interference of

competing models, etc.

187

3.7—THEORETICAL CONSEQUENCES AND CONCLUSIONS . The immanent perspective required by the current conception of AEM separates it from most theoretical analyses of velarinsertion. It more elegantly handles the data in its synchronic and diachronic context as demonstrated in the preceding sections. From a broad perspective, the current investigation corroborates the same general analogical path as proposed in previous analyses, namely the extension of the velarinsert from stemfinal nasal to lateral to rhotic to palatal to sibilant, then being limited to dialectal productivity in verbs containing the stemfinal segments. On the basis of individual stages and forms, though, it becomes clear that the analogical process is not so clear cut. In other words, there is too much diachronic overlap to suggest that velarinsertion moved distinctly from verbs having one stemfinal phoneme to those having another. Instead, it is crucial to show how this one contextual variable of the AM is broadened over time to include not a specifically lateral or rhoticfinal verb stem, but rather both of them simultaneously. In doing so, the weakening of the AM becomes apparent. This accounts for the decrease in productivity over time, and the lexically diffuse, gradual, and incomplete nature of the analogical process. Additionally, the diachronic process is marked by sometimes long periods of synchronic variation of a particular form. The messiness of this analogical development means that, by nature, it tends to fall outside the realm of most theoretical linguistic models.

The current analysis of velarinsertion demonstrates how AEM is able to incorporate the analogical process into the theoretical model. It supposes that analogy is central to and processing and categorizes input through its shared

188 characteristics. For velarinsertion, these include specific phonological, morphological and paradigmatic factors considered with respect to the immanent view of language change. The input for the AM is reshaped by subsequent generations of speakers according to the examination of available input, the function of frequency and the imperfections of human nature (including memory) and environment. It also explains the lack of systematicity in language change demonstrated by the synchronic variation that affects particular forms and the diachronic changes in velarinsert class membership, including its modern dialectal extension. A consequence of the nonteleological nature of analogy allows for multiple directions of change through the influence of competing models.

The following chapter explores aspects of the preceding analysis that require further discussion, including the function of the paradigmatic template, or morphome, type and token frequency, and the possible directions of language change. Also addressed is the hypothetical basicderived relationship between forms in a paradigm.

Each of these is relevant to the velarinsertion process and, more broadly, to the analogical process at the heart of AEM.

189

CHAPTER 4: AEM IN THEORY AND PRACTICE

4.1—THEORETICAL AND PRACTICAL ISSUES . The previous chapter demonstrated the relevant synchronic stages included within the diachronic development of the velarinsert subclass in Spanish. Each ‘slice’ of the larger process of velarinsertion adds information about its diachronic development and illustrates the intermediate changes in linguistic processes, especially with respect to the broadening of the AM and subsequent extension of the velar to new analogical targets. It has been demonstrated that the analogical development of velarinsertion is gradual yet dynamic, best handled by a theoretical model that utilizes analogy as a primary means of language organization and consequently, change. This is a central aspect of AEM which constructs an AM on the basis of multiple variables shared between exemplars. The probabilisitc nature of the theory permits but does not require the influence of a model on a potential target. It also means that competing models may come into play. These factors were demonstrated in various synchronic stages of the preceding analysis. The effect of probability and competing models explain both synchronic variation and the nonsystematic, nonteleological nature of change. At the same time, AEM approaches language from an immanent stance to ensure that the speaker is actively involved in all aspects of its development.

In AEM, the relevance of the inflectional paradigm becomes readily apparent, as it creates strong multidimensional associations between its members. For

190

velarinsertion, these associations include phonological, morphological, and lexical

characteristics among others. Thus, the 1s/SBJ distribution of allomorphy in the

velarinsert class has been identified as a crucial variable of the AM that determines the productivity and stability of this lexical subcategory.

The following sections treat this and several other specific issues dealing with the

analogical process within AEM as well as a number of topics that are not only quite

common in discussion of linguistic theory but are also extremely relevant to

velarinsertion. First, analogy is reviewed in relation to abuctive reasoning. Next, the

function of the paradigm in analogy, and specifically in the AM for velarinsertion, is

examined in greater detail. Afterwards, the role of frequency is briefly explored using

examples from the velarinsertion data presented in chapter 3. Questions about

markedness and the basic/derived relationships between paradigmatic forms are also

addressed. Finally, an analysis of possible directions of language change involving

velarinsertion are examined against data from other Romance languages.

4.1.1—THE ELEGANT COMPLEXITIES OF ANALOGY . Analogy has been defined in previous research as an abductive process, i.e. a loose generalization based on specific forms and abstract generalization (cf. Sims 2006, Andersen 1973). Accordingly, velarinsertion requires a complex abductive analysis since a rulebased account is difficult to formalize because there is no systematic rule which applies synchronically or diachronically to all velarinsert forms to the exclusion of those verbs not affected by it without appealing to abstract, special underlying representations and/or minor rules. As a simplified example, consider the following explanation for the velar in 1s/SBJ forms of

191 poner : the RESULT that poner is a (monosyllabic) second conjugation verb having a

nonhigh stem vowel and stemfinal coronal nasal /n/ in 1s/SBJ marked by nonfront

inflectional vowels, /o, a/ is combined with the LAW declaring that velarinsertion occurs in 1s/SBJ forms of the second and third conjugations which are marked by nonfront inflectional vowels, /o, a/ and (monosyllabic) stems ending in a nonhigh vowel plus coronal nasal /n/ (this LAW is an inductive generalization built from the AM tengo/tenga , vengo/venga ). The combination of RESULT and LAW implies the following CASE : that the

85 1s/SBJ forms of poner are (likely) to show velarinsertion. The preceding LAW and entire abductive process are derived from a complex series of abductive, deductive and inductive inferences since each individual characteristic of the AM requires (constant) validation. The highly involved nature of these ‘perceptual judgments’ demonstrates the complexity of cognition and linguistic organization through analogy. In the same way, the innumerable variables and the frailty that are a consequence of abductive generalizations (as well as the inductive and deductive processes built upon them) explain the difficulties typically associated with the incorporation of analogy into linguistic theory.

Another factor which only complicates the matter further is the function of immanence. The realtime analysis (and validation) of information by a linguistic individual is wrought with imperfections that interfere with the entire cognitive process.

Thus, the AM is in a constant state of flux as it emerges for each particular instance of potential application (cf. Whittlesea 1987). Likewise, the influence of competing models or changes in the salience of particular variables in the AM may, in such instances, result

192

in different directions for language change, including no change whatsoever (due to the

high associative strength of an already active AM).

The broadening of the AM for velarinsertion is the result of new abductive, inductive and deductive processes. These are the result of invalidation, insecurity, competition or imperfections in the associative mechanism utilized by a speaker at a given point in time. The consequences of such complex analogical processes are demonstrated in the sychronic and diachronic variation that characterizes the velarinsert verb class as well as the highly restricted productivity of velarinsertion in potential targets throughout the history of the language.

4.2—THE COGNITIVE PARADIGM. An important component of the preceding abductive equation, not to mention of the current overall analysis, is the inflectional paradigm. By definition, it creates a strong association of form and meaning between it members (cf.

Joseph & Sims 2006, Joseph 2006, Clahsen et al. 2001, Ristad 1993, Martín Vegas 2007).

In general, the relationship between the members of a paradigm involves lexemic identity/uniformity in the shape of a shared stem. The function of the paradigm in linguistic organization is logical from the standpoint of AEM, as it is captured perfectly by the concept of the ‘exemplar cloud’ (see §3.3, cf. Bybee 1995).

The paradigm is structurally emergent from the cognitive organization of

language and is crucially involved in many analogical processes (cf. Pirrelli & Battista

2000a, Harris 1973, Malkiel 1967, Menéndez Pidal 1980, Joseph 2009). Evidence for the

function of the paradigm as a cognitive structural unit is found in Clahsen et al. (2001),

which examines several studies on language impairment. These studies show that errors

193

are bound by grammatical category and constrained by paradigmatic structure, often

within a single morphological dimension like person or number. Additionally, Sims

(2006) indicates that: (1) inflectional forms are not random but are based on related

forms, demonstrating the existence of inflectional (and derivational) classes; (2) leveling

occurs between related forms; (3) syncretism (identity between forms in a paradigm)

affects related forms; (4) there is a similarity in syntactic function across related forms;

(5) innovative forms are automatically associated with a paradigm; (6) analogical change

is able to affect paradigmatic forms to the exclusion of nonparadigmatic ones (cf. Joseph

& Sims 2006, Stump 2001, Carstairs 1983, Joseph 1997b, to appear). Notably, many of

these points have been observed in the velarinsertion process. Obviously, there is a

morphosyntactic relationship between forms in the inflectional paradigm. More

importantly, leveling affects the AM set of –NGERE verbs and several potential and

‘partial’ members of the velarinsert class. Also, velarinsertion may be considered an innovation in previously unaffected forms which is consequently associated with their particular inflectional paradigm. The most crucial point, however, is that velarinsertion is an analogical change that depends on paradigmatic structure, i.e. it excludes nonparadigmatic forms. This is central to the composition of the AM proposed here since it relies on the paradigm to restrict its potential application in the lexicon.

The paradigm in relation to its use in the AM for velarinsertion is examined further below. First, the question concerning the function of the paradigm in language is analyzed in more detail, especially as it pertains to velarinsertion. This includes its use in analogy by way of inter and intraparadigmatic effects. Next, some of the various roles assigned to the paradigm in previous research are then compared to identify

194

commonalities. Finally, the ‘morphologization’ of the 1s/SBJ template or ‘morphome’ is

exemplified by data from several Romance languages.

4.2.1—PARADIGM THEORY . There are two main problems with depending on the inflectional paradigm in explaining language change: What exactly is its function and how can it be incorporated into linguistic theory? The first question is multifaceted and is treated below. As for the second question, paradigmatic structure has long been mentioned as being relevant to linguistic organization, hence also to theoretical conceptualizations. §§1.2.2 and 2.5.1 describe recent, valuable attempts at incorporating the paradigm into linguistic models, e.g. OOFaithfulness, Paradigm Uniformity, etc.

However, such analyses do not resolve the matter of incorporating the paradigm into linguistic theory in any truly innovative and elegant way. Thus, the proposal of this investigation is that paradigmatic structure emerges as a logical function of AEM and as such it is utilized by the speaker not only as a means of organizing the lexicon, but also of analyzing new (inflectional) input.

Earlier discussion of velarinsertion demonstrated that the paradigm is a crucial variable of the AM for velarinsertion because: (1) it accounts for several of the most salient conditioning characteristics involved in this analogical process (e.g. verb, following nonfront vowel, inflectional markers, etc.); (2) it serves as a means of including and excluding exemplars (directly and indirectly) from both the AM and target data sets (e.g. nonverbs, 2s/3s/1p/2p/3p, etc.); (3) it is representative of a type of gangeffect in its templatic structure, i.e. a supracontext where all the exemplars act in a highly similar fashion, thereby increasing the likelihood that they may be selected as the

195

model for a target (see below); (4) it accounts for the relationship between 1s and SBJ

forms. Hence, the paradigm is relevant to the phonological, morphological, lexical and

syntactic aspects of the AM. Also, the forms in a paradigm are related to each other and

those of other paradigms according to associative strength which is exemplified by the

emergent distributional patterns of allomorphy, e.g. the 1s/SBJ morphome.

With regard to AEM then, the paradigm can be defined as a higher order

exemplar cluster (cf. Fondow 2007). The reason is that the paradigm is comprised of a

highdensity localized exemplar cloud/dataset, characterized by the inherent associative

strength of shared phonological, morphological, semantic and syntactic features. This is

a consequence of lexical storage according to AEM since, “[a] set of exemplars that are judged to be similar phonetically and represent the same meaning are clustered together

and are represented as a word or phrase [here, a paradigm]” (Bybee 2007:716). For an

individual verb form, it will generally be the case that the other cells in its paradigm will participate in the AM due to ‘proximity’, considering that the activation of one exemplar

in a cluster (here, one cell within a paradigm) increases the activation or readiness of all

other exemplars in that cluster (e.g. cells in a paradigm or entire paradigms themselves

functioning as exemplars) (cf. §1.3.2, Pierrehumbert 2001, Skousen 2003).

The analogical function of the paradigm is not entirely innovative as illustrated by

the following quote from Menéndez Pidal (1980):

El verbo se sujeta en general a las mismas leyes fonéticas que otra palabra cualquiera. Pero se comprende que las múltiples formas que reviste un mismo tema en la conjugación, estando nidas estrechamente entre sí por la unidad esencial de significado, no pueden dejar de influir unas sobre otras más a menudo que dos palabras extrañas en su origen; así que la fuerza de la analogía...es mucho más activa en la conjugación que en ninguna otra parte del dominio gramatical, y continuamente veremos formas que tuercen su desarrollo fonético para seguir la analogía con otras del mismo

196

paradigma conjugable. El hablante...tiende a mirar el tema como invariable, pues invariable es la idea que expresa; y si unas veces la lengua conserva intactas las alteraciones fonéticas de un tema...otras veces busca la uniformidad. (269) 86

In other words, the forms in the verbal paradigm share a stronger connection than do nonparadigmatic ones. The stems of these forms tend to be identical in phonetic shape, meaning and syntactic function, differentiated morphologically by the inflectional markers attached to them. They often work as a linguistic unit such that change may occur exclusively within or outside a paradigm and they may provide the AM for change in other paradigms or parts of the lexicon (cf. Malkiel 1967). Therefore, there is a need to describe the promotion of or resistance to change within a paradigm, both linguistically and psychologically, something which is directly derived from AEM (cf. Hippisley et al.

2004).

Even in cases where such forms are not identical, as is the case with the allomorphy that characterizes velarinsertion, there are generally stable patterns of distribution, or ‘templates’ (see §4.2.3). It is only logical then for speakers to make associations based on such patterns. For example, although the palatal ~ nonpalatal alternation is phonemic in the language (e.g. /kabal/ ~ /kabaʎ/ as in verbal cabalgar ‘ride a horse’ ~ nominal caballo ‘horse’), it does not have this same function inside the verbal paradigm, since no verb forms are distinguished intraparadigmatically by such an alternation. Instead, verbs show more salient types of alternation, e.g. /g/ ~ Ø, dipthong ~ monophthong, etc. (cf. Malkiel 1967).

Accordingly, the ‘irregularity’ associated with many inflectional paradigms is, in fact, better qualified as an ‘entropic subregularity’ (cf. Maiden 1992). ‘Entropy’ here refers to the fact that the distribution of the alternants highlights the division in the

197 paradigm and can lead to its strengthening or ‘entrenchment’. It is ‘(sub)regular’ in the

sense that although there is an alternation (traditionally an ‘irregularity’), its distribution

is nonetheless systematic (or ‘regular’) within the affected paradigms.

4.2.2—PARADIGMATIC PROCESSES. In spite of the ‘subregularities’ that may characterize a paradigm, more often than not, they are susceptible to loss/regularization through reanalysis, i.e. an analogical process known as leveling. Martín Vegas (2007) defines leveling as the elimination of morphophonological alternation through the imposition of one of the alternants in the context of the other(s) and contends that it is more frequent than the extension of an alternation. Countless examples of such reanalysis of allomorphic variation are found in the preceding chapter, including the restructuring of the velar ~ palatal alternation in the set of –NGERE verbs which provides the primary AM for velarinsertion (e.g. 1s tango >> taño ), and several cases where the

velarinsert is replaced by a nonvelar throughout the paradigm, as in verbs like herir (3p

SBJ fiergan >> hieran ), huir (3p SBJ fuygan >> huyan ), doler (3s SBJ duelga >> duela ),

etc., as well as the exceptional case of asir , which appears to be undergoing leveling in

modern usage.

The traditionally accepted view of leveling is that it is a morphologically

conditioned change which has as its goal the regularization of somehow “undesirable”

irregular information (cf. Lenfest 1978, Malkiel 1969). Such hypotheses seem too strong,

especially in sometimes contradictory situations such as those listed above when

compared against velarinsertion. However, the fact that leveling crucially depends upon

morphology provides additional support for the function of the paradigm as a cognitive

198

unit of linguistic organization. This is because the process by which inflectionally (or

derivationally) marked forms are made more similar to each other is restricted to the

domain in which such forms are related, i.e. the paradigm.

Another wellknown aspect of leveling is that it involves the loss of the less

frequent alternant in favor of the more frequent one. Hence, there is an obvious function

of frequency in many instances of leveling. However, the topic of frequency is a

complex one since the assumption herein is that leveling is the result of reanalysis

through a combination of intra and interparadigmatic analogical effects. These types of paradigmatic pressures involve both token and type frequency (see below). Before

exploring frequency in more detail, it is first necessary to intra and

interparadigmatic effects since they are intimately related to the former.

The most obvious type of analogical influence involving the paradigm is from

within, i.e. one form or set of forms within a paradigm affecting others. This is known in

the literature as an ‘intraparadigmatic’ effect (cf. Penny 2002). On the surface,

intraparadigmatic analogy is the process responsible for leveling. However, as

suggested previously, such an explanation is insufficient because in all cases, it is likely

that the innovative pattern is supported by additional, external models. In the

aforementioned case of leveling of the velar ~ palatal alternation in the set of –NGERE

verbs, for instance, it can be argued that the loss of the velar in favor of a systematic palatal stem is not solely due to the overwhelming pressure exerted by the much larger

number of palatalfinal forms but also by the very strongly established pattern across all

conjugations in favor of a single verb stem within inflected paradigms, i.e. an

‘interparadigmatic’ effect (cf. Penny 2002, Martín Vegas 2007). 87

199

Similar arguments are proposed by many investigations concerning velarinsertion. Ford (1966) and Spaulding (1971) suggest that the velarinsert targets like tener , venir , salir , etc. resist palatalization of the nasal in the 1s because of the lack

of palatal nasals in the other forms of the paradigm. While this is a strong motivation for

the lack of palatalization, and support of the role of the paradigm in linguistic processes,

it is also necessary to mention that a palatal ~ nonpalatal alternation in the inflected paradigm is not supported by any verbs in the language. 88 On the other hand, the velar ~ nonvelar alternation is already available in verbs like decir , where it arises through

regular sound change, and in its analogical extension to verbs like hacer . In fact, both of

the analyses under discussion highlight the influence of decir , hacer , etc. and the analogical influence of forms from both within and outside the targeted paradigms.

Another oftcited effect of intra and interparadigmatic pressure is that inflected verbal systems tend toward, or see as more natural, paradigmatic uniformity, i.e. the use of a single (uniform) root to which inflection is added, as opposed to paradigmatic

(sub)regularity (cf. Harris 1973, Pérez Saldanya 1995a). Consider, for example, the types of OT constraints acknowledged earlier (cf. §§1.2.2 and 2.5.1). The 1s/SBJ template that characterizes velarinsertion, as well as the paradigms of countless other verbs in Spanish

(see below), flies in the face of this tendency since these high frequency verbs show no signs of shifting towards uniformity of their stems (cntr. the increased likelihood of uniformity in lower frequency verbs like asir , cf. Phillips 1984, KraskaSzlenk 2007).

Although velarinsertion provides a counterexample, all is not lost. What is most important in this case is that paradigmatic uniformity is a tendency, not a hardandfast rule, and that it is intimately related to frequency. The probabilistic nature of AEM that

200

emerges from the dynamic nature of language use in the form of competing AM means

that linguistic theory is obliged to make use of such tendencies. If not, the grammar will

always ‘leak’, requiring exceptions and ad hoc techniques to resolve problems.

The tendency towards paradigmatic uniformity is demonstrated in Harris (1973).

It involves a shift from OSp. cuego , cuezes to leveled cuezo , cueces . From a rulebased

standpoint, this shift necessitates a mere change in rule order, from marked to unmarked.

In this way, the process is not viewed as analogical. However, it should be noted that a

change in rule order should not be considered without appealing to the use of such forms by speakers. It is not the case that speakers simply changed the order of the rules without

reason. Instead, the motivation for such change in rule order comes from both intra and

interparadigmatic influence, which coincide with (token and type) frequency.

Another example is the case of first conjugation PRS SBJ forms treated in §2.3.3.

These forms resist regular sound change due to the associative strength of their paradigms and the influence of similar patterns in other verbs (cf. Webster 1987). For

example, MSp. 1s/3s SBJ pague (not * pace ) << 1s pago, 3s paga (INF pagar ). Harris

(1973) notes that any solution will show some ad hoc or marked/rare characteristics but

all demonstrate the active presence of paradigmatic uniformity. It is true that AEM

shows the tendency in this case towards uniformity but it does not require any special

explanation since analogy within and across paradigms is the means by which this change

occurs.

A final and very crucial case of paradigmatic influence involves the yod effects

discussed in §2.3. Recall that 1s/SBJ forms of verbs like tener and venir are marked in

CL by a thematic vowel, e.g. 1s TENEŌ , VENIŌ . The front vowel later loses its vocalic

201

status, making this change a type of intraparadigmatic leveling of the phonological

shape/syllable structure, i.e. CVCVV > CVCGV, more closely resembling 2s, 3s, 3p

PRS IND CVCVC. Afterwards, the regular effect of the newly formed yod is to palatalize the preceding nasal, e.g. SENIŌRE > señor “sir”. However, tener and venir show no such palatalization in the recorded history of Spanish. Although no solid conclusion was reached earlier concerning the existence of prehistorical palatalized forms of these verbs, paradigmatic effects shed light on one possible interpretation:

In the case of the yod…the loss of the forms containing this element [prior to palatalization] can easily be explained in terms of adaptation to the other forms in the paradigm; whether or not new analogical forms containing a yod are created depends on the necessary conditions being present in a particular language, such as the existence of a group of verbs forming a distinct class to which others can be attracted. The same thing can be said of…the substitution of a velar, c, or g, for the final consonant of the stem, or of the insertion of a velar or its modern reflex between the stem and the ending…evidenced in the case of the n, l and r stems. (Wilkinson 1981:80)

In other words, the fact that there are no such palatalized forms of these verbs in the

language may be the result of intra and interparadigmatic analogy. On one hand, the

forms which were susceptible to palatalization would resist this change on the basis of the

other cells within their paradigms that had no such palatalizing yod. On the other hand,

the already existing 1s/SBJ patterns containing the velar, alongside the absence of palatal/nonpalatal stem allomorphy, supports the loss of yod without palatalization.

Another consequence in Spanish (and other Romance languages) is the effect of

leveling on first conjugation verbs against those of the second and third conjugations. In part, because the first conjugation is the most frequent and highly productive of the three

classes, it is characterized by a tendency towards uniformity in inflectional paradigms. 89

202

This contrasts with the second and third conjugations which seem to be defined by

allomorphy (cf. Martín Vegas 2007).

The first conjugation is historically distinguished from the other two by the relative invariance of the verbal root. In Italo and IberoRomance, there are no consonantal alternations in this conjugation, and [practically] none of the idiosyncratic root ablaut and consonantal alternation between the preterite and other tenses, which, in ItaloRomance in particular, characterizes most second, and many third, conjugation verbs. (Maiden 1992:291)

Although leveling is the most commonly examined instance of intra and

interparadigmatic effects, it is not the only type of change that can occur as the result of

such analogical influence. Earlier, it was indicated the velarinsertion goes against the

general tendency towards paradigm uniformity. One of the explanations, which defined

the proposed AM, is that the verbs affected are from the second and third conjugations,

which permit allomorphy more frequently than the excluded first conjugation. Another piece of the puzzle is the fact that the distribution of the stem alternants in the paradigm

is supported by already available models. The result is the entrenchment of a 1s/SBJ paradigm template that functions as a cognitive means of organizing allomorphy in the

inflectional paradigm. This is an interparadigmatic effect that is a central component to

this and several recent linguistic analyses.

4.2.3—PARADIGMATIC FUNCTION . Malkiel (1967) suggests that paradigmatic patterns

may be salient enough to provide the source for analogical change (cf. ‘constructional

iconicity’, Pérez Saldanya 1995a) . These patterns or templates, though necessarily paradigm internal, are always, or nearly always, grounded in the inflectional paradigms

of multiple verbs, giving them an interparadigmatic foundation. As demonstrated in

203

§§2.5.2.2 and 3.2.3, appeal to these patterns of allomorphy takes many forms in the

literature, each providing unique nuances of their function in language. 90 Nonetheless, what they all share is an acknowledgement of the reality of the paradigm as a method of categorization that is available to speakers, especially for analogical processing.

A crucial difference between the template/morphome stance taken here, and other investigations deals with the specific function of the velarinsert within the paradigm. 91

Pérez Saldanya (1995a), for example, indicates that the velar is a marker of depalatalization (applicable to various stemfinal elements) which increases paradigmatic uniformity, in the sense that it limits allomorphy across the verbal system (cf.

MeyerLübke 1895, Fanciullo 1998, Elson 1988). Such an analysis applies only at the earliest stages of velarinsertion when the possibility of palatalization still exists and it seems to imply a rulebased approach such that the forms having a velar are marked to prohibit palatalization from occurring. Rather than appeal to such lexical marking, the current proposal is that the distribution of the velar, originally a consequence of regular sound change in the –NGERE AM set, is structurally extended analogically to the same paradigmatic slots in highly similar verbs. On the surface, its function then is to

distinguish the 1s/SBJ forms from the remainder of the PRS tense and nothing more. The

implication is that the growing type frequency of the pattern results in its use within the

AM in the form of a template. Interestingly, this is very similar to the conclusion reached

in Pérez Saldanya (1995b) in the analysis of Catalan (cf. 2.5.2.2.3).

The relevance of the 1s/SBJ template in Catalan, as well as in Spanish

(exemplified below), has its origins in Latin (see Appendix J). This explains the vast

extension of such patterning across Romance (see Appendices F, G, H, I; cf. Pirrelli &

204

Battista 2000a, Maiden 2003, 2004, Webster 1987). The effects of changes in the vocalic

status of the theme vowel in the verbs related to Spanish velarinsertion demonstrates the

1s/SBJ template in the parent language (as seen by the shaded boxes in Appendix F as

compared against the same forms in Appendix J, for example). The hiatus between

adjacent vowels in 1s/(3p)/SBJ, a consequence of syllable structure, distinguishes them

from the rest of the PRS, e.g. CL 1s TE NE Ō vs. 2s TE NĒS . After loss of hiatus between adjacent vowels, the syllable structure changes but these forms continue to be marked through the appearance of yod, i.e. 1s */tenjo/ (CVCGV) vs. 2s IND */tεnes/ (CVCVC).

This 1s/SBJ template fortifies similar patterning in different subgroups of verbs (a type of interparadigmatic influence), e.g. inchoatives, velarinserts, etc., leading to what is often deemed its ‘morphologization’.

4.2.4—PARADIGMATIC MORPHOLOGIZATION . The establishment of the 1s/SBJ template

or morphome is an instance of morphologization because it involves the restricted

application of phonological variation within a morphological domain (cf. Martín Vegas

2007, Malkiel 1974). For velarinsertion specifically, the distribution of the stem

alternants is systematic though it does not have an exclusively phonological basis because

there are exceptions both within and outside the verbal paradigm, i.e. cases where

velarinsertion does not occur where potentially applicable. Notwithstanding, this

systematicity gives it a psychological function for the speaker that is observed in its

analogical extension to previously unaffected verbal paradigms.

The motivation for morphologization of the 1s/SBJ template is strengthened by

the other variables that characterize the AM. Bybee (1995) suggests that connections

205 between forms that are ‘relevant’ to a shared stem are stronger than those that are not.

This is a natural byproduct of the relationship between exemplars in AEM. The

velarinsert forms comprise an exemplar cloud that has relatively greater associative

strength amongst themselves than to the other members of their paradigm (cf.

Pierrehumbert 2001). The connections between all forms within the paradigm,

nonetheless, are relevant for the construction of the distributional pattern of the template.

The question of morphologization relates to suppletion. Namely, is it possible

that the velarinsert forms are derived from some external source? Such a hypothesis

seems quite unlikely. Though the allomorphic patterning is idiosyncratic, it is built upon

an already existing paradigmatic template (see below). This templatic generalization is

costly in rulebased accounts because it does not follow morphological/paradigmatic

lines, i.e. it crosses mood (IND and SBJ) and person (1s vs. remaining persons). Also,

the fact that there is an obvious generalizable pattern, one would be hard pressed to prove

that such allomorphy were suppletive since it would require the speaker to overlook a

very real (and sometimes productive) generalization that extends well beyond the case of

velarinsertion.

4.2.5—A BROADER VIEW OF THE 1S/SBJ TEMPLATE . Up to this point, discussion of the

1s/SBJ template in Spanish has revolved around velarinsertion. However, as a means of

organizing paradigmatic allomorphy, it has much greater reach (cf. Martín Vegas 2007).

Appendix L demonstrates the expansive application of the 1s/SBJ template in Spanish.

The most obvious examples include verbs like decir and hacer which have been essential

in the analysis of the AM for velarinsertion (cf. §2.3.2.3). As indicated in chapter 2,

206

these forms are not included in the set of velarinserts because they are historically

removed from the process by the etymological nature of the velar (presuming loss of yod

in CL FACERE ). Nonetheless, they show identical 1s/SBJ distribution of the velar stem

alternant in the PRS.

The verbs marked by the 1s/SBJ template do not all involve /g/. The previously

cited ‘inchoative’ class has identical distribution of a voiceless velar /k/. In this case, the

Spanish subclass includes verbs having both etymological and analogical sources for

their 1s/SBJ allomorphy. On the basis of this information, it is possible to contend that

the paradigmatic distribution of this element is also part of the AM for extension of the

‘inchoative’. This is one of the primary reasons that some earlier investigations have

argued in favor of a single analogical development involving both velarinserts and

inchoatives (cf. §2.4.2.3).

Besides the velarinsert and inchoative cases, there are also instances of allomorphy in isolated verbs which show identical paradigmatic patterning, e.g. caber and the nonvelar, antihiatic paradigms of raer and roer . It is even possible to argue that some cases of suppletion follow the same pattern, e.g. ir .92 A psycholinguistic view of paradigmatic allomorphy explains this broader functioning of the template:

The final property concerns the distribution of phonologically distinct stems over the morphosyntactic cells within a paradigm. The claim is that the patterning of these stems is not unconstrained but is determined by the morphological system in place. In other words, the distribution of phonologically unrelated stems in cases of suppletion follows the distribution of stems of more regular lexemes. (Hippisley et al. 2004:398)

Not only does this apply to suppletion, but also to the verbal subclasses under discussion

here. They follow a pattern of allomorphy based on regular sound change that over time

207

extends well beyond its etymological domain. This extension is not systematic, however,

and so depends upon additional factors like frequency, examined below.

4.2.6—HOMOPHONY . Before dissecting frequency effects in more detail, a final point

that needs to be addressed regarding the function of the paradigm, and other variables,

within AEM involves the traditionally perceived effect of homophony or, more to the point, the avoidance of homophony, in analogy. Malkiel (1974) believes that individual

lexical items with a shared lexicosyntactic function may prove influential, though

secondarily so, in the velarinsertion process:

[I]t is still permissible to search for appropriate ‘models’ involving either primary consonant clusters or differently arrivedat secondary groups. In a minor way, vengo VINDICŌ paved the road for vengo VENIŌ (indirectly, also for tengo , pongo ) – while Ptg. vingo did nothing for venho ; Sp. cuelgo 1 COLLOCŌ and OSp. cuelgo 2 COLLIGŌ blazed the trail for duelgo and tuelgo (less palpably, also for salgo , valgo ); It. colgo and scelgo drove a wedge for dolgo , salgo , sciolgo , (s)velgo , tolgo , valgo , volgo . (Malkiel 1974:345)

Similar hypotheses are also found in Alvar & Pottier (1983) and Skousen (1989) among

others. The associative relationship between homophones is also a consequence of the

analogical basis for AEM. However, it is most definitely a secondary influence, taking

more of a supporting role, in the larger analogical set which is more homogeneous with

respect to the identified variable since they share the additional elements of identical paradigmatic distribution and morphological properties. In other words, a form like Sp.

1s vengo (INF vengar ) will only be part of the AM for 1s vengo (INF venir ) if the paradigmatic distribution of the velar element within conjugational classes is ignored.

This means one of two things, either the AM does not include the paradigm as a

contributing variable, or this (supra)context is not homogeneous since the relevant

208

variables are unique for different members of the AM set. The first of these points is not

desirable as there are no cases of velarinsertion in Spanish which have violated the

established paradigmatic/morphomic pattern that is promoted in this investigation. 93

With regard to the second possibility, the homogeneous supracontext is a concept which limits the AM in order to provide some degree of measurability (in most cases). If one were to disregard this limitation on analogical influence, then we are left with something much closer to the classic, unrestricted conception of analogy. Again, the result is one which is not altogether in line with the goals of this analysis. Finally, the weakening of the restrictions on the AM lessen the cognitive relevance of the numerous distinctions observed between conjugational classes (i.e. ar vs. –er /ir as seen above). Therefore, it is necessary to consider such analogical associations as secondary in nature to the more relevant primary analogical relationships which combine all of the necessary variables and fall into the welldefined category of the homogeneous supracontext. Essentially, these secondary models are irrelevant to the influence of the AM on the target but they do provide (a sort of cognitive/subconsious) support for such a model upon its extension.

4.3—FREQUENCY EFFECTS . Before delving to the role of frequency in velarinsertion,

there are several important, though rather obvious characteristics of frequency that need

to be stated. Frequency, in the most general sense, plays a role in language change (cf.

Bybee 2000, Hooper 1976, Jurafsky 2003, KraskaSzlenk 2007, Lloyd 1987, Maiden

1992, Phillips 1999, Pierrehumbert 2001, Scarborough et al. 1977, Skousen 1995, 2002).

Such a simple statement has important consequences in AEM since input, i.e. information

having some sort of tangible frequency, is the basis for the cognitive organization of

209

language and the emergence of grammar. An implication of this is that analogy itself

relies on frequency. This inherently relates to AEM and ‘lexical strength’:

[E]very time a word or larger linguistic unit…is processed, it is mapped onto, or superimposed on, some existing representation…If meaning and phonological form of the word in processing matches a stored representation, then it is mapped onto that representation, which is consequently strengthened…high frequency words will have a greater lexical strength than low frequency words; this will account for their relatively greater ease of activation… (Bybee 1995:232; cf. Johnson 1997, Pierrehumbert 2001)

If the mapping is not exact, then partial mapping occurs, creating lexical connections,

resulting in the emergence of morphological structure, e.g. inflectional paradigms.

In addition, there are several different types of frequency one must consider in

analyzing paradigmatic forms. The primary division in the following sections is based on

type and token frequency. The former is organizational in the sense that it applies to the

number of items characterized by a particular morphophonological pattern. An example

is the frequency of paradigms marked by the 1s/SBJ template as it classifies verbs

according to the allomorphic distribution of stem alternants. Type frequency depends

upon token frequency and corresponds to increased salience or activation of a pattern

(KraskaSzlenk 2007). Token frequency involves the number of occurrences of a particular form (or, to complicate matters, lexeme) in language use. 94 High token

frequency also corresponds to increased salience, entrenchment and recency (cf. Bybee

1995, Gagarina & Gülzow 2006, Hippisley et al. 2004, Johnson 1997).

The inflectional paradigm shows both internal and external frequency effects.

Intraparadigmatic type frequency is the relative occurrence of one type of allomorphy to

another within a single paradigm. Consider the type frequency of the 1s/SBJ forms,

summing a total of seven (7) paradigmatic slots, against the five (5) remaining slots of the

210

PRS tense. Intraparadigmatic token frequency is the number of occurrences of the forms

in each individual paradigmatic slot, a true indicator of frequency in language use. This

contrasts with interparadigmatic token frequency which corresponds to the number of

occurrences of a lexeme in all inflected and uninflected forms. Finally,

interparadigmatic type frequency is the relative occurrence of a pattern of allomorphy to

others across the lexicon (e.g. the 1s/SBJ template vs. the invariant stem type).

Verb Frequency position 95 Verb Frequency position

ser 9 pasar 74 haber 12 llegar 76 estar 22 venir 78 tener 24 creer 81 ir 28 parecer 85 decir 29 deber 89 hacer 32 hablar 90 poder 34 dejar 91 ver 39 llevar 98 dar 42 llamar 106 querer 44 poner 107 saber 59 Figure 4.1—Most frequent verbs by token (Juilland & ChangRodríguez 1964).

Several other important aspects of frequency have been well documented in the literature. For example, the resistance to change in ‘irregular’ or suppletive morphology is typically attributed to frequency (e.g. Bybee 1995, Bybee 2007, Bybee & Brewer 1980,

Gagarina & Gülzow 2006, KraskaSzlenk 2007, Maiden 1992, Malkiel 1974, Martín

Vegas 2007, Phillips 1984, Skousen 1989, Webster 1987, Zuraw 2003). Consider the data in Figure 4.1. Among the 107 most frequent words by token in Juilland &

ChangRodríguez (1964), there are 8 verbs having a single stem throughout their inflectional paradigms as opposed to 15 showing some type of allomorphy. Accordingly,

211

14% of the highest frequency lexemes and, more importantly, 65% of the most frequent

verbs are characterized by stem allomorphy. In other words, frequency is not

(necessarily) equivalent to regularity. 96 This is an important aspect of linguistic organization in AEM since it essentially promotes the ‘subregularities’ found in actual language use, as opposed to complete systematicity. It highlights the localized, nonteleological nature of analogy and its dependence on the coincidence of numerous factors including not only the particular variables of the AM, but also things like frequency and lexical strength.

Since members of paradigms differ in their token frequency, they also differ in their lexical strength; the more frequent member(s) will be easier to access than the less frequent ones…If the less frequent members of a paradigm are not accessible, a speaker would have to create a new form on the basis of the one that is available, regularizing on the basis of the most frequent form of the paradigm. This would be particularly likely to happen if the whole paradigm is low in frequency. (Bybee 1995:237)

Immanence and frequency are also implicated in leveling. Leveling is a type of

‘gangeffect’, thought to occur more often in forms having low token (and type) frequency, where the contextual space involved includes both intra and interparadigmatic influence (see above; cf. Phillips 1984, Zuraw 2003). This includes forms that are unavailable to a speaker whether due to imperfect recall or lack of input

(including decreased probability of being transmitted from one generation of speakers to another). Consider, for example, the high token frequency of tener and venir against the much lower occurrence of raer , roer and asir (cf. §4.2.2). 97 While all these verbs are part

of the standardized velarinsert subclass in Modern Spanish, the latter group appears to

favor leveling in actual usage (cf. §§3.5.53.5.6). Low token frequency also provides a

212

means of explaining the large number of targets unaffected by velarinsertion (see below).

4.3.1—TYPE FREQUENCY . Type frequency is essential to both regularity and allomorphy in the inflectional paradigm. Eddington (2002) argues that type frequency is seen as a better predictor of analogical processes and productivity than token frequency. This emerges from the variables associated with the AM for velarinsertion treated earlier as they relate more to patterns that apply across the exemplar models than to individual forms. Nonetheless, it is important to keep in mind the importance of token frequency as demonstrated in the preceding paragraph.

Type frequency, exemplified by the 1s/SBJ template, relates to the cognitive organization of language. It is possible, then, to argue that there are several applications of type frequency that are relevant to velarinsertion and the inflectional paradigm as a whole. This view of type frequency already implies the categorization of the forms under discussion as verbs. The data in Figure 4.2 demonstrate the type frequency of verbs in contrast to other grammatical categories. Among the 5,000 most frequent lexemes in the dataset, verbs account for 21% of the total. While this statistic does not appear particularly relevant, its has farreaching implications concerning the composition of the

AM for velarinsertion as explained below.

Another important example of type frequency involves the PRS. The relative frequency of a random sample of Modern Spanish verb forms in Bull et al. (1947) shows that the PRS IND is the most frequent of all types and it accounts for 40% of the tokens appearing in the dataset. The PRS SBJ, meanwhile, is the 7 th most frequent verb type,

213

though it only comprises 3% of the total number of verbs. The combination of the two

tenses, then comprises nearly half of the entire dataset and so shows the potentially high

token frequency of the velarinsert in the 1s/SBJ forms.

Relatedly, the high type frequency of ‘regular’/leveled paradigms provides an

argument for reanalysis of verbs like raer , roer and asir discussed earlier. The

overwhelming majority of inflectional paradigms in Spanish utilize a single,

nonalternating stem which represents a type of morphological patterning. Among other

things, this relates to the high lexemic frequency of the productive, highly ‘regular’ first

conjugation in relation to the second and third conjugations. Figure 4.2 provides

frequency data concerning (sub) classifications of verbs. As can be seen in the raw

frequency numbers, the combined frequency of the second and third conjugations is not

even half that of the first conjugation. However, the type frequency of these

conjugational classes shows a marked distinction. The only allomorphy associated with

the –ar class is diphthongization of stressed stem vowels in the PRS of a lexical subclass,

e.g. volar ‘fly’, 1s vuelo , 2s vuelas , cntr. 1p volamos , 1s PRET volé . The other two

conjugations, meanwhile, show many more types of allomorphy, not only in the PRS, but

also across all tenses and moods. 98 Notwithstanding, there are also –er and –ir verbs with a single stem like coser and bruñir , as illustrated in Appendix A. The end result is that the ‘regular’ distribution of an inalterable or ‘iconic’ stem in the inflectional paradigm is by far the most frequent pattern or type. The high token frequency of this ‘regular’ pattern is extremely salient and active in the mind of the speaker, thereby providing an everpresent model for change (cf. Bybee 1995, Jurafsky 2003, Scarborough et al. 1977).

214

Class Raw frequency 99 % of Verbs % of data

I, -ar 753 70% 15% II, -er 150 14% 3% III, -ir 166 15% 3%

Total Verbs 1074 21%

II & III, -er /-ir 316 29% 6% Suppletive 100 5 >1% >1% Velar-insert 44 (dial. 55) 4% (5%) >1% Velar target 32 3% >1% Inchoative 43 4% >1% Figure 4.2—Frequency of verb classes (Davies 2006).

In spite of the tremendous frequency advantage of the regular pattern, the 1s/SBJ pattern is wellrepresented. According to the data in Figure 4.2, approximately 12% of verbs fall into this paradigmatic template. Within the second and third conjugations, the

1s/SBJ occurs in 28% of all lexemes. These statistics demonstrate that as specific variables are added to the equation, the salience of the 1s/SBJ pattern increases because this template affects various groups of verbs that are part of the second and third conjugations. By including only these two conjugational classes, the probability of the

1s/SBJ template affecting a particular verb is higher than if no such constraint existed.

For present purposes then, the probability of the selection of the velarinsert as the AM for a particular target is quite high, especially if all of the variables outlined in the previous chapter are considered. 101 The increase in probability directly correlates to frequency as a determinant in activating the analogical process (cf. KraskaSzlenk 2007,

Martín Vegas 2007). In the AEM, this corresponds to the ‘randomselection’ rule of usage for determining the probability of a specific outcome (Skousen 2003). The token frequency of exemplars of a particular lexeme contributes to the type frequency that

215

characterizes it. Thus, token and type frequency are embedded in the possible outcomes

within the homogeneous supracontext. These numbers then provide the means for

calculating the probability of a particular outcome. Suppose, hypothetically, that for a particular outcome, there are 12 exemplar models of velarinsert verbs against 24

nonvelar models at the same supracontextual level. This means that the nonvelar has an

80% chance of being chosen as the AM for this particular verb. 102 Although the likelihood of the nonvelar is heavily outweighed by the alternative outcome, this method correctly highlights the probabilistic nature of analogy and emphasizes the relevance of frequency for predicting language change.

Because forms with higher token frequency generally retain their identity and resist change/regularization, they may show lower type frequency relative to iconic forms. In paradigmatic terms, allomorphic variation of any sort almost always involves a lower type frequency than a paradigm showing no such variation. This is especially true for Spanish. However, the lexemes showing alternation tend to be of high token frequency. In AEM, the sheer volume of tokens could hypothetically be argued to towards type frequency. In this way, the contrast in type frequency between regular and alternating patterns is minimized. Since forms are often attracted to higher type frequencies, most often leveling or regularization, the preceding hypothesis becomes more intriguing because it partially accounts for the analogical influence of the 1s/SBJ pattern. “De este modo, la type frequency puede ser el motivo de la extensión de una alternancia y, en otros casos, la token frequency puede ser el motivo de la estabilidad de una alternancia” (Martín Vegas 2007:10). 103

216

Additional arguments for the function of frequency in analogical change come from KraskaSzlenk (2007), which examines the relationship between high type frequency and the use of a template (showing allomorphy) as part of an AM.

Additionally, the ‘distinctiveness’ of a template and a balance in the frequency of alternants helps not only to stabilize the use of a template but may also result in its extension (cf. Albright 2008, Maiden 1992). "[T]he nuances of behavior with respect to stem/pattern analogy highly correlate with various frequency criteria, such as the size of a class, nonvacuous representation of a class among the highest frequencies and…nonvacuous representation of a minor alternant at highmedium frequencies"

(KraskaSzlenk 2007:123). All of these points are relevant to velarinsertion. First, in all applicable contexts, the 1s/SBJ template is highly distinct from nonalternating paradigms and is wellestablished among high frequency verbs, as seen in Figure 4.1 above (cf. Maiden 2004). According to the data in Figure 4.2, it is among the most frequent types of stem allomorphy in verbs even though it represents an extremely reduced amount of the data. If it is considered in the context of the second and third conjugations, the numbers are much better, as standard velarinserts account for 14% of these verb classes (17% if dialectal data is included). More broadly, the percentages for the 1s/SBJ template are even more telling, as this pattern accounts for nearly 1/3 of all second and third conjugation verbs and the forms included are among the higher token frequencies. Within the PRS paradigm, there is also excellent balance between velar and nonvelar alternants (see above). On the basis of these frequency effects, the use of the template has extended over time, as demonstrated by the data in chapter 3. In the broadest sense, the 1s/SBJ pattern is found among the most frequent subclasses of verbs

217

(interparadigmatic type frequency), it is strongly represented in the highest frequency

verbs (interparadigmatic token frequency) and the balance of alternants in the PRS tense

implies high occurrence in usage (high intraparadigmatic token frequency).

4.3.2—TOKEN FREQUENCY . Several consequences of high token frequency have already been addressed. First, a form is considered more salient if it occurs more frequently. In part, this explains the stability of (often ‘irregular’) distributional patterns in high

frequency verb forms (see above). Next, the more exemplars of a particular form, the

greater its lexical strength (cf. Skousen 1989). Together, these frequency effects

constitute ‘entrenchment’ of a form in the lexicon. Additionally, a form that occurs more

often than another form will tend to be more recent (cf. Johnson 1997). As a result,

frequent forms are more accessible, more highly anticipated and can be retrieved more

quickly from memory: “High frequency words correlate with high processing speeds,

evidenced by, for example, lexical decision times, suggesting a memory effect: the more

frequently an item is accessed, the stronger the memory traces, hence the shorter the

response times” (Hippisley et al. 2004:392; cf. Gagarina & Gülzow 2006, Pierrehumbert

2002, Stemberger 1994).

These points are crucial to the survival of the 1s/SBJ template and the velarinsert

subclass. The high frequency of the verbs affected by velarinsertion contribute to their

lexical strength and the entrenchment of their allomorphic distribution. An aspect of this

that has not been considered up to this point is the token frequency of verbs derived from

the members of the velarinsert class. Each of these derivatives increases the strength of

the established 1s/SBJ pattern, especially in the forms from which they originate. The

218

rather exhaustive list of forms in Appendix B.2 should be considered in conjunction with

the examples of the inchoative found in Appendix L, as they also contribute to the

entrenchment of the paradigmatic template. Thus, the earlier findings that allomorphy or

irregularity is stable in high frequency forms and that it is generally associated with the

second and third conjugations in Spanish are a consequence of their frequency. On the

other hand, low frequency verbs are less entrenched, leading to longer processing times

and an increased probability of analogical reanalysis on the basis of a salient competing

AM.

Such is the case for verbs like raer and roer . These verbs technically belong to

the velarinsert class but standard practice also assigns them to other subclasses, e.g. 1s

roo , 2s roes , having a completely regular stem, and 1s royo , 2s roes , showing a stemfinal palatal in the 1s/SBJ. The history of roer demonstrates that its low token frequency

makes it susceptible to reanalysis from one generation to another (cf. §3.5.5). Figure 4.3 provides the token frequencies of the 3s/3p SBJ synchronic variants. It shows the

diachronic tendency among generations of speakers towards the leveled stem, as the palatal stem has fallen completely out of use and the velar has been relegated to dialectal

use (Alvar & Pottier 1983).

Stem 1450-1500 1500-1550 1550-1600 1600-1650

roy 3 14 9 3

roig 0 0 1 2

ro 1 3 6 10 Figure 4.3—Token frequency of 3s/3p SBJ synchronic variants of roer (per CORDE).

219

The frequencies of these variants relative to each other is indicative of leveling of the stem. This is supported by the extremely low frequency of the verb. In the CREA database, the INF of appears only 125 times and the velar and palatal stems are not attested, in contrast to 6 examples of the vowelfinal stem (Real Academia Española,

2007). Similar numbers are found for raer , for which there are 6 tokens of the INF, 2 of the palatal stem and no examples of the velar. Also, neither of these verbs is among the

5,000 most frequent lexemes per Davies (2006). Thus, it is clear that if a form is of such a low token frequency that it is not commonly used, the analogical pressure of the most frequent pattern is likely to affect it. That is not to say, however, that it is impossible for less frequent analogical types to prove influential, as demonstrated by the available variants in these examples as well as the truly extraordinary case of the extremely infrequent verb, asir (see below, cf. §3.5.6).

While there are examples of low frequency verbs being affected by nonregularizing changes, it has been suggested that forms having average token frequency are the best predictors of the direction of language change (Eddington 2002).

This is because the highest frequency set includes many irregular/suppletive forms (cf.

Hippisley et al. 2004). In contrast, those in the middle are fairly homogeneous in their patterning, giving them a sort of type frequency. The lowest frequency forms are excluded on the basis of their limited activation, especially from an immanent point of view. Given the previously indicated tendency for high frequency forms to be marked by allomorphy or irregularity, an explanation emerges for the distribution of velarinsertion across high and low frequency lexemes. Velarinsertion is more likely to occur in high frequency items precisely because speakers have some awareness of the inherent

220

‘irregularity’ (of a highly frequent type) associated with them (cf. Maiden 1992, 2004).

Low frequency forms tend to lack such variation and are usually modeled on middlefrequency exemplars, most of which are regular (see above).

An interesting example of this comes from the yodfinal stems affected by velarinsertion, e.g. caer , oír , traer (1s caigo , oigo , traigo ). These verbs are considered to be of average frequency (respectively numbers 245, 263 and 289 in terms of lexemic frequency, following Davies 2006) and are all characterized by a nonfront stem vowel, cntr. reír , ver , creer , which are affected only dialectally by velarinsertion. The frequency of caer , oír , traer also contrasts them with lower frequency verbs like atribuir

(1539), concluir (1221), disminuir (1982), that show very isolated examples of velarinsertion but are generally unaffected by it. Similarly, the extremely high frequency verbs haber (11) and ir (30), though also showing low level effects of this analogical process, are characterized by different types of allomorphy, including suppletion (e.g. 1s voy ). The fact that all of the verbs above are affected, to some degree and at some point in time, by velarinsertion signifies that such evidence is only indicative of an overall tendency and does not apply systematically in any way. 104

Verb Frequency position 105 Prefixed Verb Frequency position

tener 24 detener 397 venir 78 suponer 409 poner 107 convenir 441 salir 115 oír 130 traer 222 caer 269 valer 459 Figure 4.4—Frequency of standard velarinserts (Juilland & ChangRodríguez 1964).

221

In the exceptional instances where velarinsertion is triggered by a low frequency

verb, as in asir , AEM contends that this is due to the probabilistic nature of the cognitive process (cf. Jurafsky 2003). In other words, velarinsertion cannot be completely

discounted in any case to which it may apply. Accordingly, the ‘randomselection’ rule

described earlier explains why the synchronic results for individual verbs are due to the

influence of an AM which may or may not be statistically dominant. Notably, the preceding example is isolated in the lexicon by its conjugational class and phonological

shape, i.e. there are relatively few stemfinal sibilant verbs in the third conjugation.

Thus, the supracontext of any proposed AM for asir will be rather distant from its target.

This increases the probability of the application of competing models and explains the

synchronic variation that characterizes this verb in modern (nonprescriptive) production.

The same arguments apply to raer and roer as they, along with asir , are the only standard

velarinserts in the modern language that are not included among the most frequent

lexemes as illustrated in Juilland & ChangRodríguez (1964). In fact, the remaining

standard velarinserts (and several of their prefixed forms) are of relatively high token

frequency. Their position among the most frequent lexemes in the dataset, as well as

those of several prefixed forms (in the second column), are shown in Figure 4.4 above.

222

Stem P # Mood 1250-1300 1300-1350

sal 1s/3s SBJ 30 9 salg 1s/3s SBJ 473 109

sal 1p SBJ 2 0 salg 1p SBJ 18 17

sal 2p SBJ 6 0 salg 2p SBJ 13 12

sal 3p SBJ 8 2 salg 3p SBJ 88 53 Figure 4.5—Token frequency of synchronic variants of salir (per CORDE).

A final aspect of token frequency that is relevant to velarinsertion involves the

token frequency of variants to each other over time. The diachronic shifts in variation provide evidence for the analogical nature of the process and its gradual lexical diffusion

(cf. §§ 1.3.2 and 3.5.6). Two examples of the relationship between frequency and time

involve the lateralfinal verbs salir and valer . The former is characterized by a much

higher token frequency of the velarinsert variant from the outset. The high token

frequency of this lexeme in relation to the high token frequency of the forms included in

its AM, explain the fairly rapid extinction of the nonvelar variant and

conventionalization of the velarinsert forms, as shown in Figure 4.5 (cf. Bybee 2007).

This verb is central to the AM for valer , a verb which shows a much longer period of

variation, initially favoring the earlier nonvelar, though ultimately leading to its

reanalysis as part of the velarinsert class. Figure 4.6 demonstrates the gradualness of

this development (cf. Joseph 1997b, to appear).

223

P # 125 0- 1300 - 1350 - 1400 - 1450 - 1500 - 1550 - 1600 - Stem Mood 1300 1350 1400 1450 1500 1550 1600 1650 val 1p SBJ 0 7 0 1 0 0 0 0 valg 1p SBJ 0 0 0 2 3 7 10 17 val 2p SBJ 3 1 0 1 0 0 0 1 valg 2p SBJ 1 0 0 0 0 1 1 4 val 3p SBJ 128 84 64 53 204 135 40 3 valg 3p SBJ 2 5 9 14 54 107 104 150 Figure 4.6—Token frequency of synchronic variants of valer (per CORDE).

In spite of the numerous aspects of type and token frequency that are relevant in

language change, alone they cannot explain the retention or loss of linguistic information

(cf. Maiden 1992). In part, this is because frequency provides no insight into the

composition of the AM relevant to a particular change. Also, there are seemingly

conflicting effects of frequency as seen in the changes affecting extremely high

frequency, relatively high frequency and low frequency verbs. It has been shown that

frequency effects are tendencies that contribute to the distribution of allomorphy and the

direction of change in each localized instance. In this way, they contribute to the

immanent view of language in AEM which incorporates memory effects and probability

into the analogical process.

Essentially, input is constantly compared against existing information in the

minds of speakers. High frequency implies lexical strength and so such forms, though

more constantly evaluated, are more resistant to reanalysis. Infrequent forms are more

susceptible to loss, a consequence of imperfect memory. When low frequency forms are

used, speakers may be uncertain about their realization, specifically because they are

actively making comparisons or analogies. The choice between two competing models is

224 probabilistic and depends upon the factors that are relevant in the mind of the speaker at

the time of utterance. The decision making process may end up even (idiolectal

variation) or may change according to external (e.g. age, input, register, etc.) or internal

factors (e.g. processing time, consciousness of analysis, salience/state of activity of a particular AM, etc.). Therefore, frequency is only one piece in the much larger puzzle of

multiple causation in analogical change.

4.4—MARKEDNESS . Frequency effects often go hand in hand with markedness. The

idea that one or more forms are seen as basic within the inflectional paradigm has long been used as a means of explaining the direction of intraparadigmatic change, i.e.

leveling. In some instances, however, type and token frequency go against what is seen

as a basicderived relationship between paradigmatic forms. Some of the complications

of involving markedness in a process like velarinsertion are treated below. Also, it is

demonstrated that the basicderived relationship does not provide an elegant solution to

the 1s/SBJ allomorphy that defines a large portion of verbs in Spanish.

Bybee (1995) discusses markedness with respect to the inflectional paradigm:

“Paradigms are represented as clusters of highly connected words. The strongest words

may be thought of as bases around which other words cluster. Especially in languages

with complex morphology, there may be more than one strong form per paradigm” (242).

The suggestion that there may be multiple ‘base’ forms within a paradigm is intended to

go beyond the traditional function of markedness to account for leveling.

In a similar fashion, Hooper (1979) attempts a partial explanation of the absence

of leveling in velarinserts:

225

There seem to be two major types [of subclassification of rules with morphosyntactic conditioning]: those which govern alternations that are viewed primarily as suppletive, and are subject to leveling, and those which govern alternations that correlate with morphological categories, and are subject to analogical extension…but we do not know how to predict which type of rule will arise from a given alternation. While it is clear that there must be some reasonably strong correlations with morphological categories, this correlation need not be perfect. (125)

Part of the motivation in this instance is that the allomorphy that defines the template

associated with velarinsertion (among other types of allomorphy) crosses the

morphological boundaries that are often used to restrict or define markedness

relationships. In other words, the 1s/SBJ distribution of the velar is not easily explained by appealing to any of the traditional basicderived categorizations, e.g. 3s >> 1s, IND >>

SBJ, (or 1s >> 3s, SBJ >> IND). Even in the case where the proposed derivation crosses

mood, i.e. 1s >> SBJ or SBJ >> 1s, the markedness relationship is difficult to support. 106

Instead, it is contended herein that the use of the template/morphome as a means of cognitive organization avoids searching for a basicderived relationship between forms that share the same stem (cf. Webster 1987).

The idea that the allomorphy found in the SBJ is derived in some way from the 1s requires a complex series of markedness relationships:

We will not claim that the 1s present indicative and the present subjunctive have any semantic properties in common (besides ‘present’), but we will suggest on the basis of the theory presented in Bybee and Brewer, 1980, some reasons why the occurrence of the velar in just these categories should be stable. First, it is normal for all person/number forms of a given tense and mood category to have phonological properties in common (Hooper, 1979a). Thus the velar throughout the present subjunctive should be an optimal feature if there is going to be a velar anywhere in the subjunctive. Secondly, it is also common for the marked members of a category to be formally derivable from the unmarked members (Vennemann, 1972). So the present subjunctive should be based on the present indicative. But the whole of the present subjunctive cannot be based on the whole of the present indicative. Rather, it is based on a

226

single form of the present indicative, the most autonomous form (Bybee and Brewer, 1980). Evidence from historical change indicates that the most autonomous forms among person/number forms are the 3s and the 1s. It is also possible for the 1s form, because of its relative autonomy, to be distinct from other person/number forms in the same tense/mood. (Bybee & Pardo 1981:958; cf. Pérez Saldanya 1995a)

From the immanent standpoint, the preceding hypothesis is rather exteme in the sense that involves a multidirectional, yet apparently procedural analysis of related forms to arrive at a particular output. It also places greater value on certain forms within the paradigm essentially on account of their higher token frequency (since autonomy relates to lexical strength/salience which are derived from frequency). The question is why should the paradigm be broken down in such an almost arbitrary fashion? Is it necessary for a speaker to examine the individual relationships between forms? This method of deriving a particular form from another ignores the broader patterns that emerge from examination of the paradigm as a whole. It also requires a completely different, and possibly contradictory, series of analyses to explain other patterns of allomorphy in the same language, such as the diphthongization of stem vowels in the

PRS, or the raising of the stem vowel in some 3s/3p PRET. In paradigms where there is a single inalterable stem, the implication is that there is only one basic form, usually the 3s

IND, and that all other forms are derived from it.

The need for a limited number of basic forms comes from the rulebased tradition of positing an underlying (or basic) representation from which all other forms may be built. In cases where a single underlying form is difficult, if not impossible, to establish, the next logical step is to propose two forms, one of which is marked with respect to the other for the application or lack of application of certain rules in order to derive all the attested surface forms. Pérez Saldanya (1995a) argues that such relationships create

227

greater stability through the retention of a set of morphological idiosyncrasies based on

their systematic and recursive nature. However, the same stability emerges from AEM in

the sense that the ‘morphological idiosyncrasies’ are built into the lexicon in the form of

exemplars. These exemplars are organized according to similarity along all potentially

salient lines and so the ‘systematic’ and ‘recursive’ patterns that define a language are

consequential. Nonetheless, they are useful as a means of linguistic organization.

Maiden (1992) indicates that the historical evidence promotes the simultaneous

appearance of the velar in 1s and SBJ such that one is not easily derived from the other:

“On available evidence, there is no more reason to derive the subjunctive from the first person singular than there is to derive the latter from the former: they are mutually

implicational and both diachronically and synchronically coherent” (307). Evidence

against deriving the SBJ from the 1s comes from the appearance of the velar in forms of

the defective paradigm of haber ‘exist’ (cntr. its other semantic functions, cf. §3.5.4). In this case, each paradigm contains only one form which may be interpreted as 3s or 3p depending upon its morphosyntactic context. Hence, it is impossible to suggest that the examples of the velarinsert in the SBJ form of this verb are derived from the 1s IND within the same paradigm. Of course, the most plausible argument is that the velar involves analogical extension from its nondefective, semantically distinct paradigms, yet the absence of the 1s >> SBJ relationship holds and so goes against markedness relations.

There are additional problems with such markedness relationships. First, the fact that the 1s/SBJ forms are intraparadigmatically of a greater type frequency than the other

PRS forms should allow for reanalysis of the latter on the basis of the former. However, no such leveling in favor of the velarinsert is attested. Similarly, why are there no

228

intermediate patterns of allomorphy involving the preceding series of relationships? For

example, there no allomorphic patterns which separate the 1s IND and 1s SBJ from the

remainder of their paradigms nor cases where the 1s IND is marked with respect to the

remainder of the IND and SBJ paradigms. If the preceding relationships are relevant to

allomorphic distribution, there should be examples of at least some of these other possible permutations. An alternative explanation is suggested in Pérez Saldanya

(1995a), which assigns the velarinsert dual functions, indicating that the velar in the SBJ

opposes it to the IND and that the velar of the 1s marks it against the 2s and 3s IND.

However, this hypothesis overlooks a very crucial and salient generalization concerning

the patterning of paradigmatic allomorphy and the phonological characteristics of the

related inflectional markers. Also, assuming that the velar has a dual function, why are

there no examples of an opposition between IND and SBJ?

That the 3s is marked with respect to the 1s has been refuted above because no

examples of velarinsertion are attested in the 3s. In contrast, if there is a markedness

relationship between the 1s and 3s such that the former is derived from the latter, then

leveling of 1s/SBJ patterns should be the preferred direction for change (cf.

KraskaSzlenk 2007). Support for this relationship comes from several examples of

velarinsertion which were leveled in favor of a single, nonvelar stem throughout their paradigms (cf. ch. 3). Another example is provided in the active tendency for speakers to

level the paradigm of low frequency verbs like asir . A problem arises, however, in the

form of synchronic variation. Asir maintains its 1s/SBJ allomorphy dialectally and prescriptively in the modern language. The same also holds true dialectally for several

other verbs affected by velarinsertion. In other words, even though there is an apparent

229

tendency towards viewing the 3s IND as containing the basic stem, it provides no

explanation as to why its influence on derived forms is incomplete. It is for this reason

that previous analyses have suggested that some paradigms have multiple base forms (see

above). However, synchronic variation and conflicting evidence such as that described

above call into question the usefulness of the basicderived relationship (cf. Clahsen et al.

2001). Therefore, it is believed that markedness does not account for the paradigmatic

distribution of the velarinsert (cf. Maiden 2004).

4.5—WHERE TO ‘GO ’? ANALOGY AND DIRECTIONS OF CHANGE . The 1s/SBJ pattern of allomorphic distribution is originally a consequence of regular sound change which is extended analogically beyond its established conditioning environment, on the basis of similarity along phonological, morphological, lexical, syntactic and paradigmatic lines.

These similarities constitute the variables associated with the forms that comprise the AM for the extension of the pattern. As such, they are also indicators of potential directions for continued change. What follows is discussion of the potential targets for velarinsertion in Modern Spanish. Added support is provided from the Catalan and

Italian cases, which show numerous differences in the extension of the velarinsert.

Before examining these issues in detail, some theoretical points should be considered. First, the potential for change is built into AEM. The probabilistic nature of the analogical process, in conjunction with the immanent role of speakers, allows for the possibility of reanalysis of individual linguistic forms. This gives the appearance of a gradual change, though the nonteleological nature of AEM implies that such changes need not be regular or systematic (cf. Wang 1969, KraskaSzlenk 2007). Hence,

230

language is a dynamic system which involves realtime decision making through

comparison with previously acquired knowledge, which may take the form of competing

AM.107

Next, the composition of the AM and the requirement of the homogeneous supracontext restrict the direction of language change. The variables that are relevant in establishing identity between model and target identify which forms may potentially be affected by a particular analogical process. For example, consider the data in Appendix

B.1, all of which are marked by the variables associated with the AM for velarinsertion as specified in chapter 3. The homogeneous supracontext, meanwhile, is more telling of which forms may serve as the model for change. If a form falls outside the indicated context, it should not participate in the analogical development and so it is not representative of a direction for change in that particular instance. In this way, it is possible to define what are and what are not potential analogical pathways.

4.5.1—POTENTIAL DIRECTIONS OF CHANGE . According to the definition of the AM for

Spanish velarinsertion, the primary targets are verbs of the second and third conjugations

whose 1s/SBJ stems end in /n, l, ɾ, j, s/ (cf. §3.2.1.1). 108 Although Spanish shares a

lexical subclass marked by velarinsertion with other Romance languages, the criteria for

its extension and the forms it affects are not the same (cf. §2.2).

Wilkinson (1978) indicates that this change occurs only in those languages and

dialects where phonetic developments involving yod effects and palatalization are

identical in some way (cf. §§2.3.22.3.3). However, it is generally accepted that, in spite

of such similarities, the outcomes found across Romance do not presume a single

231

development but rather characteristics that originate in some common linguistic stage

which would lend themselves to the limited range of (sometimes identical) results in the

Romance verb systems.

What becomes readily apparent is that the analogical extension of the velar in the individual Romance languages takes its own path on the basis of the sociohistorical and linguistic situation particular to each (cf. Figure 3.1). Consider, for example, the extension of the velar to yodfinal stems in Spanish. This development is due to a series of coincidental phonological changes which resulted in similarity of the shape of the verb stem and mutations in the AM leading to its broader application. In other Romance languages which did not undergo these same (thus localized) changes, the set of verbs which is susceptible to velarinsertion is not the same. Instead, other paradigmatic analogical effects may have led to distinct developments in these languages. The following sections specify the distinct changes affecting Catalan and Italian followed by more explanation concerning the potential for change in Spanish.

4.5.2—CATALAN . A marked distinction between Catalan and the other velarinsertion processes is the expansive analogical extension of the velar in the former as the result of

the convergence of various phonological changes in a number of sections of the

inflectional paradigm (cf. §2.2.2). In this way, the velarinsert has become associated

with numerous subclasses of verbs.

In addition to third conjugation tenir and venir and and some other irregular

verbs, including estar , dir , dur , parts of the second conjugation in Catalan are marked by

the velar /k/ in the 1s and the voiced /g/ in the SBJ, as well as the IMPF SBJ, PRET,

232

PART (and IMPTV) (Badia i Margarit 1962). As in the Spanish case, the salient variable

associated with these verbs is the shape of the end of their stem, in conjunction with their

conjugational subcategorization: (1) –ldre /ndre (e.g. valdre ~ valer , vendre , pondre,

caldre ~ caler , doldre ~ doler ); (2) –ure (where –u represents Span. –b, v, e.g. beure ,

deure , escriure , moure , ploure, haure ~ haver ); (3) –ure (where –u is not representative

of Span. –b, v, e.g. caure , plaure , creure , seure , riure , cloure , coure , treure , jeure ); (4)

most verbs ending in –èixer (corresponding to the Spanish inchoative, e.g. conèixer , parèixer ).

As Catalan velarinsertion involves several conjugational paradigms not included

in Spanish (or Italian), its extension is much more widespread, in part as a result of its

high intraparadigmatic frequency. Consequently, changes in the variables associated

with the AM have resulted in its targeting of verbs having other stemfinal segments,

including vowels, e.g. 1s bec (INF beure ), 1s ric (INF riure ). Notably, dialectal examples

include stemfinal palatal laterals, e.g. 1s vullc (std. vull , INF voler ), and all rhotics, e.g.

1s morc , corrc (std. moro /mor , corr , INF morir , córrer ). In all cases, the extension of the

velar maintains certain characteristics also associated with the Spanish AM, namely it

follows the Sonority Hierarchy and does not apply after stemfinal consonant clusters.

The dialectal extension of the velar in Catalan is astounding. Its influence is even

felt in uninflected verb forms, e.g. valguer , tinguer , poguer , etc. (Moll 1952). Another

very crucial analogical extension involves the presence of the velar in first conjugation

verbs in some dialects, e.g. 1s /kantuk/ (INF cantar ), including some

more peculiar instances, such as its appearance after the inflectional marker, 1s /donuk/

(INF donar ) (Costanzo 2006; cf. Costanzo 2007, forthcoming).

233

Finally, some evidence related to the dialectal expanses of velarinsertion in

Catalan is its subsequent borrowing in neighboring languages. Saura Rami (2000) provides countless examples of velarinsertion in the dialectal Aragonese spoken in regions neighboring Cataluña, including 3s SBJ ega /aiga (MSp. haber ), baiga (MSp. ir ),

corga (MSp. correr ), creuga (MSp. creer ), fuiga (MSp. huir ), muergo /muerga (MSp.

morir ), etc.

All of these examples provide possible directions for change in Spanish, presuming that they fall within the context of a welldefined AM. It should be noted that

the Catalan case differs greatly in terms of intraparadigmatic and token frequency and so

is at an advantage with respect to the amount of input available to a speaker. Also,

several of the examples discussed here correspond to actually attested cases of

velarinsertion in Spanish, either historically or dialectally. Similar arguments are made

evident in treatment of the velarinsertion process in Italian.

4.5.3—ITALIAN . Unlike Catalan, the Italian analogical process follows that of Spanish much more closely, especially as it concerns the paradigmatic distribution of the velar in the PRS. A point of contrast between Italian and Spanish, however, is the path taken by velarinsertion in each language. In large part, this is due to the languagespecific phonological changes affecting them and, crucially, the composition of the AM for extension of the velar. An example of the phonological contrast is the extension of the velarinsert to yodfinal stems in Spanish. The sound changes affecting Italian verbs like rado , rada (MSp. raer ), rido , rida (MSp, reír ), credo , creda (MSp. creer ), which result in the presence of stemfinal consonants in these examples, mean that they do not meet

234

the phonotactic requirements for similar extension of the velarinsert since it would result

in incompatible consonant clusters (see Appendix I; cf. Wilkinson 1980). Relatedly, the

Italian AM is not generally supported by the appearance of the velar in high frequency

verbs like dire (1s dico ) and fare (1s faccio ) (MSp. decir , hacer ). 109 Instead, support comes from the large number of etymological velars which are marked by an identical pattern of allomorphy as a result of regular sound changes, e.g. 1s leggo , 2s leggi .

Another contrast of this type is the greater frequency and survival of the CL –LGERE set in Italian, e.g. 1s colgo , scelgo . The availability of these forms as models for analogical extension of the velar explains its presence not only in forms like 1s valgo , salgo , but also

1s dolgo , svelgo , etc (cf. Rohlfs 1968). These differences in the AM affecting the two languages, among others, make it apparent why they have taken unique paths in the analogical extension of the velarinsert.

Consequently, Italian also shows various dialectal extensions of the velar through replacement of rootfinal consonants that distinguish this process from that occurring in

Spanish and which show stronger resemblances to Catalan, e.g. 1s pargo (std. paro ), 1s beggo (std. bevo ) respectively. These examples show the extension of the velar to rhotic and vowelfinal stems (through loss of the stemfinal C), much like the examples in the previous section. There are also dialectal examples of the extension of the velarinsert pattern to the verbs dare (1s dago ), stare (1s stago ), and andare (1s vago ).

Maiden (2001) asserts that it is not possible in such cases to appeal to the phonological characteristics associated with the AM but rather to the extension of the paradigmatic template though to the detriment of morphological transparency. This exemplifies the

235 potential continuation of the analogical process of velarinsertion and the reinterpretation

of the relevant variables of the applicable AM.

A final point worth considering in distinguishing the velarinsertion processes of

Spanish and Italian is that Spanish characterizes allomorphy in a more salient fashion

than does Italian (Malkiel 1974). For example, allomorphy in the PRS paradigm in

Spanish often involve differences in multiple distinctive features including, in some

cases, the presence or absence of an entire phoneme, as in 1s digo , 2s dices and 1s salgo

~ 2s sales respectively. Italian, on the other hand, shows more instances of the former as

opposed to the latter, e.g. 1s dico ~ 2s dici , 1s spengo ~ 2s spegni . If this hypothesis is

true, then between Spanish and Italian, it is possible to assume a gradient view of

minimal allomorphic contrast, i.e. Spanish tends to oppose allomorphs by more

distinctive features than Italian.110

4.5.4—POTENTIAL , COMPETITION AND FREQUENCY IN SPANISH . Examination of

Appendix K shows that many of the potential targets for velarinsertion in Modern

Spanish resemble, or are identical to, those affected by the process in Catalan and/or

Italian. Otherwise, they follow the patterns established by the already available

velarinserts in Spanish. The reason is that the forms that contribute to their respective

AM share a common set of relevant variables. The variables for the Spanish case are

detailed in §3.2, though they are highly similar to those involved in the Catalan and

Italian processes. The remainder of this section attempts to describe the characteristics of

the AM for velarinsertion in Spanish that are most salient for a particular set of

analogical targets, including those which are part of the velarinsert subclass dialectally.

236

In addition, the function of frequency and the relevance of competing models are

examined in order to explain, in part, the lack of velarinsertion in many of the forms

outlined here.

§3.6 provides examples of modern dialectal instances of velarinsertion. Like these forms, the potential analogical targets listed in Appendix K share a high degree of identity with the verbs included in the AM. All of the potential targets included here are second and third conjugation verbs, implying a 1s/SBJ pattern on the basis of the nonfront inflectional vowel, having a stemfinal nasal, lateral, rhotic, glide, or sibliant. 111

The categorization of these forms by their stemfinal phoneme is indicative of the broadening of the AM over time as described in the preceding chapter.

Another contributing factor is the generally low token frequency of these targets, as shown by the very small portion of them among the most frequent lexemes in the language (see notes to Appendix K). In cases of higher token frequency, the lack of velarinsertion is due to the combined influence of their frequency and their longlasting membership in another lexically strong competing AM (cf. discussion of unir and morir ,

§3.6.2). Low token frequency can be seen as motivation against velarinsertion as

exemplified by the status of the prescriptive velarinserts asir , raer and roer (cf. §4.3.2).

Since high token frequency tends to be associated with the potential for irregularity, low

frequency severely decreases the likelihood that velarinsertion will apply to a particular

target. This is supported by the token frequency of the standard velarinserts in Figure

4.4 above, which suggests that even among the higher frequency verbs, there is a tipping point below which velarinsertion is less likely to apply and if it has, it is less likely to be

maintained or will only do so in synchronic variation with a nonvelar stem. Aside from

237

the previously mentioned lowfrequency examples, there are numerous others including

verbs like doler , parir , huir , and coser which may be considered midfrequency verbs (cf. notes to Appendix K). Although these verbs show or have shown a velarinsert form in the past, they are not part of the modern subclass or are so only dialectally.

The localized activation of the analogical process is visible in these varying levels of success in affecting potential targets. Analogy is not motivated by systematic application across all potential targets and so there are numerous cases where velarinsertion could but has not occurred, or where it has been limited historically or dialectally. Up to this point, one of the primary explanations for such limited productivity is the presence of competing AMs. In many ways, this is related to the idea of ‘selective attention’ described in Nosofsky (1986). It highlights the fact that the

‘contextdependent’ similarities between forms which draw the attention of the speaker in realtime may vary (i.e. salience). While this takes on a seemingly random nature, it is based a degree of activation of the stored exemplars. Not only does this provide insight into the nature of competing models in cognitive linguistic organization, but it also helps to explain the diachronic changes in the characteristics of the AM for velarinsertion seen in chapter 3.

Another aspect of the extension of the velarinsert to these potential targets is the relationship between token frequency and time. For low frequency verbs, velarinsertion does not apply in large part because of their lowlevel usage over time, especially during the most productive applicable stage of the analogical process. If a form does not occur frequently enough, or at all, during these stages, it is less likely to undergo velarinsertion and hence, more susceptible to regularization/leveling on the basis of a competing AM

238

(e.g. abolir and pulir , §3.6.2, cf. §4.3). The same arguments hold true for the

velarinserts showing continued variation between velar and nonvelar forms, e.g. asir

(§3.5.6), and raer , roer (§3.5.5), while also demonstrating the probabilistic nature of

AEM.

Hippisley et al. (2004) suggests that high token frequency, inflectional categorization and distribution of alternation (the latter two arguably referring to type frequency) combine to detemine the resistance to regular sound change and/or the stability of irregularity/suppletion/allomorphy. In effect, this provides an explanation for the absence, loss, or limited availability of velarinsert forms in low frequency verbs.

Their susceptibilty to analogical change that promotes allomorphy is limited by frequency. Since AEM is nonteleological, such forms are not forced to change in spite of meeting all the necessary criteria that define the AM. This is due, in some ways, to the intimate involvement of the speaker which includes the patterning and frequency of input they use to construct associations between exemplars.

Early on, for example, the verb remanir /remaner was perfectly susceptible to the analogical influence of tener , venir , poner , and possibly their analogical predecessors as well. Thus, one finds OSp. 1s/3s SBJ remanga (cf. Tuscan 1s rimango ) (Malkiel

1974:325). However, “[i]n numerous instances such candidacies lapsed, because the verb at issue either fell into desuetude, or was regularized through levelling, or else was pressed into the mould of an inchoative…” (Malkiel 1974:325). The preceding example is an interesting one because it shows that a high degree of identity with the AM is not a sufficient means of survival as remanir has generally been lost. It has, however, left behind some etymological vestiges. The regular verb manir , an ideal target for

239

velarinsertion, is still used in the modern language. Given its highly specialized

meaning, extremely low diachronic and synchronic frequency, and highly defective paradigm, it has resisted velarinsertion in favor of a competing and extremely high (type

and token) frequency model showing no paradigmatic allomorphy whatsoever. Few

speakers are likely to have experience with this verb and so it does not carry the same

lexical strength as the much more frequent members of the velarinsert class. Although

there remains a small probability that such a verb could be affected by velarinsertion, it

seems that frequency, and consequently lexical strength and salience, play a part in its

actual outcome. Additionally, under such circumstances, a competing AM comprised of

midrange frequency verbs, most of which show no such allomorphic variation, provides

the motivation for the leveled paradigms found in these potential analogical targets (cf.

the ‘robustness’ of analogy, Skousen 2002). Though this hypothesis is not without its

exceptions, these properties are implicit in the highly localized distribution of the

velarinsert throughout the lexicon.

A possible consequence of the effects of frequency and time involves the later

reintroduction of an alternation, as discussed in KraskaSzlenk (2007):

Frequency of use of a given lexeme or a particular wordform may considerably change with time. Analogical leveling may affect a particular word when it has a limited scope of use, while at a later time the word may expand its usage due to a meaning extension or generalization, or a cultural or any other unpredictable reason. Once an alternation is eliminated from an unproductive pattern, it will not reappear, even though it would be tolerated well at the present stage. (40)

Although it may be argued that velarinsertion is unproductive, some possible

counterexamples to this hypothesis have been mentioned previously. According to the

tenets of AEM, the lack of productivity of a pattern should not be viewed in such a static

240

fashion. This is because the constant, realtime processing of language implied by the

immanent view leaves open the possibility that a pattern may become productive again

according to its salience in the minds of speakers.

A final direction for change that is not included in the data for this section

involves the extension of the velar to first conjugation verbs. 112 Such an analogical development would require a tremendous restructuring of the AM to the point that many of the relevant variables associated with it would be eliminated. In such a case, the preceding phonological conditioning environment would still be applicable, as would the

requirement that such forms involve paradigmatic distribution of inflected forms, i.e. that

they be verbs. However, the 1s/SBJ template, so strongly linked to the distribution of the

velarinsert, among other patterns, would no longer be relevant. This would necessitate

reanalysis of the AM as not being characterized by any existing following phonological

criteria such as the alternation between front and nonfront inflectional vowels that

defines the 1s/SBJ pattern. Similar hypotheses may be constructed with respect to its

extension to nonverbs.

In conclusion, the paths for analogical extension of the velarinsert are limited by

the variables characteristic of the AM. For the most part, the potential targets share a

high degree of identity with those forms that have been affected by velarinsertion

throughout the history of the language. In extreme cases, the likelihood of velarinsertion

requires changes in the definition of the AM. The probabilistic nature of AEM accounts

not only for the activation of velarinsertion but also its variable effect in localized

contexts and its lack of activation in other cases. Generally, these differing degrees of

241

velarinsertion involve both timing and frequency effects in conjunction with the salience

of various AMs in the continuous analysis of information on the part of the speaker.

4.6—CONCLUSIONS . The variation and lack of systematicity that are visible in the

synchronic stages and diachronic development of the velarinsert class in Spanish are

indicative of an analogical process. Instead of ignoring the subregularities of this lexical

class or handling them in some complicated way, AEM handles them in an elegant and

accessible manner by incorporating the emergence of cognitive linguistic structure

through analogy based on actual, concrete input. Language acquisition and change are

analogical, in that they arise from associations between input and stored information in

the form of exemplars.

Given its analogical nature, Spanish velarinsertion involves the coincidental confluence and contribution of a number of phonological, morphological, paradigmatic, lexical and historical factors. Among the multiple variables defining the relevant AM, the paradigm is considered a crucial unit of organization, itself made up of exemplars that tend to share features from multiple levels of linguistic analysis. In addition, the paradigm provides a means for both internal (intraparadigmatic) and external

(interparadigmatic) analogical influence in language on the basis of its categorical nature. It has been shown that the patterning of allomorphy is salient in the Romance verb system as a result of a shared history though independently utilized in each language. The 1s/SBJ template associated with Spanish velarinsertion, among other processes, is a consequence of phonological and morphological characteristics associated with the second and third conjugational classes.

242

Through analysis of the synchronic stages which show productivity of the velarinsert, it is possible to observe diachronic changes in the composition of the AM, as well as in the members of the velarinsert class. The analysis of the affected verbs according to first attested appearance of the velar demonstrates how the generic explanation of the analogical process moving from nasal to lateral to rhotic to yod to sibilant paints too broad a diachronic picture. Such a view ignores the relevant details regarding the everchanging definition of the AM. In this way, AEM, and its immanent view of language, explain the synchronic variation, diachronic changes and limited or localized productivity of the process. The ‘immanent perspective’, i.e. the relevance of the “nonideal” speaker in language change (through analogy) anticipates less than systematic changes and the involvement of probabilistic decisionmaking.

It is hoped that the analysis provided herein contributes to the discussion of velarinsertion in Spanish and proves insightful in defining the process of analogy in a more specific, robust and detailed manner. Rather than carelessly appealing to analogy as a means of explaining exceptional or outlying linguistic developments, this cognitive mechanism has instead been placed at the forefront of language change, an idea which is not only logical, but has its basis in the immanent, psychological processing of information.

243

APPENDIX A: MODERN SPANISH REGULAR CONJUGATIONAL CLASSES (P RESENT TENSE )

MSp Class INF Gloss IND SBJ # P #

I (1) cenar dine ceno cene 1 1s cenas cenes 2 2s cena cene 3 3s cenamos cenemos 4 1p cenáis cenéis 5 2p cenan cenen 6 3p

II (2) coser sew coso cosa 1 1s coses cosas 2 2s cose cosa 3 3s cosemos cosamos 4 1p coséis cosáis 5 2p cosen cosan 6 3p

III (3) bruñir polish bruño bruña 1 1s bruñes bruñas 2 2s bruñe bruña 3 3s bruñimos bruñamos 4 1p bruñís bruñáis 5 2p bruñen bruñan 6 3p

244

APPENDIX B.1: MODERN SPANISH (N ON -ETYMOLOGICAL ) VELAR -INSERT VERB 113 CLASS

MSp Gloss Notes

asir grasp Velarinsert per RAE, not generally productive caer fall oír hear poner put, place raer scrape One of two accepted variants per RAE roer gnaw, corrode One of three accepted variants per RAE salir leave tener have, possess traer bring valer be worth venir come

bullir boil Nonstandard, dialectally attested concluir conclude Nonstandard, dialectally attested coser sew Nonstandard, dialectally attested creer believe Nonstandard, dialectally attested dar give Nonstandard, dialectally attested doler ache, hurt Nonstandard, dialectally attested haber have (aux) Nonstandard, dialectally attested huir flee Nonstandard, dialectally attested ir go Nonstandard, dialectally attested moler grind Nonstandard, dialectally attested parir give birth Nonstandard, dialectally attested reír laugh Nonstandard, dialectally attested soler be accustomed to Nonstandard, dialectally attested ver see Nonstandard, dialectally attested zambullir plunge (into water) Nonstandard, dialectally attested

245

APPENDIX B.2: MODERN SPANISH VELAR -INSERTS DERIVED BY PREFIXATION

MSp INF Gloss mantener sustain; maintain manutener uphold (legal) abstener abstain obtener obtain abstraer abstract oponer oppose advenir come posponer postpone anteponer place in front of predisponer predispose aponer index (linguistic) preponer put before dial. aprevenir prepare; prevent presuponer presuppose asalir meet prevaler prevail atener abide prevenir prepare; prevent atraer attract proponer propose avenir reconcile provenir arise from componer form, make up recaer fall back contener contain recomponer recompose; repair contraer contract reconvenir reproach contraponer oppose reponer replace contravenir infringe resalir protrude convenir arrange retener hold, detain corroer corrode retraer bring back decaer decline retrotraer predate deponer lay down revenir return desasir release sobreponer superimpose desatraer separate sobresalir project, stick out desavenir cause to break up sobrevenir happen, occur descaer decline sostener support, hold up descomponer decompose steal; remove; substraer desconvenir disagree subtract desimponer remove a condition subvenir help; defray desoír ignore superponer super(im)pose detener detain supervenir happen, occur detraer detract suponer suppose devenir happen, occur steal; remove; sustraer disconvenir disagree subtract disponer dispose transponer move (behind) distraer distract trasoír mishear entreoír hear vaguely trasponer move (behind) entretener entertain yuxtaponer juxtapose equivaler be equal exponer exhibit extraer extract imponer impose indisponer upset interponer interpose intervenir intervene maltraer mistreat

246

APPENDIX C: PRESENT PARADIGMS OF STANDARD MODERN SPANISH VELAR -INSERT VERBS

MSp INF Gloss IND SBJ # P #

asir grasp asgo asga 1 1s ases asgas 2 2s ase asga 3 3s asimos asgamos 4 1p asís asgáis 5 2p asen asgan 6 3p

caer fall caigo caiga 1 1s caes caigas 2 2s cae caiga 3 3s caemos caigamos 4 1p caéis caigáis 5 2p caen caigan 6 3p

oír hear oigo oiga 1 1s oyes oigas 2 2s oye oiga 3 3s oímos oigamos 4 1p oís oigáis 5 2p oyen oigan 6 3p

poner put, place pongo ponga 1 1s pones pongas 2 2s pone ponga 3 3s ponemos pongamos 4 1p ponéis pongáis 5 2p ponen pongan 6 3p

raer scrape raigo raiga 1 1s raes raigas 2 2s rae raiga 3 3s raemos raigamos 4 1p raéis raigáis 5 2p raen raigan 6 3p

247 roer gnaw, corrode roigo roiga 1 1s roes roigas 2 2s roe roiga 3 3s roemos roigamos 4 1p roéis roigáis 5 2p roen roigan 6 3p salir leave salgo salga 1 1s sales salgas 2 2s sale salga 3 3s salimos salgamos 4 1p salís salgáis 5 2p salen salgan 6 3p tener have, possess tengo tenga 1 1s tienes tengas 2 2s tiene tenga 3 3s tenemos tengamos 4 1p tenéis tengáis 5 2p tienen tengan 6 3p traer bring traigo traiga 1 1s traes traigas 2 2s trae traiga 3 3s traemos traigamos 4 1p traéis traigáis 5 2p traen traigan 6 3p valer be worth valgo valga 1 1s vales valgas 2 2s vale valga 3 3s valemos valgamos 4 1p valéis valgáis 5 2p valen valgan 6 3p venir come vengo venga 1 1s vienes vengas 2 2s viene venga 3 3s venimos vengamos 4 1p venís vengáis 5 2p vienen vengan 6 3p

248

114 APPENDIX D: ATTESTED FORMS OF REMANIR

Sp INF Gloss IND SBJ # P #

remanir remain remanga 1 1s 2 2s remangat 3 3s 4 1p 5 2p remangan 6 3p

249

APPENDIX E: VERBS WITH ATTESTED VELAR IN SYNCHRONIC VARIATION

Sp. Verb Gloss Example

atribuir attribute atribuigo bullir boil buigas concluir conclude concluiga coser sew cosga creer believe creigo dar give daiga destruir destroy destruiga disminuir diminish disminuygan doler ache, hurt duelga excluir exclude excluigo extruir extrude estruyga haber have (AUX) aiga herir wound fiergan huir flee fuygan instituir institute instituigo instruir instruct instruigan ir go vaiga moler grind muelga parir give birth parga reír laugh reiga soler be accustomed to suelgo sustituir substitute sustituiga toller deprive tuelgan ver see veigas zambullir plunge (into water) zambuiga

250

APPENDIX F: ROMANCE PATTERNS CORRESPONDING TO SPANISH TENER , VENIR

Lg INF IND SBJ # P #

MRom a ine in / ti ṷ in 1 1s a inea ii ii 2 2s ine ină 3 3s inem inem 4 1p inei inei 5 2p in ină 6 3p

MRom a veni vin / vi ṷ vin 1 1s vii vii 2 2s vine vină 3 3s venim venim 4 1p venii venii 5 2p vin vină 6 3p

MEPtg ter tenho tenha 1 1s tens tenhas 2 2s tem tenha 3 3s temos tenhamos 4 1p tendes tenhais 5 2p têm tenham 6 3p

MEPtg vir venho venha 1 1s vens venhas 2 2s vem venha 3 3s vimos venhamos 4 1p vindes venhais 5 2p vêm venham 6 3p

OFr tieng / ting tiegne 1 1s tiens tiegnes 2 2s tient tiegne 3 3s tenons tiegniens 4 1p tenez tiegniez 5 2p tienent tiegnent 6 3p

MFr tenir tiens tienne 1 1s

251

tiens tiennes 2 2s tient tienne 3 3s tenons tenions 4 1p tenez teniez 5 2p tiennent tiennent 6 3p

OFr vieng / ving 1 1s viens 2 2s vient 3 3s 4 1p 5 2p 6 3p

MFr venir viens vienne 1 1s viens viennes 2 2s vient vienne 3 3s venons venions 4 1p venez veniez 5 2p viennent viennent 6 3p

MOcPrG tener teni / tenc tengui / tenga 1 1s tenes tengas 2 2s ten tenga 3 3s tenèm tengam 4 1p tenètz tengatz 5 2p tenon tengan 6 3p

MOcPrG venir vèni / venc / venh vengui / venga 1 1s vènes vengas 2 2s vèn venga 3 3s venen / venèm vengam 4 1p venès / venètz vengatz 5 2p vènon vengan 6 3p

MCat tenir tinc tingui 1 1s tens tinguis 2 2s té tingui 3 3s tenim tinguem 4 1p teniu tingueu 5 2p tenen tinguin 6 3p

MCat venir vinc vingui 1 1s véns vinguis 2 2s ve vingui 3 3s venim vinguem 4 1p veniu vingueu 5 2p 252

vénen vinguin 6 3p

MIt tenere tengo / tegno (dial.) tenga / tegna (dial.) 1 1s tieni tenga / tegna (dial.) 2 2s tiene tenga / tegna (dial.) 3 3s teniamo teniamo / tegniamo (dial.) 4 1p tenete teniate / tegniate (dial.) 5 2p tengono tengano / tegnano (dial.) 6 3p

MIt venire vengo / vegno (dial.) venga / vegna (dial.) 1 1s vieni venga / venga (dial.) 2 2s viene venga / vegna (dial.) 3 3s veniamo veniamo / vegniamo (dial.) 4 1p venite veniate / vegniate (dial.) 5 2p vengono vengano / vegnano (dial.) 6 3p

253

APPENDIX G: MODERN EUROPEAN PORTUGUESE VERBS CORRESPONDING TO STANDARD MODERN SPANISH VELAR -INSERTS

MEPtg INF Gloss IND SBJ # P #

asir grasp aso asa 1 1s ases asas 2 2s ase asa 3 3s asimos asamos 4 1p asis asais 5 2p asem asam 6 3p

cair fall caio caia 1 1s cais caias 2 2s cai caia 3 3s caímos caiamos 4 1p caís caiais 5 2p caem caiam 6 3p

ouvir hear ouço (oiço) ouça (oiça) 1 1s ouves ouças (oiças) 2 2s ouve ouça (oiça) 3 3s ouvimos ouçamos (oiçamos) 4 1p ouvis ouçais (oiçais) 5 2p ouvem ouçam (oiçam) 6 3p

pôr put, place ponho ponha 1 1s pőes ponhas 2 2s pőe ponha 3 3s pomos ponhamos 4 1p pondes ponhais 5 2p pőem ponham 6 3p

raer scrape 1 1s 2 2s rái 3 3s 4 1p 5 2p 6 3p

254 roer gnaw, corrode 1 1s 2 2s 3 3s 4 1p 5 2p 6 3p sair leave saio saia 1 1s sais saias 2 2s sai saia 3 3s saímos saiamos 4 1p saís saiais 5 2p saem saiam 6 3p ter have, possess tenho tenha 1 1s tens tenhas 2 2s tem tenha 3 3s temos tenhamos 4 1p tendes tenhais 5 2p têm tenham 6 3p trazer bring trago traga 1 1s trazes tragas 2 2s traz traga 3 3s trazemos tragamos 4 1p trazeis tragais 5 2p trazem tragam 6 3p valer be worth valho valha 1 1s vales valhas 2 2s vale valha 3 3s valemos valhamos 4 1p valeis valhais 5 2p valem valham 6 3p vir come venho venha 1 1s vens venhas 2 2s vem venha 3 3s vimos venhamos 4 1p vindes venhais 5 2p vêm venham 6 3p

255

APPENDIX H: STANDARD MODERN CATALAN VERBS CORRESPONDING TO STANDARD MODERN SPANISH VELAR -INSERTS

MCat INF Gloss IND SBJ # P #

caure fall, lower caic caigui 1 1s caus caiguis 2 2s cau caigui 3 3s caiem caiguem 4 1p caieu caigueu 5 2p cauen caiguin 6 3p

oir hear oeixo oeixi 1 1s oeixes oeixis 2 2s oeix oeixi 3 3s oïm oïm 4 1p oïu oïu 5 2p oeixen oeixin 6 3p

pondre lay (an egg) ponc pongui 1 1s pons ponguis 2 2s pon pongui 3 3s ponem ponguem 4 1p poneu pongueu 5 2p ponen ponguin 6 3p

raure scrape rac ragui 1 1s raus raguis 2 2s rau ragui 3 3s raem raguem 4 1p raeu ragueu 5 2p rauen raguin 6 3p

tenir have, possess tinc tingui 1 1s tens tinguis 2 2s té tingui 3 3s tenim tinguem 4 1p teniu tingueu 5 2p tenen tinguin 6 3p

256 traure bring trac tragui 1 1s traus traguis 2 2s trau tragui 3 3s traiem traguem 4 1p traieu tragueu 5 2p trauen traguin 6 3p valer be worth valc valgui 1 1s vals valguis 2 2s val valgui 3 3s valem valguem 4 1p valeu valgueu 5 2p valen valguin 6 3p venir come vinc vingui 1 1s véns vinguis 2 2s ve vingui 3 3s venim vinguem 4 1p veniu vingueu 5 2p vénen vinguin 6 3p

257

APPENDIX I—STANDARD MODERN ITALIAN VERBS CORRESPONDING TO STANDARD MODERN SPANISH VELAR -INSERTS

MIt INF Gloss IND SBJ # P #

cadere fall cado cada 1 1s cadi cada 2 2s cade cada 3 3s cadiamo cadiamo 4 1p cadete cadiate 5 2p cadono cadano 6 3p

udire hear odo oda 1 1s odi oda 2 2s ode oda 3 3s udiamo udiamo 4 1p udite udiate 5 2p odono odano 6 3p

porre put, place pongo ponga 1 1s poni ponga 2 2s pone ponga 3 3s poniamo poniamo 4 1p ponete poniate 5 2p pongono pongano 6 3p

radere shave rado rada 1 1s radi rada 2 2s rade rada 3 3s radiamo radiamo 4 1p radete radiate 5 2p radono radano 6 3p

rodere gnaw rodo roda 1 1s rodi roda 2 2s rode roda 3 3s rodiamo rodiamo 4 1p rodete rodiate 5 2p rodono rodano 6 3p

258 salire go up salgo salga 1 1s sali salga 2 2s sale salga 3 3s saliamo saliamo 4 1p salite saliate 5 2p salgono salgano 6 3p tenere keep, hold tengo tenga 1 1s tieni tenga 2 2s tiene tenga 3 3s teniamo teniamo 4 1p tenete teniate 5 2p tengono tengano 6 3p trarre draw, pull traggo tragga 1 1s trai tragga 2 2s trae tragga 3 3s traiamo traiamo 4 1p traete traiate 5 2p traggono traggano 6 3p valere be worth valgo valga 1 1s vali valga 2 2s vale valga 3 3s valiamo valiamo 4 1p valete valiate 5 2p valgono valgano 6 3p venire come vengo venga 1 1s vieni venga 2 2s viene venga 3 3s veniamo veniamo 4 1p venite veniate 5 2p vengono vengano 6 3p

259

APPENDIX J—CLASSICAL LATIN REGULAR CONJUGATIONAL CLASSES (P RESENT TENSE )

CL Class INF Gloss IND SBJ # P #

I (1) CĒNĀRE dine CĒNŌ CĒNEM 1 1s CĒNĀS CĒNĒS 2 2s CĒNAT CĒNET 3 3s CĒNĀMUS CĒNĒMUS 4 1p CĒNĀTIS CĒNĒTIS 5 2p CĒNANT CĒNENT 6 3p

II (2) TENĒRE hold, keep TENEŌ TENEAM 1 1s TENĒS TENEĀS 2 2s TENET TENEAT 3 3s TENĒMUS TENEĀMUS 4 1p TENĒTIS TENEĀTIS 5 2p TENENT TENEANT 6 3p

III (3) DĪCERE say, tell DĪCŌ DĪCAM 1 1s DĪCIS DĪCĀS 2 2s DĪCIT DĪCAT 3 3s DĪCIMUS DĪCĀMUS 4 1p DĪCITIS DĪCĀTIS 5 2p DĪCUNT DĪCANT 6 3p

IIIi (3i stem subclass) FACERE do, make FACIŌ FACIAM 1 1s FACIS FACIĀS 2 2s FACIT FACIAT 3 3s FACIMUS FACIĀMUS 4 1p FACITIS FACIĀTIS 5 2p FACIUNT FACIANT 6 3p

IV (4) VENĪRE come VENIŌ VENIAM 1 1s VENĪS VENIĀS 2 2s VENIT VENIAT 3 3s VENĪMUS VENIĀMUS 4 1p VENĪTIS VENIĀTIS 5 2p VENIUNT VENIANT 6 3p

260

APPENDIX K—SOME POTENTIAL TARGETS FOR VELAR -INSERTION IN MODERN SPANISH

Conjugation INF Gloss Notes 115

3 abolir abolish defective 3 adherir adhere D # 4890 3 adquirir acquire J # 530 3 adurir burn uncommon 3 argüir deduce; argue 3 asaborir savor; flavor uncommon 3 aterir stiffen (in cold) defective 3 atribuir attribute attested; D # 1539; J # 1456 3 banir turn in (a criminal) anticuated 2 caler be necessary uncommon 2 cerner sift 3 cernir sift 3 circuir surround 3 colorir color defective 2 compeler compel 3 compelir compel 3 concernir concern 3 concluir conclude attested; D # 1221; J # 1925 3 conquerir conquer anticuated 3 constituir constitute J # 495 3 construir construct D # 689; J # 838 3 contribuir contribute D # 1183; J # 1461 2 coser sew attested; D # 3799; J # 2678 2 creer believe attested; dialectal; D # 91; J # 81 3 cusir stitch (poorly) colloquial 1 dar give dialectal; D # 39; J # 42 3 definir define D # 974; J # 1332 3 derruir demolish 3 desleír disolve 3 destituir dismiss (from office) 3 destruir destroy attested; D # 1207; J # 2030 3 digerir digest 3 diluir dilute D # 4029 3 discernir discern 3 disminuir diminish attested; D # 1982; J # 1659 3 distribuir distribute D # 1503 2 doler ache, hurt attested; dialectal; D # 2037; J # 2033 3 enfusir stuff (sausage) 3 engreír make vain

261

3 enserir insert (in text) 3 entregerir mix uncommon 2 esleer elect uncommon 3 esleír elect uncommon 3 estatuir establish 3 excluir exclude attested; D # 3121 2 expeler expel 3 ferir wound attested; anticuated 3 finir finish 3 fluir flow D # 3350; J # 4962 3 freír fry 3 fruir enjoy 3 fuir flee uncommon 3 gruir call (of a crane) 3 guarir heal anticuated 3 haber exist defective; attested; dialectal; D # 11; J # 12 3 herir wound attested; D # 3109; J # 2709 3 huir flee attested; dialectal; D # 1574; J # 725 3 imbuir imbue 2 impeler impel 3 incluir include D # 720; J # 1840 3 influir influence D # 2022; J # 1473 3 ingerir ingest 3 injerir insert 3 inmiscuir mix 3 inquirir inquire 3 inserir insert 3 instituir institute attested 3 instruir instruct attested; D # 4918 3 intuir intuit D # 4638 3 ir go suppletive; dialectal; D # 30; J # 28 3 irruir invade 2 leer read D # 244; J # 232 3 luir redeem; rub, shine uncommon 3 manir remain; age (meat) anticuated; defective 2 moler grind attested; dialectal 3 morir die D # 293; J # 257 3 muir milk 3 obstruir obstruct 3 ocluir occlude 2 oler smell D # 3036; J # 1902 3 parir give birth D # 3602; J # 4812 2 peer fart 2 poseer possess J # 515 3 preferir prefer D # 541; J # 991 3 prever foresee J # 4212 3 prostituir prostitute 2 proveer provide J # 4515 3 pulir polish 3 punir punish 2 querer want D # 57; J # 44 262

3 reconstruir reconstruct D # 2328; J # 3960 3 recluir seclude 3 referir refer J #322 3 reír laugh attested; dialectal; D # 1493; J # 529 3 remanir remain attested; anticuated 2 repeler repel 3 requerir require D # 1081; J # 1858 3 restituir restore 3 retribuir compensate 3 reunir reunite J # 856 2 seer be; be seated anticuated 3 sepelir bury uncommon 2 ser be suppletive; D # 8; J # 9 2 soler be accustomed to attested; dialectal; D # 487; J #358 3 sonreír smile J # 1194 3 substituir substitute J # 1739 3 sustituir substitute attested; D # 1786 3 sugerir suggest D # 1327 2 toser cough 3 tribuir attribute uncommon 3 unir unite D # 918; J # 1067 2 ver see dialectal; D # 37; J # 39 3 zambullir plunge (into water) dialectal

263

APPENDIX L—1S/SBJ TEMPLATE IN STANDARD MODERN SPANISH

Type INF Gloss IND SBJ # P #

Etymological hacer do, make hago haga 1 1s haces hagas 2 2s hace haga 3 3s hacemos hagamos 4 1p hacéis hagáis 5 2p hacen hagan 6 3p

decir and derived forms, afacer, contrafacer, refacer, arefacer, rarefacer, licuefacer, grandifacer, malfacer, ienfacer, desfacer Other examples satisfacer, yacer and derived forms, restructured forms, e.g. – NGERE , cocer , esparzer , erzer , etc.

Velar-insert poner put, place pongo ponga 1 1s pones pongas 2 2s pone ponga 3 3s ponemos pongamos 4 1p ponéis pongáis 5 2p ponen pongan 6 3p

asir, caer, oír, raer, roer, salir, tener, traer, valer, venir , Other examples derived forms (Appendix B.2), dialectal forms (Appendix B.1); synchronic variants (Appendix E)

Inchoative nacer be born nazco nazca 1 1s naces nazcas 2 2s nace nazca 3 3s nacemos nazcamos 4 1p nacéis nazcáis 5 2p nacen nazcan 6 3p

abastecer, ablandecer, aborrecer, abravecer, adolecer, adonecer, adormecer, afeblecerse, agradecer, aloquecerse, altivecer, amanecer, amarecer, amarillecer, amodorrecer, amorecer, amortecer, anochecer, apetecer, arbolecer, Other examples arborecer, aridecer, aterecer, avanecerse, blanquecer, calecer, carecer, clarecer, colucuecer, condolecerse, conocer, convalecer, crecer, defenecer, desbravecer, descaecer, desflaquecer, desmorecerse, desvanecer, embarbecer,

264

embarnecer, embastecer, embebecer, embellaquecerse, embellecer, embermejecer, emblandecer, embobecer, embosquecer, embravecer, embrutecer, empecer, empequeñecer, emplebeyecer, emplumecer, empobrecer, empoltronecerse, emputecer, enaltecer, enamarillecer, enardecer, encabellecerse, encallecer, encandecer, encanecer, encarnecer, enceguecer, enloquecer, encorecer, encrudecer, encruelecer, endentecer, endurecer, enfierecerse, enflaquecer, enfranquecer, enfurecer, engrandecer, engravecer, engrumecerse, enloquecer, enlustrecer, enllentecer, enmagrecer, enmalecerse, enmarillecer, enmollecer, enmudecer, enmugrecer, ennudecer, enorgullecer, enralecer, enrarecer, enrigidecer, enriquecer, enrojecer, enronquecer, enrudecer, enruinecer, ensandecer, ensarnecer, ensilvecerse, ensorberbecer, ensombrecer, ensordecer, entenebrecer, enternecer, entestecer, entigrecerse, entontecer, entorpecer, entristecer, entumecer, envaguecer, envanecer, envejecer, envilecer, enzurdecer, escaecer, escalfecerse, escarnecer, establecer, estremecer, evanecer, evanescer, excandecer, fallecer, favorecer, fenecer, florecer, fortalecer, fosforecer, fosforescer, frutecer, guarecer, guarnecer, herbecer, hermanecer, humedecer, juvenecer, lagrimacer, languidecer, lentecer, lividecer, lobreguecer, lozanecer, lucir, merecer, mohecer, nacer, negrecer, noblecer, obedecer, obscurecer, ofrecer, onecer, oscurecer, pacer, padecer, palidecer, parecer, perecer, permanecer, pertenecer, pimpollecer, placer, plastecer, podrecer, prevalecer, pubescer, reblandecer, recrudecer, remanecer, remostecerse, resplandecer, retoñecer, revejecer, robustecer, tallecer, tardecer, terrecer, tullecer, verdecer, derived forms and OSp. forms. ducir: aducir, conducir, deducir, educir, inducir, introducir, producir, reconducir, reducir, reproducir, retraducir, seducir, traducir.

Anti-hiatic yod raer scrape rayo raya 1 1s raes rayas 2 2s rae raya 3 3s raemos rayamos 4 1p raéis rayáis 5 2p raen rayan 6 3p

Other examples roer

Suppletive ir go voy vaya 1 1s vas vayas 2 2s va vaya 3 3s vamos vayamos 4 1p vais vayáis 5 2p van vayan 6 3p 265

Suppletive 116 caber fit quepo quepa 1 1s cabes quepas 2 2s cabe quepa 3 3s cabemos quepamos 4 1p cabéis quepáis 5 2p caben quepan 6 3p

Other examples 117 saber

266

Bibliography

e,i ALARCOS LLORACH, EMILIO . 1954. Resultados de G en la Península. Archivum, 4.33042.

ALBRIGHT, ADAM . 2008. Inflectional paradigms have bases too: Arguments from . Inflectional Identity, ed. by Asaf Bachrach and Nevins, 271312. : Oxford.

ALLEN, ANDREW S. 1988. The Lexical Derivational and Inflectional Continuum in Verb Morphology. Georgetown University Round Table on Languages and Linguistics 1988, 17. , D.C.: Georgetown University Press.

ALVAR, MANUEL and P OTTIER, BERNARD . 1983. Morfología histórica del español. Madrid: Gredos.

ANDERSEN, HENNING . 1973. Abductive and deductive change. Language, 49.76593.

ANTTILA, RAIMO . 2005. Analogy: The Warp and Woof of Cognition. The Handbook of Historical Lingusitics, ed. by Brian Joseph and Richard Janda, 42540. Malden: Blackwell Publishing.

ARBIB, MICHAEL A. and H ILL, JANE C. 1988. Language Acquisition: Schemas Replace . Explaining Language Universals, ed. by John A. Hawkins, 5672. Oxford: Blackwell.

ARONOFF, MARK . 1994. Morphology by Itself. Cambridge: MIT Press.

ASKI, JANICE . 1995. Verbal suppletion: An analysis of Italian, French, and Spanish to go. Linguistics, 33.40332.

BAAYEN, H. 2003. Probabilistic approaches to morphology. Probabilistic Linguistics, ed. by R. Bod, J. Hay and S. Jannedy, 22987. Cambridge: MIT Press.

BACHRACH, ASAF and N EVINS, ANDREW . 2008. Introduction: Approaching inflectional identity. Inflectional Identity, ed. by Asaf Bachrach and Andrew Nevins, 128. New York: Oxford.

267

BADIA I MARGARIT, ANTONI M. 1950. El habla del valle de Bielsa (Pirineo aragonés). : Consejo Superior de Estudios Científicos. Instituto de Estudios Pirenaicos.

—. 1962. Gramatica catalana Madrid: Editorial Gredos.

BAIST, GOTTFRIED . 1888. Die spanische Sprache. Strassburg.

BEC, PIERRE . 1973. Manuel pratique d'occitan moderne. Paris: Picard.

BELLO, ANDRÉS . 1964. Gramatica de la lengua castellana. : Sopena Argentina.

BLASCO FERRER, EDUARDO . 1984. Grammatica storica del catalano e dei suoi dialetti con speciale riguardo all'algherese. Tubingen: Gunter Narr Verlag Tübingen.

BOD, RENS , H AY, JENNIFER and J ANNEDY, STEPHANIE (eds.) 2003. Probabilistic lingusitics. Cambridge: MIT Press.

—. 2003. Introduction—Probabilistic Linguistics. Probabilistic lingusitics, ed. by Rens Bod, Jennifer Hay and Stephanie Jannedy, 110. Cambridge: MIT Press.

BOOMERSHINE, AMANDA REITER . 2005. Perceptual Processing of Variable Input in Spanish: An exemplarbased approach to speech perception, Department of Spanish and Portuguese, The Ohio State University.

BOSQUE, IGNACIO and P ÉREZ FERNÁNDEZ, MANUEL . 1987. Diccionario inverso de la lengua española. Madrid: Gredos.

BOURCIEZ, EDOUARD . 1967. Elements de linguistique romane. Paris: Klincksieck.

BOYDBOWMAN, PETER . 1980. From Latin to Spanish in Sound Charts. Washington D.C.: Georgetown University Press.

BULL, E., C ANTÓN, ALFREDO , C ORD, WILLIAM , F ARLEY, RODGER , F INAN, JOHN , JACOBS, SUZANNE , J AEGER, , K OONS, MARIE and T UEGEL, BARBARA . 1947. Modern Spanish VerbForm Frequencies [Nov]. , 30.45166.

BURZIO, LUIGI . 2004. Paradigmatic and Syntagmatic Relations in Italian Verbal Inflection. Contemporary Approaches to : selected papers from the 33rd Linguistic Symposium on Romance Languages (LSRL), Bloomington, Indiana, April 2003, ed. by Julie Auger, J. Clancy Clements and Barbara Vance, 1744. Amsterdam: John Benjamins.

268

—. 2005. Sources of Paradigm Uniformity. Paradigms in Phonological Theory, ed. by Laura J. Downing, Tracy A. Hall and Renate Raffelsiefen, 65106. Oxford: Oxford University Press.

BUSQUETS, LORETO and B ONZI, LIDIA . 1993. Los verbos en español. Madrid: Verbum.

BUSTOS GISBERT, E. 1989. Algunas observaciones sobre las alternancias vocálica y consonántica en el lexema verbal del presente. Philologica. Homenaje a D. Llorente, ed. by Julio Borrego Nieto, José Jesús Gómez Asencio and Santos Río, 25570. : Ediciones Universidad de Salamanca.

BYBEE, JOAN and B REWER, MARY . 1980. Explanations in Morphophonemics: Changes in Provençal and Spanish Preterite Forms. Lingua, 52.20142.

BYBEE, JOAN and P ARDO, ELLY . 1981. On Lexical and Morphological Conditioning of Alternations: A NonceProbe Experiment with Spanish Verbs. Linguistics, 19.93768.

BYBEE, JOAN and S LOBIN, DAN I. 1982. Rules and Schemas in the development and use of the English . Language, 58.26589.

BYBEE, JOAN and M ODER, CAROL LYNN . 1983. Morphological classes as natural categories. Language, 59.25170.

BYBEE, JOAN . 1985. Morphology: A Study of the Relation Between Meaning and Form. Amsterdam: John Benjamins.

—. 1995. Diachronic and Typological Properties of Morphology and Their Implications for Representation. Morphological Aspects of Language Processing, ed. by Laurie Beth Feldman, 22546. Hillsdale, : Erlbaum.

—. 1996. Productivity, Regularity and Fusion: How Language Use Affects the Lexicon. Trubetzkoy's Orphan: Proceedings of the Roundtable 'Morphonology: Contemporary Responses', ed. by Rajendra Singh, Richard Desrochers and Etienne Tiffou, 24769; 8083. Amsterdam: Benjamins.

BYBEE, JOAN , G OAD, HEATHER , T IFFOU, ETIENNE , K IPARSKY, PAUL and J ANDA, RICHARD . 1996. Productivity and the lexicon: Discussion. Trubetzkoy's Orphan: Proceedings of the Montreal Roundtable 'Morphonology: Contemporary Responses', ed. by Rajendra Singh, Richard Desrochers and Etienne Tiffou, 28494. Amsterdam: Benjamins.

269

BYBEE, JOAN . 2000. The Phonology of the Lexicon: Evidence from Lexical Diffusion. Usagebased models of language, ed. by Suzanne Kemmer and Michael Barlow, 6585. Stanford: CSLI Publications.

—. 2002. Phonological Evidence for Exemplar Storage of Multiword Sequences. SSLA, 24.21521.

—. 2007. From usage to grammar: The mind's response to repetition. Language, 82.71133.

CAMPBELL, LYLE . 1974. On Conditions on Sound Change. , ed. by John M. Anderson and Jones, 8997. Amsterdam: NorthHolland.

CANO AGUILAR, RAFAEL . 1999. El español a través de los tiempos. Madrid: Arco/Libros.

CARRERA DE LA RED, MICAELA . 1999. Morfofonología del español en Santo Domingo. Una perspectiva de cinco siglos. Actas del XI Congreso Internacional de la ALFAL, ed. by J. A. Samper, 184965. : Universidad de Las PalmasALFAL.

CARSTAIRS, ANDREW . 1983. Paradigm Economy. Journal of Linguistics, 19.11525.

—. 1990. Phonologically Conditioned Suppletion. Contemporary Morphology, ed. by Wolfgang Dressler, H. C. Luschützky, O. E. Pfeiffer and J. R. Rennison, 1723. Berlin: Mouton de Gruyter.

CHANDLER, STEVE . 1993. Are rules and modules really necessary for explaining language? Journal of Psycholinguistic Research, 22.593606.

CLAHSEN, HARALD , S ONNENSTUHL, INGRID , H ADLER, MEIKE and E ISENBEISS, SONJA . 2001. Morphological Paradigms in Language Processing and Language Disorders [Nov]. Transactions of the Philological Society, 99.24777.

CLARET (ed.) 2002. Els verbs conjugats. Barcelona: Claret.

CLARK, EVE V. 1986. The acquistion of Romance, with special reference to French. Hillsdale: Lawrence Erlbaum Associates.

COLANERI, JOHN and L UCIANI, VINCENT . 1992. 501 Italian Verbs. Hauppauge: Barron's.

COROMINAS, JOAN and P ASCUAL, JOSÉ A. 1980. Diccionario crítico etimológico castellano e hispánico. Madrid: Gredos.

270

COSTANZO, ANGELO . 2006. Analogy in the Formation of Verb Classes in Catalan. Paper presented at 3rd Annual OSU MLK Day Symposium in Linguistics, Columbus.

—. 2007. Variable Analogical Extension of Velar Augments in Dialectal Catalan. Paper presented at 18th International Conference on Historical Linguistics, Montreal.

—. forthcoming. Romance Conjugational Classes: Learning from the Peripheries. Department of Linguistics, The Ohio State University.

CRESSEY, WILLIAM . 1972. Irregular Verbs in Spanish. Generative Studies in Romance Linguistics, ed. by Jean Casagrande and Bohdan Saciuk, 23646. Rowley: Newbury House Publishers, Inc.

CROCCOGALEAS, GRAZIA . 1989. Suppletivismo debole negli entici italiani: la nozione di schema. Parallela 4 Morfologia / Morphologie, ed. by Monica Berretta, P. Molinelli and A. Valentini, 21929. Tübingen: Günter.

CULICOVER, PETER and N OWAK, ANDRZEJ . 2003. Dynamical Grammar. Oxford: Oxford University Press.

DAVIES, MARK . 2006. A Frequency Dictionary of Spanish. New York: Routledge.

DELL, FRANÇOIS and E LMEDLAOUI, MOHAMED . 1996. On Consonant Releases in Imdlawn Tashlhiyt Berber. Linguistics, 34.35795.

DIEZ, FRIEDRICH . 18741876. Grammaire des langues romanes. Paris: A. Franck.

—. 1887. Etymologisches Wörterbuch der romanischen Sprachen. Bonn: A. Marcus.

DOWNING, LAURA J., H ALL, T. ALLEN and R AFFELSIEFEN, RENATE . 2005. Introduction: The Role of Paradigms in Phonological Theory. Paradigms in Phonological Theory, ed. by Laura J. Downing, Tracy A. Hall and Renate Raffelsiefen, 116. Oxford: Oxford University Press.

— (eds.) 2005. Paradigms in Phonological Theory. Oxford Studies in Theoretical Lingusitics. Oxford: Oxford University Press.

DWORKIN, STEVEN N. and M ALKIEL, YAKOV . 1983. The Fragmentation of the Latin Verb tollere in Hispano(Including Luso) Romance. Romance Philology, 37.16574.

EDDINGTON, DAVID . 2002. Issues in modeling language processing analogically. Analogical Modeling Reseach Group at BYU

ELCOCK, W. D. 1960. The Romance Languages. London: Faber and Faber Limited. 271

ELSON, MARK J. 1988. The Synchronic Status and the Evolution of the g in Spanish vengo, salgo , etc. revisited [May 1988]. Hispania, 71.392400.

ESPINOSA, C. 1955. La excepción del verbo inmiscuir de los verbos irregulares de la décima clase a la que sirve de modelo la conjugación del verbo huir. Boletín de la Academia Cubana de la Lengua, 4.10916.

FANCIULLO, FRANCO . 1998. Per una interpretazione dei verbi italiani a 'inserto' velare. Archivio glottologico italiano, LXXXIII.188239.

FAULHABER, CHARLES B., G ÓMEZ MORENO, ANGEL , C ORTIJO OCAÑA, ANTONIO and P EREA RODRÍGUEZ, OSCAR . 2002. PhiloBiblon BETA (Bibliografía Española de Textos Antiguos) [online]. Berkeley: The Regents of the University of

FEDERICI, STEFANO and P IRRELLI, VITO . 1997. Analogy, computation and linguistic theory. New Methods in Language Processing, ed. by D. B. Jones and H. L. Somers, 1634. London: Univeristy College of London.

FONDOW, STEVEN R. 2007. Velarinsertion and variation in the Old Spanish verbal paradigm. Paper presented at 18th International Conference on Historical Linguistics, Montreal.

FORD, J. D. M. 1966. Old Spanish Readings. : Ginn and Co.

FOUCHÉ, PIERRE . 1976. Morphologie historique du français. Le verbe. Paris: Klincksieck.

FRADEJAS RUEDA, JOSÉ MANUEL . 1997. Fonología histórica del español. Madrid: Visor Libros.

GAGARINA, NATALIA and G ÜLZOW, INSA . 2006. Introduction—The Acquisition of Verbs and their Grammar : The Effect of Particular Languages. The Acquisition of Verbs and their Grammar : The Effect of Particular Languages, ed. by Natalia Gagarina and Insa Gülzow, 111. Dordrecht: Springer.

GARCÍA DE DIEGO, VICENTE . 1951. Gramática histórica española. Madrid: Gredos.

—. 1985. Diccionario etimológico español e hispánico. Madrid: EspasaCalpe.

GASSNER, ARMIN . 1897. Das altspanische verbum. Halle a.S.: M. Niemeyer.

GOLDINGER, D. 1998. Echoes of Echoes? An Episodic Theory of Lexical Access. Psychological Review, 105.25179.

272

GOSSEN, CARL . 1970. Grammaire de l'ancien picard. Paris: Klincksieck.

GRANDGENT, CHARLES H. 1907. An Introduction to Vulgar Latin. Boston: Heath and Co.

HALL, ROBERT A. 1983. ProtoRomance Morphology. Amsterdam: John Benjamins.

HALL, KATHLEEN C. 2006. Sound change in an exemplarbased model of grammar. Paper presented at 3rd Annual OSU MLK Day Symposium in Linguistics, Columbus.

HALL, KATHLEEN C. and B OOMERSHINE, AMANDA REITER . 2006. Life, The Critical Period: An ExemplarBased Model of Language Learning: The Ohio State University

HANSSEN, FREDERICH . 1913. Gramática histórica de la lengua castellana. Halle: M. Niemeyer.

HARRIS, JAMES . 1972. Five classes of irregular verbs in Spanish. Generative Studies in Romance Languages, ed. by Jean Casagrande and Bohdan Saciuk, 24771. Rowley: Newbury House.

—. 1973. On the Order of Certain Phonological Rules in Spanish. A Festschrift for Morris Halle, ed. by Stephen R. Anderson and Paul Kiparsky, 5976. New York: Holt, Rinehart and Winston.

HARRIS, MARTIN and V INCENT, NIGEL (eds.) 1988. The Romance languages. New York: Oxford University Press.

HIPPISLEY, ANDREW , C HUMAKINA, MARINA , C ORBETT, GREVILLE G. and B ROWN, DUNSTAN . 2004. Suppletion: Frequency, categories and distribution of stems. Studies in Language: International Journal Sponsored by the Foundation 'Foundations of Language', 28.387418.

HOCK, HANS HEINRICH . 1991. Principles of Historical Linguistics. Berlin: Mouton de Gruyter.

—. 2005. Analogical Change. The Handbook of Historical Lingusitics, ed. by Brian Joseph and Richard Janda, 44160. Malden: Blackwell Publishing.

HOGG, RICHARD . 1980. Analogy as a Source of Morphological Complexity. Folia Linguistica Historica, 1.27784.

HOOPER, JOAN BYBEE . 1976. Word frequency in lexical diffusion and the source of morphophonological change. Current progress in historical linguistics, ed. by William M. Christie, 96105. Amsterdam: North Holland.

273

—. 1979. Substantive principles in Natural Generative Phonology. Current approaches to phonological theory, ed. by Dinnsen. Bloomington: Indiana University Press.

INDUSTRIAL SOFT . 2009. English<>Romanian Dictionary Dictionar Englez Roman [online]. Montreal: Industrial Soft

INSTITUT D' ESTUDIS . 2004. Diccionari de la llengua catalana [online]. Barcelona: Institut d'Estudis Catalans

IORDAN, IORGU and M ANOLIUMANEA, MARIA . 1972. Manual de linguistica romanica vol. 1. Madrid: Gredos.

JANDA, RICHARD and J OSEPH, BRIAN . 2005. On Language, Change, and Language Change—Or, Of History, Linguistics, and Historical Linguistics. The Handbook of Historical Linguistics, ed. by Brian Joseph and Richard Janda, 3180. Malden: Blackwell.

JOHNSON, KEITH . 1997. Speech perception without speaker normalization: An exemplar model. Talker Variability in Speech Processing, ed. by Keith Johnson and John W. Mullennix, 14566. : Academic Press.

—. 2005. Speaker normalization in speech perception. The Handbook of Speech Perception, ed. by D. B. Pisoni and R. Remez, 36389. Oxford: Blackwell.

JOSEPH, BRIAN and J ANDA, RICHARD . 1988. The How and Why of Diachronic Morphologization and Demorphologization. Theoretical Morphology: approaches in modern linguistics., ed. by M. Hammond and M. Noonan, 193210. San Diego: Academic Press.

JOSEPH, BRIAN . 1982. Multiple causation in change [microfiche]. Educational Resources Information Center.

—. 1983. The Synchrony and Diachrony of the Balkan Infinitive: A Study in Areal, General and Historical Linguistics. Cambridge: Cambridge University Press.

—. 1992. Diachronic explanation: Putting speakers back into the picture. Explanation in Historical Linguistics, ed. by G. W. Davis and G. K. Iverson, 12344. Amsterdam: John Benjamins.

—. 1997a. How General are our Generalizations? What Speakers Actually Know and What They Actually Do. ESCOL 1996. Proceedings of the Thirteenth Eastern States Conference on Linguistics, ed. by A. Green and V. Motopanyane, 14860. Ithaca: Cascadilla Press. 274

—. 1997b. On the linguistics of marginality: The centrality of the periphery. Papers from the 33 rd Regional Meeting of the Chicago Linguistic Society, ed. by Kora Singer, Randall Eggert and Gregory Anderson, 197213. Chicago: Chicago Linguistic Society.

—. 2001. Historical Linguistics. The Handbook of Linguistics, ed. by Mark Aronoff and J. ReesMiller, 10529. Malden: Blackwell.

—. 2004. Optimality, Optimization, and Analogy—A Reconsideration: Back to Basics (and Beyond). Columbus: The Ohio State University

—. 2006. Partial Analogies within Paradigms: Diachronic Perspectives and Synchronic Consequences. Paper presented at 3rd Annual OSU MLK Day Symposium in Linguistics, Columbus.

—. 2009. Greek dialectal evidence for the role of the paradigm in inflectional change [June 2009]. Morphology, 19.4557.

—. To appear in: A localistic approach to universals and variation. Universals and Variation, ed. by Peter Siemund, 394414. Berlin: Mouton de Gruyter.

JOSEPH, BRIAN and J ANDA, RICHARD (eds.) 2005. The Handbook of Historical Linguistics. Oxford: Blackwell.

JOSEPH, BRIAN and S IMS, . 2006. Mechanisms of Paradigmatic Change: Diachronic Evidence for Paradigms. Paper presented at 80th Annual Meeting of the Linguistic Society of America, Albuquerque.

JUILLAND, ALPHONSE and C HANGRODRÍGUEZ, EUGENIO . 1964. Frequency Dictionary of Spanish Words. The Hague: Mouton.

JURAFSKY, DAN . 2003. Probabilistic Modeling in : Linguistic Comprehension and Production. Probabilistic lingusitics, ed. by Rens Bod, Jennifer Hay and Stephanie Jannedy, 3995. Cambridge: MIT Press.

KASTEN, LLOYD and C ODY, FLORIAN . 2001. Tentative Dictionary of Medieval Spanish. New York: Hispanic Seminary of Medieval Studies.

KENDRIS, . 1996a. 501 . Hauppauge: Barron’s.

—. 1996b. 501 Spanish Verbs. Hauppauge: Barron's.

KENSTOWICZ, MICHAEL . 1994. Phonology in . Cambridge: Blackwell. 275

—. 1996. BaseIdentity and Uniform Exponence: Alternatives to Cyclicity. Rutgers Optimality Archive

—. 2005. Paradigmatic Uniformity and Contrast. Paradigms in Phonological Theory, ed. by Laura J. Downing, T. Allen Hall and Renate Raffelsiefen, 14569. Oxford: Oxford University Press.

KING, ROBERT D. 1969. Historical Linguistics and Generative Grammar. Englewood Cliffs: PrenticeHall.

KIPARSKY, PAUL . 1974. Remarks on Analogical Change. Historical Linguistics, ed. by John M. Anderson and Charles Jones, 25575. Amsterdam: NorthHolland.

KLAUSENBURGER, JÜRGEN . 1984. The morphology of the velar insert in . Romanitas: studies in Romance linguistics, ed. by Ernest Pulgram, 13251. Ann Arbor: Romance Studies.

KRASKASZLENK, IWONA . 2007. Analogy: The Relation between Lexicon and Grammar. Muenchen: Lincom.

KUHL, PATRICIA and I VERSON, PAUL . 1995. Linguistic Experience and the ‘Perceptual Magnet Effect. Speech Perception and Linguistic Experience: Issues in CrossLanguage Research, ed. by Winifred Strange, 12154. Timonium: York Press.

KURYŁOWICZ, JERZY . 1949. La nature des procès dits 'analogiques'. Acta Linguistica, 5.1537.

LABOV, WILLIAM . 1994. Principles of Linguistic Change.vol. 1. Oxford: Blackwell.

LAHIRI, ADITI . 2000. Introduction. Analogy, Levelling, Markedness: Principles of Change in Phonology and Morphology, ed. by Aditi Lahiri, 114. Berlin: Mouton de Gruyter.

— (ed.) 2000. Analogy, Levelling, Markedness: Principles of Change in Phonology and Morphology. Berlin: Mouton de Gruyter.

LAPESA, RAFAEL . 1980. Historia de la lengua española. Madrid: Gredos.

LATHROP, T. A. 1986. The Evolution of Spanish. An Introductory Historical Grammar. Newark: Juan de la Cuesta.

276

LENFEST, DONALD E. 1978. An Explanation of the /G/ in 'Tengo,' 'Vengo,' 'Pongo,' 'Salgo,' and 'Valgo'. Hispania, 61.894904.

—. 1993. TengoVengo, an Update. Hispania, 76.63444.

LINARES, RAFAEL ENRIQUE , R ODRÍGUEZFORNELLS, ANTONI and C LAHSEN, HARALD . 2006. Stem allomorphy in the Spanish mental lexicon: Evidence from behavioral and ERP experiments [April, 2006]. Brain and Language, 97.11020.

LLOYD, PAUL M. 1987. From Latin to Spanish. Vol. I, Historical phonology and morphology of the Spanish language. Philadelphia: American Philosophical Society.

LOMBARD, ALF and C ONSTATIN, GÂDEI . 1981. Dictionnaire morphologique de la langue roumaine: permettant de connaître la flexion entière des mots qui en possèdent une, substantifs, adjectifs, pronoms, verbes. Lund: Gleerup.

MAIDEN, MARTIN . 1992. Irregularity as a determinant of morphological change. Journal of Linguistics, 28.285312.

—. 1993. The Role of Paradigms in the Phonetic Detail of Sound Change. Historical Linguistics 1989: Papers from the 9th International Conference on Historical Linguistics, ed. by Henk Aertsen and Robert J. Jeffers, 28396. Amsterdam: John Benjamins.

—. 2001. Di nuovo sulle alternanze velari nel verbo italiano e spagnolo. Cuadernos de filologia italiana, 8.3961.

—. 2003. Verb Augments and Meaninglessness in Early Romance Morphology. Studi di grammatica italiana, ed. by Nicoletta Maraschio, Francesco Sabatini, Harro Stammerjohann and Domenico De Martino, 161. Firenze: Le Lettere.

—. 2004. When lexemes become allomorphs: On the genesis of suppletion. Folia Linguistica: Acta Societatis Linguisticae Europaeae, 38.22756.

MAIDEN, MARTIN , O 'NEILL, PAUL and S WEARINGEN, ANDREW . 2007. Imperative Morphology in Diachrony: Evidence from the Romance Languages. Paper presented at International Conference on Historical Linguistics, Montreal.

MAIDEN, MARTIN and O 'NEILL, PAUL . 2008. On Morphomic Defectiveness: Evidence from the Romance languages of the Iberian Peninsula. Oxford

MALKIEL, YAKOV . 1958. Los interfijos hispánicos: Problema de lingüística histórica y estructural. Miscelánea homenaje a André Martinet. Madrid: Gredos. 277

—. 1967. The Inflectional Paradigm as an Occasional Determinant of Sound Change. Directions for Historical Linguistics: A Symposium, ed. by Winfred P. Lehmann and Yakov Malkiel, 2164. Austin: U of .

—. 1969. Morphological Analogy as a Stimulus for Sound Change. Lingua e stile, 4.30527.

—. 1974. New Problems in Romance Interfixation (I): The Velar Insert in the Present Tense (with an Excursus on zer/zir Verbs). Romance Philology, 27.30455.

—. 1982. Interplay of Sounds and Forms in the Shaping of Three Old Spanish Medial Consonant Clusters. Hispanic Review, 50.24766.

—. 1989. Divergent development of 'inchoatives' in late Old Spanish and Old Portuguese. Studia linguistica et orientalia memoriae Haim Blanc dedicata, ed. by P. Wexler, A. Borg and S. Somekh, 20018. Wiesbaden: Harrassowitz.

MAŃCZAK, WITOLD . 1958. Tendances générales des changements analogiques. Lingua, 7.298325.

MARTÍN VEGAS, ROSA ANA . 2007. Morfofonología histórica del español. München: Lincom.

MCCARTHY, JOHN . 2003. Optimal Paradigms. Rutgers Optimality Archive

MENÉNDEZ PIDAL, RAMÓN . 1976. Orígenes del español. Madrid: EspasaCalpe.

—. 1980. Manual de gramática histórica española. Madrid: EspasaCalpe.

MEYERLÜBKE, WILHELM . 1895. Grammaire des langues romanes: morphologie. Paris: Welter.

—. 1972. Romanisches Etymologisches Wörterbuch. Heidelberg: Carl Winter.

MINSKY, MARVIN . 1975. A framework for representing knowledge. The Psychology of Computer Vision, ed. by P. Winston. New York: McGrawHill.

MOLL, FRANCESC DE BORJA . 1952. Gramatica historica catalana. Madrid: Editorial Gredos.

—. 1975. Gramatica Catalana. Palma de : Moll. [Reprinted in , Univesitat de València, 1991].

278

MONDÉJAR CUMPIÁN, JOSÉ . 1995. Los presentes de subjuntivo anómalos en g, y ig. Intento de explicación. Estudis de linguistica i filologia oferts a Antoni M. Badia i Margarit, 1123. Barcelona: Publicacions de l'Abadia de Monserrat.

MOYA, JOSÉ ANTONIO , O RTIZ, MARÍA JESÚS and O RTIZ, GREGORIO . 2003. Diccionario del revés. : ECU.

MYERS, JAMES . 1999. Lexical phonology and the lexicon: Rutgers Optimality Archives

—. 2002. Exemplardriven analogy in Optimality Theory. Analogical Modeling: An exemplarbased approach to language, ed. by Royal Skousen, D. Lonsdale and D. Parkinson, 265301. Amsterdam: John Benjamins.

NAPOLI, DONNA JO and V OGEL, IRENE . 1990. The Conjugations of Italian [Winter]. Italica, 67.479502.

NEBRIJA, ANTONIO DE . 1989. Gramática de la lengua castellana. Estudio y edición. Madrid: Editorial Centro de Estudios Ramon Aredes.

NITTI, JOHN J. 1995. 501 Portuguese Verbs…. Hauppauge: Barron’s Educational Series.

NOSOFSKY, ROBERT M. 1986. Attention, similarity, and the identificationcategorization relationship. Journal of Experimental Psychology: General, 115.3957.

—. 1988. Exemplar based accounts of relations between classification, recognition, and typicality. Journal of Experimental Psychology: Learning, Memory and Cognition, 14.70008.

NOSOFSKY, ROBERT M. and AKI, SAFA R. 2002. Exemplar and Prototype Models Revisited: Response Strategies, Selective Attention, and Stimulus Generalization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28.92440.

OELSCHLÄGER, VICTOR . 1940. A Medieval Spanish WordList. A Preliminary Dated Vocabulary of First Appearances up to Berceo. Madison: University of Wisconsin Press.

OHALA, JOHN . 2005. Phonetics and Historical Phonology. The Handbook of Historical Lingusitics, ed. by Brian Joseph and Richard Janda, 66986. Malden: Blackwell Publishing.

OLIVETTI, ENRICO . 2009. Italian Verbs [online]: Olivetti, Enrico

OTERO, C. P. 1971. Evolución y revolución en romance. Barcelona: Seix Barral. 279

OXFORD ENGLISH DICTIONARY . 1989. OED [online]. Oxford: Oxford University Press

PANOCCITAN .ORG . 2009. panOccitan.org La conjugasion occitane [online]: panOccitan.org

PENNY, RALPH . 1969. El habla pasiega : Ensayo de dialectología montañesa. London: Tamesis.

—. 2002. A history of the Spanish language. Cambridge: Cambridge University Press.

PENSADO RUIZ, CARMEN . 1997. On the Spanish Depalatalization of /n'/ and /l'/ in . Issues in the Phonology and Morphology of the Major , ed. by Fernando MartínezGil and A. Morales Front, 595618. Washington D.C.: Georgetown University Press.

PÉREZ SALDANYA, MANUEL . 1995a. Un problema de morfologia verbal romanica: les formes verbals velaritzades. Actas del I Congreso de Historia de la Lengua Española en América y España, ed. by María Teresa Echenique Elizondo, Milagros Aleza and María José Martínez, 41020. Valencia: Tirant lo Blanch.

—. 1995b. Analogia i canvi morfològic: A propòsit de les formes verbals velaritzades. Caplletra, 19.279305.

PHILLIPS, BETTY S. 1984. Word Frequency and the Actuation of Sound Change [June 1984]. Language, 60.32042.

—. 1999. The Mental Lexicon: Evidence from Lexical Diffusion. Brain and Language, 68.10409.

PIERREHUMBERT, JANET . 2001. Exemplar Dynamics: Word Frequency, Lenition and Contrast. Frequency and the Emergence of Linguistic Structure, ed. by Joan Bybee and Paul Hopper, 13757. Amsterdam: John Benjamins.

—. 2002. WordSpecific Phonetics. Laboratory Phonology, 7, ed. by Carlos Gussenhoven and Natasha Warner. 124. Berlin: Mouton de Gruyter

PIRRELLI, VITO . 2000. Paradigmi in morfologia. Un approccio interdisciplinare alla flessione verbale dell'italiano. Pisa: Istituti editoriali e poligrafici internazionali.

PIRRELLI, VITO and B ATTISTA, MARCO . 2000a. The Paradigmatic Dimension of Stem Allomorphy in Italian Verb Inflection. Rivista di linguistica, 12.30780.

280

—. 2000b. On the interaction of paradigmatic and syntagmatic stem alternation in . Acta Linguistica Hungarica, 47.289314.

PRIOR, RICHARD E. and W OHLBERG, JOSEPH . 1995. 501 Latin Verbs…. Hauppauge: Barron’s Educational Series.

REAL ACADEMIA ESPAÑOLA . 2004. Diccionario de la lengua española [online]. Madrid: Real Academia Española

—. 2005. Corpus diacrónico del español (CORDE) [online]. Madrid: Real Academia Española

—. 2007. Corpus de referencia del español actual (CREA) [online]. Madrid: Real Academia Española

REBRUS, PÉTER and T ÖRKENCZY, MIKLÓS . 2005. Uniformity and Contrast in the Hungarian Verbal Paradigm. Paradigms in Phonological Theory, ed. by Laura J. Downing, Tracy A. Hall and Renate Raffelsiefen, 26395. Oxford: Oxford University Press.

REISS, CHARLES . 1997. Explaining Analogy. Rutgers Optimality Archive: Rutgers University

RESNICK, MELVYN C. 1981. Introducción a la historia de la lengua española. Washington D.C.: Georgetown University Press.

RISTAD, S. 1993. The Language Complexity Game. Cambridge: MIT Press.

ROHLFS, GERHARD . 1968. Grammatica storica della lingua italiana e dei suoi dialetti: Morfologia.vol. 2. Torino: Giulio Einaudi.

SANTANA SUÁREZ, OCTAVIO . 2002. Manual de la conjugación del español. 12790 verbos conjugados. Madrid: Arcos/Libros.

SAPIR , EDWARD . 1921. Language, an introduction to the study of speech. New York: Harcourt, Brace and Company.

SAURA RAMI, JOSE ANTONIO . 2000. Aspectos de fonética y de analogía en la flexión verbal benasquesa. Alazet, 12.14759.

SCARBOROUGH, D. L., C ORTESE, C. and S CARBOROUGH, H. S. 1977. Frequency and repetition effects in lexical memory. Journal of Experimental Psychology, 3.117.

SCHEDE, HILDEGARD . 1987. Die Morphologie des Verbes im Altspanischen. Frankfurt am Main: Lang. 281

SIMS, ANDREA . 2006. Analogy and Paradigms in Current Morphological Theory. Paper presented at 3rd Annual OSU MLK Day Symposium in Linguistics, Columbus.

SKOUSEN, ROYAL . 1989. Analogical modeling of language. Dordrecht: Kluwer.

—. 1992. Analogy and structure. Dordrecht: Kluwer.

—. 1995. Analogy: A nonrule alternative to neural networks. Rivista di linguistica, 7.21332.

—. 2002. An Overview of Analogical Modeling; Issues in Analogical Modeling. Analogical modeling: an exemplarbased approach to language, ed. by Royal Skousen, Deryle Lonsdale and Dilworth B. Parkinson, 1148. Amsterdam: John Benjamins.

—. 2003. Analogical Modeling: Exemplars, Rules and Quantum Computing. Paper presented at 29th Annual Meeting of the Berkeley Linguistics Society, Berkeley.

SPAULDING, ROBERT K. 1971. How Spanish Grew. Berkeley: University of California Press.

STEMBERGER, JOSEPH PAUL . 1994. Ruleless morphology at the phonologylexicon interface. The reality of linguistic rules, ed. by Susan D. Lima, Roberta L. Corrigan and Gregory K. Iverson, 14769. Amsterdam: John Benjamins.

STERIADE, DONCA . 1996. Paradigm Uniformity and the PhoneticsPhonology Boundary. Paper presented at 5th Conference in Laboratory Phonology, Evanston, .

—. 2000. Paradigm Uniformity and the PhoneticsPhonology Boundary. Papers in Laboratory Phonology V: Acquisition and the Lexicon, ed. by Michael Broe and Janet Pierrehumbert, 31335. Cambridge: Cambridge University Press.

STUMP, GREGORY . 2001. Inflectional Morphlogy: A theory of paradigm structure. Cambridge: Cambridge University Press.

SUANCES, JAIME . 2000. Diccionario del verbo español, hispanoamericano y dialectal. Barcelona: Herder.

TEKAVČIĆ, PAVAO . 1980a. Grammatica storica dell'italiano: I. Fonematica. Bologna: Società editrice il Mulino.

—. 1980b. Grammatica storica dell'italiano: II. Morfosintassi. Bologna: Società editrice il Mulino. 282

TOMASELLO, M. 2001. First steps in a usagebased theory of acquisition. Cognitive Linguistics, 11.6182.

URRUTIA CARDENAS, HERNÁN and A LVAREZ ALVAREZ, MANUELA . 1983. Esquema de morfosintaxis histórica del español. Bilbao: Universidad de Deusto.

VINCENT, NIGEL . 1988. Italian. The Romance Languages, ed. by Martin Harris and Nigel Vincent, 279313. Oxford: Oxford University Press.

WANG, WILLIAM S. Y. 1969. Competing sound changes as a cause of residue. Language, 45.925.

WANNER, DIETER . 2005. Diachronic Morphology as Analogy. Romance Verb Forms. 34th Linguistic Symposium on Romance Languages. Salt Lake City

—. 2006a. Diachronic Morphology as Analogy. Romance Verb Forms. Columbus: The Ohio State University

—. 2006b. The Power of Analogy: An Essay on Historical Lingusitics: Trends in Linguistics. Berlin: Mouton de Gruyter.

—. 2006c. The Logic of Analogy. Exemplars of analogical change in morphology and syntax. Paper presented at 3rd Annual OSU MLK Day Symposium in Lingusitics, Columbus.

WEBSTER, LEE CHARLES, JR. 1987. Historical Morphology of the GalicianPortuguese Verb: The Direction of Analogical Change, University of : Ph.D. dissertation.

WHEELER, MAX W. 1995. La primera persona del present d'indicatiu, por haverhi més a dirne? Estudis de linguistica i filologia oferts a Antoni M. Badia i Margarit, 41126. Barcelona: Publicacions de l'Abadia de Monserrat.

WHITTLESEA, BRUCE W. A. 1987. Preservation of specific experiences in the representation of general knowledge. Journal of Experimental Psychology: Learning, Memory and Cognition, 13.317.

WIESE, BERTHOLD . 1928. Altitalienisches Elementarbuch. Heidelberg: C. Winter.

WILKINSON, HUGH E. 1978. Palatal vs. velar in the stem of the Romance present (I). Ronshu 19.1935.

—. 1979. Palatal vs. velar in the stem of the Romance present (II). Ronshu 20.1935. 283

—. 1980. Palatal vs. velar in the stem of the Romance present (III). Ronshu 21.4162.

—. 1981. Palatal vs. velar in the stem of the Romance present (IV). Ronshu 22.6785.

—. 1982. Palatal vs. velar in the stem of the Romance present (V). Ronshu 23.11536.

—. 1983. Palatal vs. velar in the stem of the Romance present (VI). Ronshu 24.17799.

WILLIAMS, EDWIN B. 1962. From Latin to Portuguese—Historical Phonology and Morphology of the . Philadelphia: University of Press.

WONGOPASI, UTHAIWAN . 1993. The Effects of Yod on the Vocalic and Consonantal Systems: from Latin to Spanish. Historical Linguistics 1989: Papers from the 9th International Conference on Historical Linguistics, ed. by Henk Aertsen and Robert J. Jeffers, 497524. Amsterdam: John Benjamins.

ZURAW, KIE . 2003. Probability in Language Change. Probabilistic lingusitics, ed. by Rens Bod, Jennifer Hay and Stephanie Jannedy, 13976. Cambridge: MIT Press.

284

Endnotes

1 The rules are what mattered since “analogy was too vague a notion to deal with in a formal model” (Myers 1999:265). For discussion of this issue, see Downing et al. 2005 (cf. Lahiri 2000).

2 The Greek tradition of scientific classification was composed of an analogical model with contiguity and similarity axes. The paradigm (index) provides the example upon which analogy (proportion, empirical regularity) is based. Analogy as a method of problem solving (i.e. hermeneutics—pattern explanation), then, was simply the filling of gaps by means of comparison across these axes. Unlike mathematical proportions, in most cases these gaps are filled by “vaguer similarities” (Anttila 2005:426).

3 There are many recent wellwritten outlines of analogy in historical linguistics, including Martín Vegas 2007, KraskaSzlenk 2007, and Hock 2005.

4 Personal example (2006). Attested as an obsolete PART from the the 16th c. (Oxford English Dictionary 1989).

5 In a sense, these types of change are the same since it is possible to provide a proportional analogical equation for cases of leveling, the forms in the proportion representing a larger analogical set (see below).

6 Sturtevant’s Paradox suggests that “phonetic laws are regular but produce irregularities. Analogic creation is irregular but produces regularity” (Hock 2005:450).

7 Any investigation concerning analogical change is incomplete without reference to the infamous ‘tendencies’ and ‘laws’ of analogy as found in Mańczak 1958 and Kuryłowicz 1949 respectively. The current investigation follows Wanner 2006a, which indicates that these treatments of analogy cannot be viewed as determinative since analogy is not required to follow the guidelines or tendencies they set forth as illustrated by available counterevidence for each tendency and law (cf. Hock 2005). Thus, these works maintain the vagaries of traditional analogy that have doomed it to remain outside the formal domain of linguistic theory.

8 "[Inference] is used in reference to the process of reasoning from parallel cases and presumptive reasoning based upon the assumption that if things have some similar attributes, their other attributes will be similar" (Lahiri 2000:2; cf. Anttila 2005). 285

9 Some examples of the rather arbitrary, less defined nature of analogy are provided in the numerous AM used in previous research to explain Spanish velarinsertion. Arguments against such models is provided in chapter 2 because there are other forms which better fulfill the analogical criteria as defined in this investigation.

10 The following example from the late 19 th c. illustrates an idea which seems to linger on in the discussion of analogy: "As one might expect, opponents of the doctrine of exceptionless sound laws were particulary hostile to the notion of analogy, which was ridiculed as a convenient fad to deal with obvious counterexamples. To be sure, if analogy can be invoked whenever a sound shape in some word a come to (partially) match the sound shape of some word b, and the relatedness between a and b is unrestricted, the empirical explanatory power of analogy is virtually nil. Critics have insisted that analogical formation is in principle always possible, but never necessary (Curtius 1885:39). While the latter is in fact also true of physiologically motivated sound laws (cf. Tobler 1879), many agreed that recourse to analogy should be an ultimum refugium to the linguist, to be invoked only if all else fails, and that such an analysis should be 'happily' abandoned for any alternative explanation, especially a strictly phonological approach (cf. Schmidt 1882; Curtius 1885). While other researchers felt that granting a fundamental priority to strictly phonological explanations was not necessarily justified, they still deplored the lack of rigid criteria to motivate the legitimate use of analogy (Scherer 1868:177)" (Downing et al. 2005:23).

11 Consider the description of the lexical entry for ‘help’ in King (1969): "Typically, in the history of a language, a lexical entry is simplified in ways that reduce the number of idiosyncratic features it carries. The change in the verb 'to help' between Middle English and Modern English is characteristic. In Middle English it was conjugated strong with the helpe(n) halp hulpen holpen ; today it is weak help helped helped . What seems to have changed here is the set of features that determine this verb's morphological behavior. In Middle English it had schematically the lexical entry: (help+ [+ Verb, + Strong, + Class III, ...]) which characterizes it as a strong verb of the third ablaut class plus whatever else (...) is required to determine its grammatical behavior completely. In Modern English it has the lexical entry: (help+ [+ Verb, Strong, ...]) which represents a simplification in that the single feature [ Strong] has replaced the two features [+ Strong, + Class III]" (62). In such a way, the rules are saved through exceptional application by lexical marking or the lack thereof.

12 Several other arguments against a rulebased approach to linguistic organization are found in Chandler 1993.

13 An extreme take on this argument is found in Reiss 1997, which indicates that analogy is a purely diachronic, nongrammatical process and therefore is not of concern in phonological theory. 286

14 The gradualness of language change is a wellknown point of linguistic discussion, cf. Kiparsky 1974, Wang 1969, Joseph 1997a.

15 A single example should suffice. Harris 1973 posits underlying /ak/ and /dik/ as the stems for MSp. hacer and decir . Their 1s forms hago and digo are derived through the voicing of intervocalic /k/. However, the same rule is used for spirantization of underlying /kok/ (INF cocer ) and /proteg/ (INF proteger ) to derive 1s cuezo and protejo . Thus, a special mechanism is hypothesized to reverse the order of application of this and other rules in the former, ‘irregular’ pair of verbs.

16 See, for example, Johnson 1997, Minsky 1975.

17 There are also constraints like Uniform Exponence which are meant to reduce surface allomorphy (e.g. Kenstowicz 1996).

18 In order to attain the Optimal Candidate Output for the fourpart equation: drive :drove :: dive :X, a highly localized OOF constraint, IdentOO ( drove ,dove ;[o]), establishes a relationship between model and target, and outranks the relevant IOF constraints for the past tense Input for dive (Myers 1999:68).

19 "The Ganong effect can be characterized by saying that a phonemesize perceptual unit is a candidate for attraction by a phonemesized mental category while at the same time the wordsized perceptual unit it is part of may also be attracted by a member of the word lexicon, and that the two attractions are cumulative" (Burzio 2005:73).

20 It has been suggested that nonrelevant linguistic and extralinguistic information is also stored in memory (see below).

21 Salience is also used to explain why phonic/form characteristics are often relevant in analogical associations (Skousen 1992, 1995).

22 As a means of illustration, consider the exemplar cloud to be a knot in a large cord of rope. The links between such paradigmatic exemplar clouds, or knots, are ropes of varying thickness stretching in multiple directions towards other knots (or clouds). Bunches of thread represent exemplars and each individual strand, a feature of that exemplar. These bunches are also of varying thickness according to the number of features shared between exemplars and clouds. The more strands and ropes involved, the stronger the connection. As a way of complicating the matter, input may be categorized according to countless pieces of information, creating a multidimensional associative space. Continuing with the rope metaphor, this is represented by individual strands of an exemplar or paradigm rope being connected to other knots or ropes in multiple directions from their source. For additional visual conceptualizations, see §3.3. 287

23 An interesting example of the robustness of linguistic processing is the ability to interpret incomplete information as found in crossword puzzles, text messages, shorthand, etc.

24 AL is typically geared more towards quantum computating and computational modeling of language. The relevant statistical methods are not utilized in AEM (see examples in Wanner 2005).

25 “There is abundant psycholinguistic evidence that imperfect learning plays a prominent role in both directing human verbal behaviour and justifying language change” (Pirrelli & Battista 2000b:296).

26 The above definition excludes predictable allomorphy, e.g. diphthongization of the stem vowel in certain present tense forms determined by the placement of . From a synchronic point of view, these ‘stemchangers’ resemble the velarinsert verbs which are the subject of this investigation as both show a clearly identifiable allomorphic distributional pattern (based on stress placement and the frontness of the following vowel respectively). However, in both instances the identification of those verbs which are members of these ‘irregular’ classes is not systematic, hence their exclusion from the regular category.

27 This discussion does not include defective verbal paradigms which are characterized by the absence of inflected forms in certain cells of the conjugational structure. Such verbs, as well as those defined as suppletive, are not addressed in this investigation.

28 There is little debate that the velar is considered part of the verbal stem. For arguments in favor of its status as an inflectional marker, see Martín Vegas 2007.

29 The Lpattern mentioned above, as described in Maiden 2001, involves the distribution of the velar element in the highlighted forms in Figure 2.1, which resembles an inverted and reversed capital . See additional information below.

30 There are a number of verbs in which rootfinal /g/ is etymological. In some cases, this consonant is found in all forms of the paradigm, as in pagar “pay”, i.e. pago, pagaba, pagué, pague, pagara, pagaré, pagaría , etc. Such forms are not considered members of the velarinsert class. In other instances where the /g/ is due to regular sound changes, i.e. ‘lautgesetzlich’, a verb may traditionally/pedagogically be included in the set of velarinsert forms due to the allomorphy found in their verbal paradigms, e.g. decir “say”, digo (dices), decía, dije, diga, dijera, diré, diría, etc. In this investigation, neither of these of these types of verbs is part of the velarinsert class although the latter is relevant to historical analysis. For present purposes then, the velarinsert verbal subclass 288

includes only those forms where the velar obstruent is not etymological nor results from regular sound change and promotes allomorphy. In this sense, the set is as highlydefined as possible.

31 The verb poner “put, place” is exceptional since its Classical Latin forms did not contain the vocalic hiatus which has been considered to trigger palatalization prior to velarinsertion (cf. Hanssen 1913, Malkiel 1974, Menéndez Pidal 1980, MeyerLübke 1895). Most previous research indicates that this form entered the velarinsert class quite early via analogy in form and function with tener and venir , (e.g. Corominas & Pascual 1980, Lloyd 1987, Penny 2002). This is due to the fact that the earliest attested forms are latinate hypercorrections, e.g. 3s SBJ poncat , a form concurrent to the first attestations of the voiced velar in both tener and venir . However, the voiceless velar can be quite easily viewed as a hypercorrected reconstruction derived from a voiced velar in the Spanish spoken at the time since the scribe producing such a form would certainly be aware of the lenition process that was a notable indicator of contrast between Latin and Spanish resulting in voicing of simple voiceless obstruents in certain phonological environments.

32 The entrance of remanir into the class is typically thought to have been later in spite of the earliest attested velars occurring during the same time period as with the other nasalfinal stems, cf. §3.5.1.

33 The diachronic dialectal data is too sparse to determine which of these is the case. Regardless of such complications, it is plausible that both are true with respect to the individual verbs under analysis, i.e. some having renewed their attraction to the group, others having merely continued to show variation, on a reduced scale, from the initial stages.

34 The data used in the current investigation, including Appendices, is derived from various sources according to language. Catalan: Badia i Margarit 1962, Blasco Ferrer 1984, Claret 2002, García de Diego 1951, Institut d’Estudis Catalans 2004, Klausenberger 1984, Moll 1952, 1975, Pérez Saldanya 1995a, 1995b, Wheeler 1995, Wilkinson 19781983. French: Allen 1988, Fouché 1976, Gossen 1970, Kendris 1996a, Maiden 2001, 2003, Malkiel 1967, 1974, Wilkinson 19781983. Italian: Burzio 2004, Colaneri & Luciani 1992, Fanciullo 1998, Lenfest 1978, Maiden 1992, 2001, 2003, Malkiel 1967, 1974, Napoli & Vogel 1990, Olivetti 2009, Pirrelli 2000, Rohlfs 1968, Tekavčić 1980b. Latin: Malkiel 1974, Prior & Wohlberg 1995. Portuguese and Galician: Malkiel 1974, Nitti 1995, Webster 1987, Wilkinson 19781983, Williams 1962. Provençal, Gascon and Occitan: Bec 1973, Klausenberger 1984, Malkiel 1967, 1974, panOccitan.org 2009, Wilkinson 19781983.

289

Romanian: Industrial Soft 2009, Lombard & Gadei 1981, Maiden 2003, MeyerLübke 1972. Finally, while the primary Spanish data for chapter 3 are taken from the CORDE (Real Academia Española 2005) and CREA (Real Academia Española 2007) databases, relevant examples and information are also found in Alvar & Pottier 1983, Badia i Margarit 1950, Bello 1964, Bosque & Pérez Fernández 1987, Busquets & Bonzi 1993, Cano Aguilar 1999, Carrera de la Red 1999, Clark 1986, Corominas & Pascual 1980, Diez 18741876, 1887, Dworkin & Malkiel 1983, Elson 1988, Espinosa 1955, García de Diego 1951, 1985, Gassner 1897, Grandgent 1907, Hanssen 1913, Harris 1972, Kasten & Cody 2001, Kendris 1996b, Lapesa 1980, Lenfest 1993, Lloyd 1987, Maiden 2003, Malkiel 1967, 1974, Martín Vegas 2007, Menéndez Pidal 1976, 1980, MeyerLübke 1972, Mondéjar 1995, Moya et al. 2003, Nebrija 1989, Oelschläger 1940, Penny 1969, 2002, Real Academia Española 2004, Santana Suárez 2002, Schede 1987, Spaulding 1971, Suances Torres 2000, Wilkinson 19781983.

35 For greater detail on those Romance languages which are not affected by velarinsertion, see Wilkinson 19781983.

36 Though this does not exclude the possibility of some degree of mutual influence or borrowing.

37 Fr. pondre ‘lay an egg’ is excluded due to its phonological shape and highly specified meaning.

38 For further treatment of the changes affecting the vowel system, see Penny 2002.

39 The inclusion of the 3p PRS IND here assumes its analogical leveling on the basis of intraparadigmatic influence. The Spanish forms are not directly derived from those of Classical Latin because they are marked by the theme vowel instead of a nonthematic inflectional morpheme. In contrast, Italian shows a straightforward etymological relationship between 3p forms from CL. For Spanish, it is assumed that this form had already been restructured prior to the process under analysis. Maiden 2001 discusses the U and Lpatterns for Italian and Spanish respectively (see below).

40 Notably, certain dialects of Modern Spanish still show yod developing from front vowels as the result of loss of hiatus, e.g. teatro , dial. [`tjatro] “theater” (Wongopasi 1993).

41 See Tekavčić 1980a:187188 for examples.

42 Spanish phonology prohibits the appearance of palatalized consonants in coda position in the native lexicon (cf. Pensado Ruiz 1997).

290

43 For arguments against yod hardening to /g/, including examples of a velar softening wordinitially, the ad hoc nature of such a hypothesis, its specificity to verbs, etc., see Lenfest 1978.

44 The actual situation leading up to the described stage is more complicated than it would appear. A detailed analysis of the preceding stages of proposed development are found in Alarcos Llorach 1954.

45 Penny 2002 indicates that 1s TRADO is an attested form relevant to the development of this verb in Spanish.

46 This palatal consonant later extended to learned forms ending in –uir , e.g. 1s construyo from construir “construct” (Penny 2002).

47 For additional examples of the set of –NGERE verbs, see Figure 2.8.

48 Forms 6 (3p) like TANGUNT were restructured in Spanish through inter and intraparadigmatic influence. The presence of the theme vowel was extended to these forms on the basis of the rest of the forms of the PRS IND (with the exception of 1s) (also possibly via the PRS SBJ where a vocalic inflectional marker is found systematically). Influence is also believed to have come from CL conjugations I and II where the theme vowel did occur in 3p forms. This is similar to the loss of the theme vowel from all 1s PRS IND forms (e.g. CL DĒBEŌ > Sp. debo ‘1s ought’; CL PŪNIŌ > Sp. puno ‘1s punish’).

49 Malkiel 1967 and 1974, among others, provide perhaps the most informative and detailed, though possibly contradictory (or preferably dual) analyses of this matter in Spanish. More recently, Martín Vegas 2007 examines these developments with a particular focus on analogical pressure and frequency. See also Alvar & Pottier 1983, Mondéjar Cumpián 1995 and Lenfest 1978, 1993. Mention should also be made of similar developments creating a major dialectal division in Italian, on the basis of the geographic distribution of the palatal nasal (central/southern piagnere ‘cry’) against the nasal plus affricate sequence (northern piangere ). For a more detailed discussion of the Italian situation, see Tekavčić 1980a.

50 For alternative (i.e. paradigmexternal) explanations of the leveling that eliminates the allomorphy in these verbal paradigms, see Lenfest 1978 and Malkiel 1974.

51 A thorough diachronic analysis of individual verbs is provided in Martín Vegas 2007.

52 There are numerous changes affecting the inflected and uninflected forms of decir that are irrelevant to the creation of the velarinsert verb class. Of particular interest to these

291

topics are Urrutia Cardenas & Alvarez Alvarez 1983 and Martín Vegas 2007, as well as the numerous traditional examinations of the history of the Spanish language.

53 The loss of yod is favored by Elcock 1960, Lenfest 1978, Martín Vegas 2007, Menéndez Pidal 1980, Mondéjar Cumpián 1995, and Wilkinson 1980 among others. The alternative view is argued in Malkiel 1974 on the basis of lack of attested evidence and an analysis that relies on analogically restructuring a weakly marked alternation (cf. Cuervo’s analysis in Bello 1964, MeyerLübke 1895, Malkiel 1969).

54 In this sense, the Spanish situation contrasts with the Italian one which shows a shift from palatal to velar, e.g. OIt. 1s tegno >> MIt. 1s tengo . See especially Malkiel 1974, Maiden 1992, 2001, 2003, and Fanciullo 1998.

55 See §2.2 and App. 2F.

56 This synchronic stage is the one analyzed in the vast majority of the literature because it is the point of departure required to explain the extension of the velarinsert diachronically. A highly detailed analysis of the extension of the velar beyond its inception is a central part of chapter 3, as it has been lacking in previous research.

57 A phonological description of these verb sets is found in §2.3.

58 The ‘inchoative’ subclass of verbs in Spanish does not have an inchoative function or meaning. The term refers to its historical role in Latin and represents the group of verbs which are marked by a voiceless velar /k/ in the 1s/SBJ, see Figure 2.9. For the development of this class, see Lloyd 1987. The same hypothesis is considered, secondarily, in Martín Vegas 2007 and Pérez Saldanya 1995a.

59 The primary motivation for grouping these classes of verbs together is to explain the variant 1s/SBJ stems of the verb yacer ‘lie (down)’ – 1s yazco ~ yazgo ~ yago . The last form coincides with the conjugation of the verb hacer and is thought by some (e.g. Martín Vegas 2007), to be the expected outcome through very early loss of the yod. The first one is analogically based on the inchoative set, e.g. conocer – 1s conozco . The middle variant, yazgo , is less clearly motivated. Per the data provided in CORDE (Real Academia Española 2005), this form appears very late (XIX c.) and could be due to a combined influence of its etymological form, the velarinsert class and the inchoative set, or it may have to do with an assimilation of voicing at the very early stages of Old Spanish when the grapheme < z> typically represented a voiced segment.

60 It is important to note that the use of valgo in the formalization of the constraint is merely a convenience. The actual shape of the Input potentially requires a more complicated explanation than is able to be explained here. See the subsequent paragraphs for the importance of formalizing the Input. 292

61 For more elaborate cases involving paradigms as output candidates, see Bachrach & Nevins 2008, Burzio 2005, Downing et al. 2005, Kenstowicz 2000, KraskaSzlenk 2007, and McCarthy 2003.

62 The proposed AM includes OSp. 1s digo ‘say’, fago ‘do’, yago ‘lie (down)’, cuelgo ‘hang’, and sigo ‘follow’ among other verbs.

63 Another reason for refuting the avoidance of homophony argument is that the proposed forms of these verbs are, in fact, not homophonous as the verbs having a palatal nasal in their stem undergo raising of their stem vowel due to yod effects, cf. 1s tengo ~ tiño , pongo ~ puño .

64 Alvar & Pottier 1983 excludes asir from the velarinsert set due to its greater phonetic similarity to other available AM. For further discussion of this verb’s exceptional status, see §3.5.6.

65 Apparently, for * teno and * veno , the insertion was therapeutic (a repair strategy) because the stem vowel had not diphthongized, "having created an imbalance in the paradigm" (Lenfest 1978:900, cf. verbs with 1s stemchange, e.g. siento (<< * sento ), alongside sientes , siente ). The repair strategy is based on phonetic weight (CVCCV ≈ CVVCV, cntr. CVCV, Lenfest 1978). “The set tengo , tienes …fell within the limits of paradigmatic tolerance which excluded the set * teno , tienes …Because gross phonetic substance, rather than the particular sequence of phonemes, was the source of uneasiness, it made no difference that the semivowel of the 2 nd pers. diphthong was balanced by a velar in the 1 st pers. sing. pres. indic.” (Lenfest 1978:901).

66 “This reanalysis [of /K/ as part of the inflectional marker for 1s/SBJ] would be the focus of diffusion in an inflectional model that is characterized by having the marker –go in the 1s PRS IND and –ga , gas in the PRS SBJ…The schema has relative lexical strength because it initially affects very frequent verbs…Some of the lesserused verbs that are part of this schema end up leveling the alternation: e.g. cingo  ciño . But the alternation is maintained in such frequent verbs as digo /dices , pongo /pones , salgo /sales …With an analysis of this type we wouldn’t speak of an insertion of –g in certain verbs, but rather of the extension of a schema/model of inflectional that is characterized because the 1s PRS IND and PRS SBJ end in –go , ga , gas …”

67 Any terms used here related to velarinsertion are not taken directly from Elson 1988 where they have been intentionally avoided. It should be pointed out that the term velarinsert is not applicable in the sense used in the current investigation given the inclusion of verbs with etymological velar allomorphs, i.e. contexts in which the velar element is not ‘inserted’ per se.

293

68 The Upattern described in Maiden 2001 equates to the 1s/3p/SBJ distribution of the velarinsert in Italian, resembling the letter turned on its side. A variant Npattern is also found in Italian in which the 1p/2p SBJ forms do not contain a velar. The Lpattern, is described in an earlier note (cf. Maiden & O’Neill 2008).

69 An interesting and unintentionally similar analysis applies to a verbal subclass of the Anglian dialect of examined in Hogg 1980.

70 Of the verbs included in the velarinsert group through the end of the 13 th c., only the final stage includes verbs having a high stem vowel, e.g. huir . In other words, the earliest AM further limits the preceding phonological structure in that the stem vowel need be [high].

71 It is no coincidence that nasal consonants are at the center of the Sonority Hierarchy (Dell & Elmedlaoui 1986). This permits analogical spread of the velarinsert to verbs having both more and less sonorous stemfinal segments, especially when considered in conjunction with the phonotactic constraints on the syllable in Spanish.

72 Although the data includes forms having a syllableinitially ( traer ), no initial consonant ( oír ), ( fiergo ) and stems ending in a yod ( caigo ), the general phonological shape of the stem holds true as these apparent exceptions fall within the expected phonological constraints on the Spanish syllable. Even more broadly, the implication here is that the nonderived stem is closed (or offgliding) and monosyllabic. Excluded are remanir and definir because their derivational sources (i.e. manir ‘remain’ and finir ‘finish; end’) do not show a velar.

73 It is possible to argue that there is a requirement that the stem be closed or at least decreasing in sonority, i.e. CVC, where the final element is simply nonsyllabic. This implies no vowelplusvelar combinations (cf. §4.5.4).

74 A counterargument that identifies these early instances of g as representative of the velar occlusive /g/ is established in Lenfest 1978. It is suggested that these cases are all examples of the antihiatic use of /g, k/ as opposed to an orthographic misrepresentation of a palatal fricative as g. The attested 3s SBJ form poncat is, for example, difficult to explain without appealing to velarinsertion/inchoative productivity, if it is not viewed, as in this investigation, as a hypercorrect reconstruction of an earlier stage.

75 Note the potential involvement of diphthongization in the outcomes affecting these pairs of verbs. Ongliding diphthongs are not associated with the standard modern velarinsert set, though they do appear dialectally (cf. Penny 1969).

294

76 The identification of nasals as [+cont] is controversial (Kenstowicz 1994). However, for present purposes, [cont] refers to the uninterrupted flow of air through either the oral or nasal cavity.

77 Wilkinson 1978 indicates that the verb parir ‘bear, give birth’ shows velarinsertion in modern dialects, though no such forms are attested in the CORDE. Nonetheless, the suggestion that this verb is affected by velarinsertion is indicative of the nature and direction of analogical change (cf. §§3.6 and 4.5).

78 Urrutia Cardenas & Alvarez Alvarez 1983 suggests first that the yod raises (or closes) the stem vowel (with velarinsertion, e.g. unattested 1s * firgo ) which is then diphthongized due to intraparadigmatic analogical effects (providing attested fiergo ). The hypothesis that this yod is lost prior to exerting any such effect and the absence of any attested forms with a high vowel permit the stance indicated above (cf. §2.3).

79 It is more in line with the current investigation to assume that this verb is reanalyzed via interparadigmatic analogy with the very large AM set of rhoticfinal diphthongized verbs not having a velarinsert, e.g. querer ‘want’, 3p SBJ quieran ; morir ‘die’, 3p SBJ mueran , including the intraparadigmatic influence of the nonvelar forms as indicated above.

80 For example, during the 14 th c., the 3s IND form of valer appears twice as often as that of soler (410 cases in 81 documents against 204 cases in 62 documents respectively). Diachronically, Davies 2006 ranks valer 387 th and soler at 487 th in weighted frequency.

81 Based on the first attested examples of INF and 3s.

82 Preliminary research indicates that this verb is characterized by a completely regular, nonvelar stem, i.e. 1s aso , 1s/3s SBJ asa . This exemplifies the robustness of analogy in AEM (cf. Skousen 2002).

83 Motivation for the incorporation of the inchoative verbs in the AM come from examples of the variation of /g/ and /k/ in forms like 1s asgo ~ asco , conozgo ~ conozco , and the analogical extension of the inchoative class, e.g. yacer , 1s yago ~ yazco ~ yazgo (id. the –ducir verb set). The variation found in the verb asir , e.g. 1s asgo ~ asco ~ aso , is seen as an ‘experimental innovation’ because of its rootfinal sibilant which, to a certain degree, represents a new direction for the velarinsertion process (Malkiel 1989). The first two variants are related to the “longdrawnout wavering between (etymological) –sco , sca and (analogical) –zco , zca in the pres. ind. and subj. of inchoative verbs” and the aforementioned similarities between the velarinsert and the inchoative (Malkiel 1974:338). The latter form, meanwhile, represents a regularized/leveled analysis.

295

84 Malkiel 1967 provides a note about additional data which demonstrated the extension of velarinsertion in some Italian dialects: “On the Berkeley campus R. Stefanini has drawn my attention to several Italian parallels and nearparallels. Thus, such northeastern dialects as Veronese and Trentino exemplify the spread of the voiced velar from digo, ga, ‘I say, may say,’ producing fago, ga, vago, ga, dago, ga, stago, ga, which match Stand. It. faccio (fo) ‘I make,’ vo ‘I go,’ do ‘I give,’ sto ‘I stand,’ and the corresponding sets of pres. subj. forms…” (64).

85 Note that the use of the terms ‘ LAW ’ is not meant to imply that this generalization is systematic nor without exception. It simply follows the conventional notations as employed in Andersen 1973.

86 “The verb is subject to the same phonetic laws as any other word. But it is understood that the multiple forms that express the same stem in conjugation, being tightly connected to each other by the essential unity of meaning, cannot stop influencing one another more often than two words that differ in their source; such that the power of analogy…is much more active in conjugation than in any other part of the grammatical domain, and we continually see forms that distort their phonetic development in order to follow an analogy alongside others from the same conjugational paradigm. The speaker…tends to see the stem as invariable, since the idea it expresses is invariable; and if at times the language preserves intact the phonetic alternations of a stem…other times it seeks uniformity”

87 The verbs included in the AM may be palatalfinal but it is possible to argue that such a variable need not be included if the appearance of an isomorphic stem throughout the entire inflectional paradigm is considered a ‘morphome’ like the 1s/SBJ.

88 The preference of the 1s/SBJ velarinsert forms over palatalized ones implies a type of intraparadigmatic effect or leveling in the sense that the verb stem maintains a nonpalatal nasal throughout (cf. Spaulding 1971, Ford 1966, Lenfest 1978).

89 Another reason for the distinction of the first conjugation is the lack of any salient phonological changes conditioned by the nonfront vowel /a/ which is the theme vowel for this conjugational class. This contrasts with the front vowels /e, i/ which represent the second and third conjugations respectively and are essential in triggering numerous sound changes like the palatalization effects examined in chapter 2.

90 Other perspectives on the function of the velar element in the paradigm which resemble the template, to greater or lesser degrees, are found in various investigations, e.g. Alvar & Pottier 1983, Elson 1988, Fanciullo 1998, Lloyd 1987, Lenfest 1978, Pérez Saldanya 1995a.

296

91 Although the term ‘morphome’ has been used interchangeably with ‘template’ up to this point, the former, as defined in Maiden 2001, following Aronoff 1994, does not traditionally correspond exactly to the latter. The difference lies in the fact that 1s/SBJ distribution of the velarinsert is, on the surface, completely defined by its phonological conditioning environment (cf. §2.1.2, KraskaSzlenk 2007).

92 It is also arguable that verbs like dar , estar and haber could be included in this group because the 1s/SBJ forms are distinguished, though not in the same way, from the remainder of the PRS, e.g. 1s doy , estoy , he , 3s SBJ dé , esté , haya , cntr. 2s IND das , estás , has .

93 Similarly, the influence of homophonous nonverbs in the AM is not possible as it would require elimination of a number of relevant variables. This is supported by the fact that velarinsertion does not seem to be attested outside the verbal paradigm.

94 Where clarification of the distinction between word frequency and lexeme frequency is indicated.

95 Based on the approximately 5,000 most frequent lexemes in the corpus dataset.

96 Consider the effect of interparadigmatic (or lexemic) frequency in the survival of the 1s/SBJ pattern in verbs like decir and hacer against the paradigmatic leveling of lower frequency verbs like tañer and teñir , as discussed in Martín Vegas 2007.

97 These three lowfrequency verbs are the only standard velarinserts not among the 5,000 most frequent lexemes as calculated in Davies 2006 and Juilland & ChangRodríguez 1964.

98 The actual patterns across the entire inflectional paradigm are numerous. In the PRS tense alone, there are at least three types of templatic patterns, not including the systematic use of a single stem. The point is that allomorphy is a distinctive characteristic of these verb classes which substantiates their unified analysis against the first conjugation (cf. §2.1.1).

99 Based on the top 5,000 most frequent lexemes in the corpus dataset.

100 Included here are dar, estar, ir, haber, ser .

101 In fact, the only nasalfinal second or third conjugations verbs included in the data are homogeneous in this respect, i.e. they all show velarinsertion in the 1s/SBJ. All nonvelar examples have a low lexemic frequency.

297

102 This is based on the square of the frequencies. 122 = 144, 24 2 = 576; Probability of velar 144/720 = 1/5; Probability of nonvelar 576/720 = 4/5. This method is contrasted with ‘selection by plurality’ which would (undesirably) result in the nonvelar being chosen in all such cases (cf. §1.3.3).

103 “In this way, type frequency can be the motive for the extension of an alternation and, in other cases, token frequency can be the motive for the stability of an alternation.”

104 See Maiden 1992 for similar arguments concerning Italian.

105 Based on the approximately 5,000 most frequent lexemes in the corpus dataset.

106 An example of problems associated with such arguments is based on token frequency. According to rough numbers taken from the CREA database, the 1s salgo appears approximately 1000 times against nearly 4000 tokens of the SBJ. Notably, the 1s/3s SBJ alone accounts for over half of all examples of the SBJ (Real Academia Española 2007).

107 Wang 1969 elaborates on the types of competing changes that occur in time to include successive/complementary changes and intersecting/overlapping changes. For present purposes, both are assumed under the same term.

108 For verbs like creer , it is assumed that velarinsertion coincides with a yodfinal stem, as observed in several dialectal examples like 3s SBJ creiga .

109 This point also has a phonological basis as the intervocalic velar in 1s dico is not subject to voicing (via lenition) as in Spanish. Similarly, the presence of yod and its palatalizing effect eliminate the possibility of the velar appearing in forms like 1s faccio (cf. §2.2.3).

110 An even more extreme example of this gradient contrast is that of Portuguese in which many examples of paradigmatic allomorphy are distinguished by a single distinctive feature, e.g. 1s te o ~ 2s te ns.

111 Some of the verbs have a vowelfinal stem. In these cases, the velar would be preceded by a glide as there are practically no attested cases of a velarinsert immediately following a vowel. This is indicative of an additional variable of the AM, namely that the syllable preceding the velar be closed or at least decreasing in sonority. In the extremely few exceptional cases where a stemfinal vowel does appear, it calls into question the pronunciation of the velar (cf. §3.4).

112 Another possibility, however, is the extension of the velarinsert to nonalternating stems in the first conjugation. An example of this sort is mentioned in §3.6.3, where the verbs traer and poner are shifted to the first conjugation as traigar and pongar in the 298

language of some Spanishspeaking children (Clark 1986). Rather than attributing such reanalyzed forms to velarinsertion, they should rather be associated with the analogical influence of the leveled patterning typical of the highly productive first conjugation on a single stem alternant.

113 The nonstandard, dialectally attested forms in this Appendix are taken from the CREA database (Real Academia Española 2007). In contrast, those forms that are marked as attested in Appendix K are from the CORDE database (Real Academia Española 2005). In sum, the data is distinguished by chronology, i.e. those forms occurring in the most modern language data against those which primarily predate the 20 th c. respectively.

114 This verb is the only to show velarinsertion exclusively that is not used in the modern language (Real Academia Española 2004). The attested forms are taken from the CORDE database (Real Academia Española 2005).

115 The numbers concerning lexemic frequency are taken from Davies (2006) and Juilland & ChangRodríguez (1964). See also comments regarding Appendix B.1.

116 The verb caber may be seen as suppletive or irregular, depending on perspective. In the modern language, it is unlikely that the typical speaker identifies an etymological relationship between the two stem allomorphs.

117 Thanks to Terrell Morgan for providing an realworld example of 1s sepo ‘taste’ which means that it corresponds, at least at the level of individual speakers, to the example provided by caber .

299