A Constraint-Based Account of Handshape Contrast in Sign

A CONSTRAINT-BASED ACCOUNT OF HANDSHAPE CONTRAST IN SIGN

LANGUAGES

A Dissertation

Submitted to the Faculty

Purdue University

Petra Nichole Eccarius

In Partial Fulﬁllment of the

Requirements for the Degree

Doctor of Philosophy

May 2008

Purdue University

West Lafayette, Indiana ii

To my mother, who taught me how to write. iii

ACKNOWLEDGMENTS

There is no way to adequately thank all of the people who have contributed to the creation of this document in the space provided here. For all of you not specifically named, I will strive to thank you in person. Particular thanks goes to my dissertation committee: to Alex Francis and Karen Emmorey for their patience and guidance as I ventured into the realm of psycholinguistic experimentation; to Ronnie Wilbur, for suggesting (during those first over- whelming weeks of graduate school) that I should join her research lab, and for acting as a mentor ever since; and especially to Diane Brentari, my chair, for her patience, guidance and relentless encouragement throughout every new opportunity—thank you for figuring out the best ways to keep me motivated. I am also grateful to all of the agencies who funded me in this endeavor. Parts of this work were supported by grants from the National Science Foundation (BCS- 0112391 and BCS-0547554 to Diane Brentari and 0414953 to Ronnie Wilbur), as well as a National Institutes of Health Training Grant (T32DC00030) and a dissertation grant from the Purdue Research Foundation. Thank you also to the sign language researchers and their assistants in Switzerland and Hong Kong for helping to make this a cross-linguistic study. Special thanks goes to Penny Boyes Braem, Gladys Tang, Felix Sze and Wong Yuet On—thank you for answering so many questions from so far away. A big thank you also goes to all of the members of the Deaf communities in Indiana, at Gallaudet, in Switzerland, and in Hong Kong, who contributed to this work by providing data, helping me in the collection process, or by just answering my incessant questions. I would especially like to thank Rita Mowl, Molly O’Hara, Drucilla Ronchen, Christa Notter, Sandra Sidler-Miserez, Brigitte Daiss- Klaug, Thomas Schmidt, Doris Hermann, Chu Kwan Ngai, Chu Hoi Yan, Tse Wing iv

Man and Lam Tung Wah for their linguistic insights; Jill Lestina for her patience and good humor as she signed my stimulus items over and over and over again; Kristi Win- ter for signing my instructions and being curious enough to help with data collection; and Robin Shay for answering all of my last minute questions with a smile. I also owe a debt of gratitude to Ruta Bajorunaite and Bruce Craig for all of their help with my statistical analysis; and to my fellow lab members for their assistance and support—especially Carolina Gonzalez and Donovan Grose for keeping me sane, Marie Nadolske for pushing me to sign more, and Christina Giﬀord for her fabulous work ethic and all of those hours at the computer. And for all of the friends who supported me (and even traveled great distances to cheer me on) in this pursuit that they didn’t always understand—thank you as well. A special thank you goes to Noemi for housing me, encouraging me, and providing me with cats to play with on my frequent trips to West Lafayette; to all of the folks at Purdue Lutheran Ministries for helping me keep my priorities straight; to Eirinn, Pieter and the rest of the madragali for giving me a musical escape during my time at Purdue; and to Chap and Jon for (among many other things) being crazy enough to want to read (and even proofread) this massive document so tangential to their own interests. Finally, I would like to thank my family, who continued to love me even at my most stressed. I am especially grateful to my mother who has always supported me in my academic pursuits (as in life) even in the most uncertain of times; and to Dennis, for sustaining me—physically and spiritually—through this and all of life’s challenges. v

TABLE OF CONTENTS

Page LIST OF FIGURES ...... viii ABBREVIATIONS ...... x ABSTRACT ...... xi 1 INTRODUCTION ...... 1 2 BACKGROUND AND LITERATURE REVIEW ...... 3 2.1 Handshape ...... 3 2.1.1 Phonological structure of handshape ...... 4 2.1.2 Structural scope of this research ...... 6 2.2 Contrast types ...... 6 2.3 Lexical stratiﬁcation ...... 10 2.3.1 Lexical components for sign languages ...... 11 2.3.2 Phonological behavior across the lexicon ...... 15 2.4 Iconicity ...... 18 2.4.1 Iconicity debate: All or nothing ...... 19 2.4.2 Iconicity debate: Middle ground ...... 21 2.5 Cross-linguistic comparisons ...... 22 2.6 Methodological approach ...... 23 3 INTERNAL EVIDENCE: DICTIONARY AND ELICITED DATA .... 24 3.1 General methodology ...... 25 3.2 Utilization of [stacked] joints ...... 26 3.2.1 Language-internal distributions ...... 27 3.2.2 Contrast types ...... 29 3.3 Utilization of three selected ﬁngers ...... 31 3.3.1 Language-internal distributions ...... 32 vi

Page 3.3.2 Contrast types ...... 37 3.4 Summary of [stacked] and 3FHS contrast differences ...... 38 3.5 Handshape contrast across classifier types ...... 39 3.5.1 Classifier types ...... 39 3.5.2 Iconic representations via phonological features ...... 42 3.5.3 Feature asymmetries across classifier types ...... 52 4 EXTERNAL EVIDENCE: EXPERIMENTAL DATA ...... 59 4.1 Background ...... 60 4.2 Methodology ...... 62 4.2.1 Stimuli ...... 62 4.2.2 Design ...... 66 4.2.3 Participants ...... 68 4.2.4 Procedure ...... 68 4.3 Results ...... 69 4.3.1 O handshape group ...... 69 4.3.2 F handshape group ...... 71 4.3.3 C handshape group ...... 73 4.4 Discussion ...... 73 4.4.1 The minimal pair problem revisited ...... 77 4.4.2 Predictions for language change ...... 79 4.4.3 Future experiments ...... 81 4.4.4 Theoretical implications ...... 82 5 OT ANALYSIS OF CONTRAST ...... 84 5.1 Background ...... 84 5.1.1 OT in general ...... 84 5.1.2 OT and contrast ...... 85 5.1.3 OT and lexical stratification ...... 86 5.1.4 OT and iconicity? ...... 87 vii

Page 5.1.5 Challenges to the analysis (and disclaimer) ...... 92 5.2 Constraints ...... 93 5.2.1 Markedness constraints ...... 93 5.2.2 Faithfulness constraints ...... 99 5.3 Synchronic analyses ...... 112 5.3.1 Selected Fingers ...... 112 5.3.2 Joints ...... 137 5.4 Diachronic Analyses ...... 142 5.4.1 Selected ﬁngers ...... 142 5.4.2 Joints ...... 144 5.4.3 Iconicity in core forms ...... 146 5.5 Summary ...... 147 6 CONCLUSION ...... 150 LIST OF REFERENCES ...... 153 A EXPERIMENTAL STIMULUS ITEMS ...... 160 B LIST OF PROPOSED CONSTRAINTS ...... 162 VITA ...... 166 viii

LIST OF FIGURES

Figure Page 2.1 The phonological structure for handshape (Brentari, 1998)...... 5 2.2 Examples of joint configurations from Brentari (1998)...... 5 2.3 Types of phonological contrast (following Clements, 2001)...... 9 2.4 Stratification of the ASL lexicon (Brentari and Padden, 2001). ... 11 2.5 Examples of core lexical items...... 12 2.6 Examples of fingerspelled letters (top) and initialized signs utilizing them (bottom)...... 14 2.7 Examples of classifier constructions...... 15 2.8 Example handshape asymmetries across ASL lexical components (adapted from Brentari and Padden, 2001)...... 16 3.1 Examples of plain and stacked handshapes...... 26 3.2 Stimuli used to elicit stacked handshapes (Zwitserlood, 2003). ... 27 3.3 The three-legged boy (left) and handshapes used to represent him (right)...... 33 3.4 Examples of three-finger handshapes (3FHSs) in the foreign, core and classifier components for ASL, DSGS and HKSL...... 36 3.5 The phonological structure for handshape (Brentari, 1998)...... 43 3.6 Size representations...... 44 3.7 WCLs representing whole object shape...... 45 3.8 Different sizes of similarly shaped objects via degree of base and nonbase joint flexion...... 46 3.9 Fingertip arrangements depicting round and flat RP shapes (top) and perimeter classifiers combining these RP shapes with a curved surface shape (bottom)...... 47 3.10 DCL and HCL orderings for size and shape combinations...... 49 3.11 Power vs. precision grips (based on Napier 1956)...... 51 ix

Figure Page 3.12 Example handshapes from the four proposed classifier types. .... 55 3.13 Continuum for joint and selected finger possibilities in the four types of classifier handshapes...... 58 4.1 Handshape groups tested...... 63 4.2 Examples of stimulus items for the O handshape group...... 64 4.3 Example presentation slides for each task...... 67 4.4 O handshape group results...... 70 4.5 F handshape group results...... 72 4.6 C handshape group results (those with minimal pairs only). .... 74 4.7 Summary of contrasts across lexical components...... 77 4.8 Average ratings in Form-Rating Task for core items without minimal pairs for all three handshape groups...... 78 5.1 Differences in the conceptualization of ‘round’ within an F handshape (left) and in the shape of a coin (right)...... 91 5.2 Faithfulness constraints in large and small scale analyses...... 103 5.3 Available iconic relationships by classifier type...... 105 5.4 Shape distinctions in terms of symmetry and angles ...... 109 5.5 Examples of character signs...... 122 x

ABBREVIATIONS

3FHS Three-finger handshape ASL American Sign Language CP Categorical Perception DASL Dictionary of American Sign Language (Stokoe et al., 1965) DCL Descriptive Classifier DSGS Swiss German Sign Language (Deutschschweizerische Gebärden- sprache) HCL Handling Classifier HKSL Hong Kong Sign Language HS X Experimental handshape variant (e.g. HS 1, HS 2) OT Optimality Theory RP Reference Point (with respect to size and shape) SCL Semantic Classifier SF Selected fingers WCL Whole Classifier xi

ABSTRACT

Eccarius, Petra N. Ph.D., Purdue University, May, 2008. A constraint-based account of handshape contrast in sign languages. Major Professor: Diane K. Brentari.

The main goal of this dissertation is to explore the nature of contrast in sign language handshapes. I first demonstrate that the distribution of handshape contrasts is not homogeneous, either within or across sign languages. By using a variety of methodologies (examination of dictionary data, elicited data, and psycholinguistic experimentation), I present examples of differences related to type of contrast (distinctive, active, and prominent—following Clements, 2001); position in the lexical substrata (following the work of Ito and Mester, 1995a, and Brentari and Padden, 2001); iconic relationships (e.g. shape, size, arrangement of parts); and cross-linguistic variation (comparing American Sign Language, Swiss German Sign Language, and Hong Kong Sign Language). I also propose that the distributional differences in handshape contrasts can be explained in terms of a confluence of pressures on language. Using the tenets of Optimality Theory (OT), these differences can be explained by determining how various languages—or lexical components within languages—rank constraints related to those pressures. Specifically, I follow Flemming’s (2002) version of OT (Dispersion Theory) in which grammars balance the pressures of articulatory ease and perceptual distinctiveness, as well as the desire to maximize the number of contrasts available for word formation. To this, I propose an additional pressure—one to maintain contrasts borrowed into the language from external sources. These external contrasts can be borrowed from other languages (directly from other sign languages, or indirectly from spoken languages via systems such as fingerspelling), or they can be borrowed from visual aspects of the real world. 1

1. INTRODUCTION

The main goal of this thesis is to explore the nature of contrast in sign language handshapes. To this end, I put forth two main hypotheses: The ﬁrst hypothesis is that the distribution of handshape contrasts is not homogeneous, either within or across sign languages. I explore this issue by:

1. examining diﬀerent types of possible contrasts (distinctive, active and prominent— following Clements, 2001) as they relate to handshape,

2. looking at the distribution of handshape contrasts in diﬀerent parts of the lexicon (following the work of Ito and Mester, 1995a, and Brentari and Padden, 2001),

3. considering the role that iconic relationships between phonological form and the visual aspects of referents play in the determination of handshape contrast, and

4. looking at how all of these factors aﬀect the distribution of handshape contrast cross-linguistically.

I also explore handshape contrasts using a variety of methodologies. These include examining language-internal evidence—dictionary entries, elicited descriptions of pictures and stories, and interview questions about phonological variation—as well as language-external evidence via the results from a psycholinguistic experiment. The second hypothesis is that the distributions of handshape contrasts can be explained in terms of a confluence of pressures on language, and that, using the tenets of Optimality Theory (OT), the differences in those distributions can be explained as differences in how various languages—or lexical components within languages—rank constraints related to those pressures. Specifically, I follow Flemming’s (2002) version 2 of OT, Dispersion Theory, in which grammars balance the pressures of articulatory ease and perceptual distinctiveness, as well as the desire to maximize the number of contrasts available for word formation. To this, I propose an additional pressure— a pressure to maintain contrasts borrowed into the language from external sources. These external contrasts can be borrowed from other languages (directly from other sign languages, or indirectly from spoken languages via systems such as fingerspelling), or they can be borrowed from visual aspects of the real world. The dissertation is organized as follows: Chapter 2 provides background and a review of the literature relevant to the issues at hand; Chapters 3 and 4 show that differences in both the distribution and type of contrasts for handshape features do exist within and across languages based both on language internal evidence (Chap- ter 3) and language external evidence (Chapter 4). Chapter 5 then unifies these differences into a single theoretical approach using Optimality Theory. 3

2. BACKGROUND AND LITERATURE REVIEW

Of the five parameters of sign language structure—handshape, movement, place of articulation, orientation and nonmanual behaviors—handshape is the parameter that is the best understood from a phonological point of view; of these five, it is the one that has been analyzed most successfully using both theoretical and experimental methodologies. Despite the large body of literature relative to other parameters, however, we know comparatively little about handshape in terms of contrast. The goal of this thesis is to contribute to this growing body of knowledge by exploring the nature of contrast in handshape, both within a single language and cross-linguistically. Here I explore three aspects of contrast not usually addressed: 1) differences in contrast type, 2) variations in contrast cross-componentially within the lexicon, and 3) the relationship between contrast and iconicity. In this chapter, I begin by providing basic information about the phonological structure of handshape itself, and then continue with a review of literature in each area.1

2.1 Handshape

Before addressing the larger topics of this work, it is important to provide some background about the parameter of handshape itself. Three main categories of handshape features have been established in the literature, (see Brentari, 1998, and references therein for evidence): 1) selected fingers, 2) joint configuration and 3) aperture. Generally speaking, (depending on orientation and other physical limitations), selected fingers (hereafter, SF) are the fingers that move and/or make contact with another part of the body during the articulation of a sign (Mandel, 1981); joint configuration is the extent to which the joints of the

1Parts of Chapters 2 and 3 are taken from Brentari and Eccarius (in press). 4

fingers are extended or flexed in a handshape; and aperture concerns the open or closed position of the hand, usually as it pertains to local movement (i.e. handshape change) during the articulation of a sign. This dissertation focuses on the first two of these feature types and leaves the third for future research.

2.1.1 Phonological structure of handshape

While there is reasonably wide consensus about the basic structure of handshape among various models of sign language phonology (e.g. Brentari, 1998; Corina, 1993; van der Hulst, 1995; Sandler, 1989, 1996), the model used here (when one is necessary) is Brentari (1998), as shown in Figure 2.1.2 This model was chosen over others because it was designed to represent a variety of forms, including those outside of the frozen lexicon (e.g. classifier constructions and fingerspelled forms, see Section 2.3); consequently, Brentari’s model is able to represent a wider range of handshapes than the others available. The two main structures for handshape in Brentari’s feature geometry are the class nodes Joints and Fingers, representing the Selected Fingers’ joint configuration and finger combination respectively. The Joints structure expresses the disposition of the joints ([stacked], [flexed], [crossed] or [spread]) as well as which joints are affected by the feature—the joints at the fingers’ base (metacarpophalangeal joints) or the nonbase joints (the proximal and distal interphalangeal joints). Examples of the joint configurations used throughout this work (i.e. the possible combinations of Joint features according to Brentari) are illustrated in Figure 2.2.3 The Fingers structure under the Selected Fingers node expresses which fingers are foregrounded, or important, in the handshape. The number of fingers is expressed by the Quantity features ([all] and [one]), and part of the hand where the SF are located

2This simplified version of Brentari’s handshape structure fits into the larger structure of the Articulator, which in turn fits into an even larger structure representing the sign’s place of articulation and movement. Since none of the other structures are relevant to this analysis, the only structure introduced here will be that of handshape. 3See Brentari (1998) for structural representations of each configuration. 5

Figure 2.1.: The phonological structure for handshape (Brentari, 1998).

Figure 2.2.: Examples of joint configurations from Brentari (1998). 6 is expressed by the Point of Reference features (a default value of radial for the index side, [ulnar] for the pinkie side, or [mid] for the middle). For example, , and all have a quantity specification of [one] but with different points of reference on the hand—radial, [ulnar] and [mid], respectively.4 When additional information is needed concerning particular features, it will be provided in later sections.

2.1.2 Structural scope of this research

Because the overall scope of this research is already quite extensive, there are many aspects of handshape that will not be addressed here. One such example is movement local to the hand; here I focus only on static handshapes.5 Given the importance of aperture as a feature type, the choice not to include handshape movement should be justified. According to an informal count of the Dictionary of American Sign Language (Stokoe et al., 1965, henceforth DASL) using the tables made by Hara (2003), only about 16% of the approximately 2000 dictionary entries involve a handshape change of some sort. Given this estimation, it is clear that even by limiting the discussion to static handshapes, my analysis will still be based on the vast majority of signs. The number of dynamic handshapes is by no means insignificant, however, and contrast differences in these handshapes (if they exist) need to be investigated in future work.

2.2 Contrast types

The phonological features used to represent signs are those that show ‘contrast’, but what does this term mean? The notion of contrast and how it contributes to our understanding of phonological units has always been extremely important in phonological theory. Over the decades, phonologists have oscillated between deﬁning these units according to language- and context-particular behavior dependent on minimal

4My thanks to Gladys Tang for the use of her handshape font for this work; additions of my own were added as necessary for later sections. 5I also do not address orientation diﬀerences here. 7

pairs, and deﬁning them universally by assigning them all the same features regardless of context and language. Both extremes have their problems, and consequently, most theories fall between the two, although the exact proportions of each are still being debated. (See Dresher et al., 1994 for an overview of these issues.) In his introduction to The Handbook of Phonological Theory (1995), Goldsmith summarizes an underlying problem in this area—contrast is not a homogeneous phenomenon. He lists ﬁve types of contrast between segments depending on the amount of force exerted on a particular feature (F) by the phonological system.

1. Contrastive segments: no pressure (synchronic or diachronic) toward a preference for one feature value over another

2. Modest asymmetry: a few asymmetries exist, but otherwise the two F values are unrelated

3. Not-yet-integrated semi-contrast: a contrast exists, but one on the outer limits of permissible behavior with a strong tendency toward pushing the one feature value to the other to conform to native linguistic tendencies

4. Just barely contrastive: contrast only exists in a small phonological and/or lexical class but otherwise is controlled by the grammar

5. Allophones in complementary distribution: the grammar is in complete control of F’s distribution (i.e. nothing is assigned to the lexical item)

Traditionally, the first four of these types have tended to be considered in the realm of ‘lexical contrast’ while the last one (allophonic distribution) is set apart as grammatical. The notion of contrast is a topic that has recently been re-animated in the literature on phonological theory (see Hall, 2001), recognizing that only a few of the possible grammatical identities for feature contrasts have been explored in any detail. Clements (2001) is one such contributor working toward the ideal of economy in the 8 theoretical representation of contrast.6 He argues that in order to represent features only when needed, researchers must first recognize that all features are not created equal in the eyes of the phonological representation—a feature’s status changes depending on context, behavior in phonological processes, language-particular factors, etc. This issue is particularly important for sign languages because their phonological representations are quite complex; if certain features or feature structures do not need to be represented, or do not need to be represented all of the time, it would certainly be a welcome innovation. Clements distinguishes between three main types of contrast (Figure 2.3). Using his terminology, a feature is distinctive when its presence results in a minimal pair, active when it is involved in a phonological process, and prominent (a subtype of active) when it is used to establish an autosegmental tier (Goldsmith, 1976), that is, the feature is either used for morphological purposes or involved in a particular type of phonological operation (e.g. spreading).7 In spoken languages, features involved in vowel height and the obstruent series of American English can be used to illustrate these three contrast types. A contrast between [high] and [low] in non-tense front vowels ([I] vs. [æ]) can be distinctive (e.g. ‘p[I]t’ vs. ‘p[æ]t’), but it can also be prominent, indicating differences in the tense and aspect of certain verbs (e.g. ‘s[I]ng’ vs. ‘s[æ]ng’). Meanwhile, the feature [continuant], which creates a corresponding fricative vs. stop pair (e.g. [s] vs. [t]), is also distinctive (e.g. ‘[s]ue’ vs. ‘[t]wo’), but can be active as well in dialects of American English that pronounce ‘tree’ as ‘[tS]ree’, where the addition of the feature is involved in the affrication of the [t]. In sign language research, the criteria used by Brentari (1998) for inclusion of a phonological structure or feature in her theoretical hierarchy are quite similar to

6See Dresher (2003) for an additional discussion of the problems of determining feature contrast in inventories and a review of attempts to characterize the scope of contrast using hierarchical structures. 7The other two criteria for establishing an autosegmental tier are long distance eﬀects, which have not been observed in sign languages, and many to one association, which is observed almost everywhere throughout the system. Consequently, neither is particularly useful for the analysis here because they would not reveal anything new. 9

Figure 2.3.: Types of phonological contrast (following Clements, 2001).

the contrast types set forth by Clements. These criteria were 1) use in a minimal pair (i.e. distinctive), 2) use in a phonological constraint (i.e. active), or 3) use morphologically in a productive way (i.e. prominent). All three contrast types are found in sign languages, but because these languages have so few minimal pairs, the first, or ‘traditional’ definition of contrast—distinctive contrast—has never been as useful to sign language researchers as the other two. The handshape feature [stacked] is a good example of a feature used in all three types of contrast, as will be shown in Section 3.2. Further discussion of contrast from a theoretical point of view is beyond the scope of this dissertation. My main point here is simply that featural contrast itself is not an easily defined concept in phonology and changes with respect to a number of factors. As with spoken languages, when analyzing sign languages we must keep this in mind, remembering that the problem of defining phonological units is a complex one. In this 10 work, when making observations about the contrast variation of handshape, I follow the work of Brentari and of Clements, using the terminology set forth by Clements (2001).

2.3 Lexical stratiﬁcation

Another problematic assumption made about the nature of contrast is that once a particular contrast has been determined, it remains constant across the whole lexicon of a language. Ito and Mester (1995a, 1995b, 2001, 2003a), among others, have shown that this kind of assumption is rarely true for spoken languages. In their work they point out that a variety of social and historical factors influence language change, and that those changes are often not uniform with respect to phonological constraints. To account for such differences (especially those in Japanese), they propose the Core-Periphery model, arguing that the lexicon is comprised of a native core and increasingly peripheral substrata. In this model, the further you move from the core, the weaker the adherence to the core constraints becomes. Brentari and Padden (2001) show that a similar stratification exists in the lexicon of American Sign Language (ASL). They argue that the ASL lexicon consists of a core lexicon, native to the language; a spatial component, peripheral to the core but also native; and a foreign (i.e. non-native) component with a series of peripheral substrata including signs using fingerspelled letters and borrowings from other sign languages (Figure 2.4). In general, as a form gets further from the core either in the direction of classifier forms or in the foreign vocabulary, it obeys fewer of the set of phonological constraints attested in core forms. In Sections 2.3.1–2.3.2, I provide a brief introduction to the lexical components proposed by Brentari and Padden for sign languages, as well as to some of the cross- componential differences in phonological behavior that have been noted in the litera- 11

Figure 2.4.: Stratiﬁcation of the ASL lexicon (Brentari and Padden, 2001).

ture. Although Brentari and Padden’s model was developed for ASL, here I assume a similar organization for other sign languages.8

2.3.1 Lexical components for sign languages

Core component

Core forms are usually understood to be the ‘frozen’ or established signs of a sign language and are considered to be native by the users (e.g. teach, strange, benefit; Figure 2.5).9 These forms consist of signs that were nativized or lexicalized

8See Chapter 3 for distributional evidence supporting this assumption. 9I use the following glossing conventions to indicate various sign types: small caps for core signs (e.g. teach), small caps with letter(s) underlined for initialized signs (e.g. society), and single quotes for English translations of classiﬁer forms (e.g. ‘vehicle drives by’) or individual ﬁngerspelled letters (e.g. ‘T’). 12 over time from the other components. Handshapes in core forms are exclusively phonological and combine with other elements to form stems.

Figure 2.5.: Examples of core lexical items.

Foreign component

The foreign component includes fingerspelling and initialized signs, as well as signs borrowed from other sign languages and cultural gestures; most of this work will focus on the former two types. Fingerspelling did not originate in the Deaf Community, but was originally adopted as a tool for educating Deaf children. The handshapes of the fingerspelling alphabet have a one-to-one correspondence with the letters of the written alphabet used by the majority culture, but fingerspelling is not an exact correlate to print since it must be produced and perceived sequentially (Padden and Le Master, 1985). In America, fingerspelling is used most often to represent common and proper nouns(especially specialized vocabulary borrowed from English), but it can also be used to signify contrastive meaning and to add emphasis. In other countries, fingerspelling is used in similar capacities, albeit much less frequently (Padden and Gunsauls, 2003). Because 13

fingerspelling is an indirect borrowing from spoken language orthography, in many cases (especially where the fingerspelled handshape was adopted to try to visually represent the written character) the handshapes can be fairly complex in comparison to those of the core lexicon. (See Figure 2.6 for examples of handshapes from fingerspelling.) Initialized signs are typically signs that use the location, movement and orientation of a native form, but replace the handshape with a fingerspelled letter to connect it more closely to a given spoken language translation (e.g. society, organization and department (Figure 2.6) from the native ASL sign group), i.e. they have a handshape of the manual alphabet as an affix, and are built from core stems. Most initialization comes from hearing educators and invented sign systems, and the acceptability of such signs depends on a number of sociolinguistic factors, for example, whether or not the initialized form creates a new lexical distinction and/or the suitability of its handshape in the context of the sign’s other parameters (Woll, 1987). These signs are often considered to be non-native and peripheral by signers and researchers alike.

Spatial component

The spatial component as defined by Brentari and Padden consists of forms commonly referred to in the sign language literature as ‘classifier constructions’, as well as spatial verbs (defined in Padden, 1988), pronominals, and predicates relating to locative direction (e.g. up, down). This work will only examine the first of these, namely, classifier forms.10 Classifier constructions are polymorphemic complexes using a movement as a verbal root in combination with location and handshape affixes (e.g. ‘vehicle arrives’, ‘long cylindrical object’; Figure 2.7). The handshapes of these constructions (the classifiers themselves), carry classificatory information about the

10There has been much discussion in the literature about whether or not ‘classiﬁer’ is the correct label for these forms (see Schembri, 2003 for an overview and discussion). I do not address those issues here and continue to use the term since it is widely recognized. 14

Figure 2.6.: Examples of ﬁngerspelled letters (top) and initialized signs utilizing them (bottom). 15

nominal referents involved (e.g. semantic category, size, shape).11 More literature concerning classiﬁers will be provided in Section 3.5.

Figure 2.7.: Examples of classiﬁer constructions.

2.3.2 Phonological behavior across the lexicon

Phonological and morphological diﬀerences across the strata of sign language lexicons have long been noted in the ASL literature, although often for the purpose of excluding exceptional forms.12 One such diﬀerence can be seen in the distribution of various handshapes across lexical components—some handshapes are common to all components, while others are found only in certain components (Brentari and Pad- den, 2001; see Figure 2.8 for examples). Distributional asymmetries like these will be addressed further in Chapter 3.

11Benedicto and Brentari (2004) show that classiﬁer handshapes also demonstrate syntactic alternations, such as transitive and intransitive. The handshapes in the other two components do not carry this type of morphosyntactic information. 12Lexical diﬀerences based on substrata have been described in the literature of other sign languages as well (e.g. Johnston and Schembri, 2007). 16

Figure 2.8.: Example handshape asymmetries across ASL lexical components (adapted from Brentari and Padden, 2001). 17

Forms from non-core components (i.e. foreign borrowings and classifiers) are often set aside by researchers when formulating phonological models or generalizations because the forms (especially with respect to handshape and movement) vary so much from that of core lexical items. For example, Sandler (1996) excludes fingerspelled handshapes from her analysis because they only occur in initialized signs and “do not appear to be relevant to the phonology of the language” due to their lack of participation in handshape change and assimilation (125, fn). She goes on to say that she assumes that these handshapes (specifically ‘R’, ‘T’, ‘N’, ‘M’, ‘W’ and ‘K’/‘P’) are represented holistically instead of using independent features (e.g. SF or joint features). That their form and behavior is different from that of core forms is clear, but I disagree with the exclusion of these handshapes from phonological analysis since they appear in commonly used signs like color terms, days of the week, and even the sign for water. Friedman (1977) also observes that certain handshapes occur only in loans from English, commenting: “we might say that they have not been fully incorporated into the phonological system of the language” (18). Although she does not exclude these handshapes from her analysis, she does consider them to be marginal—a view more consistent with the core-periphery approach than that of Sandler. Classifiers, too, are often excluded from analyses on the basis of their aberrant behavior. For instance, Aronoff et al. (2003) have argued that classifier constructions do not adhere to the constraints found for prosodic words in ASL or Israeli Sign Language. They state:

[C]lassiﬁer constructions freely violate all of these constraints—monosyllabicity, selected ﬁnger, symmetry, and dominance—regardless of internal morphological structure. (71)

Sandler (1989) also cites some of the same behavioral differences in the description of her phonological model for ASL saying that they render the classifier system “beyond the scope of the/[her] current discussion” (97), and Cogill-Koez (2000) argues that 18 because of these differences, classifier predicates should be analyzed as a separate and non-linguistic communication mode based instead on schematic visual representations. Again, the observation that classifiers behave differently than core forms is not in question, but given the frequency with which classifier constructions are used by signers, and their ability to make grammaticality judgments about them, I argue that they should not be entirely discounted in discussions of phonological representation. Brentari and Padden take an important first step in unifying the varying behavior of forms throughout the lexical components under a single theory by showing that, similar to Ito and Mester’s analysis, the phonological and prosodic constraints of the core are weakened as you move out into the peripheral levels. Here I expand upon their work, attempting to find specific differences in handshape contrasts between lexical components (see Chapter 3). I then present an approach for analyzing these differences using Optimality Theory (Chapter 5), presenting a unified analysis for handshape that accounts for cross-componential differences instead of leaving them for future research, thereby showing that they do, indeed, have a place in the overall grammar.

2.4 Iconicity

Iconicity can generally be deﬁned as a perceived similarity between the form of a sign and its meaning. Since relation to iconicity is only part of my examination of handshape and is not the main thesis of this work, and since the literature on the topic of iconicity in sign languages is vast, I will only be including a small portion of that literature here (see Taub, 2001, for a more comprehensive review). In fact, even including all of the iconicity literature pertaining speciﬁcally to phonology is beyond the scope of this work (see Brentari, to appear, for an overview). Because I am interested in ascertaining the role of iconicity in handshape contrast and representing it theoretically, I will limit my discussion to the literature in this area pertaining to the phonological representation of sign languages. 19

Use of iconicity is not restricted to sign languages; it is also utilized by spoken languages, although to a much smaller extent (e.g. quantity via vowel length, speed via speech rate, temporal ordering via syntactic structure; see Okrent, 2002, for a discussion). However, when iconicity is used by spoken languages, it rarely occurs at the level of the phoneme—let alone at the level of the phonological feature—as it does in sign languages. The most comparable examples of iconicity in speech to those in sign languages is onomatopoeia. In these cases, the relationship between the linguistic form (the spoken word) and that of the referent (the sound) is parallel in terms of modality, similar to the representation of visual information by signs. For example, Taub (2001) provides an example of waveform plots for both the ringing of a bell and the English word ‘ding’ and showed that the two were shaped very similarly. She points out, however, that despite this acoustic similarity, /dIN/ conforms to English phonotactic constraints while the sound of a bell does not. She further notes that more productive, non-conventionalized examples of onomatopoeia have the potential to be even more faithful to the actual noise they represent (e.g. [dNNNN]), but only at the expense of conformity to linguistic constraints. This has interesting parallels to the analysis presented in Chapter 5, as we shall see.

2.4.1 Iconicity debate: All or nothing

When linguistic research in sign languages began in the 1960s, admitting that iconicity was important to the formation of signs was seen as a challenge to the idea that signed languages, like spoken languages, had a duality of patterning and were indeed languages in their own right. As a result, the inclusion or exclusion of iconicity from analyses was a highly debated topic. On one side of the debate, there were those who placed a great deal of importance on iconicity’s role in sign formation. For example, DeMatteo (1977) argues for incor- porating visual imagery as a basis for a theory of ASL grammar, because he believes that ASL can be “both iconic and continuous in its code representations” (134). He 20

thus implies that in these cases there is no intervening abstraction between the real world and its depiction of it, as is necessitated in the spoken modality. On the other side of the debate are those who strive to eliminate iconicity from the phonology, either by accounting for all iconic forms using only discrete phonological units (e.g. Supalla, 1982 for classiﬁers), or by assigning all iconic elements to another part of the grammar. Examples of the latter are Channon (2002), who excludes all “predictably iconic elements” from her analysis on the basis that they are phonetic instead of phonological, and both van der Kooij (2002) and van der Hulst and van der Kooij (2006), who account for unpredictable, semantically driven properties (i.e. iconic exceptions to the rest of the phonology) by ‘pre-specifying’ them at the lexical level. Van der Hulst and van der Kooij (2006) consider such semantic motivation to be in conﬂict with phonology:

Whereas phonology implies a finite and minimal set of discrete meaning- less building blocks that are arranged in an autonomous, compositional structure, then semantic motivation (especially iconic motivation) may create an infinite array of gradual, holistic shapes that mirror the shapes of referents (object or action, directly or via metaphor). In short, phonology flourishes where meaning is absent. (267)

Furthermore, in their discussion of analyses where semantic motivation operates via the phonology (e.g. Boyes Braem, 1981) they argue that by proposing semantically motivated features for some forms (e.g. [round] for the handshape in the ASL sign diploma), these theories are implying that those same features could be distinctive in all forms, even though they sometimes never occur in arbitrary signs. By taking the view (as I do here) that the phonology does not behave the same way across the lexicon (i.e. that it prioritizes its adherence to iconically motivated constraints in 21

diﬀerent ways across components) this argument no longer poses a problem for the incorporation of iconicity into the phonology.13

2.4.2 Iconicity debate: Middle ground

The need for balance between purely iconic analyses and those that discount it entirely has often been discussed in the literature (McDonald, 1982; Taub, 2001; Brentari, to appear), and some researchers have attempted to incorporate varying degrees of iconicity into their analyses (e.g. Mandel, 1977; Klima and Bellugi, 1979; Boyes Braem, 1981; McDonald, 1982). Friedman (1977) is one such example, recognizing the need for some kind of incorporation of iconicity into linguistic analysis:

It is clear that although there are certainly ﬁnite sets of arbitrary elements in the language, a crucial element of ASL’s formational structure lies in the nature of visual representation or suggestion of referents or part of referents—in its iconicity. . . If we fail to consider the role of iconicity and insist on analyzing ASL with reference only to its arbitrary elements, we will fail to grasp the essential nature of its formational properties. (49-50)

In such discussions, the authors not only acknowledge the influence of iconicity on the phonological form of signs, but also point out that this influence is limited by linguistic constraints. For instance, Taub (2001) points out that there are language specific restrictions on iconic choices, exemplified in the use of an extended index finger ( ) to represent long-thin objects as opposed to an equally straight ring finger ( ), and the use of in tracing cylindrical objects instead of one of the other available ring-shaped finger-thumb combinations (e.g. ). The analysis presented in this thesis attempts in part to explain why ASL chooses the extended index finger and not an extended ring finger to represent long, thin objects, and why

13Van der Hulst and van der Kooij do not give examples of handshape features, only location. Future work will be needed to discover whether or not the current approach is useful for parameters other than handshape. 22

the index-thumb ring is chosen above the others for tracing cylinders. I argue that it is not, in fact, an arbitrary choice, but rather the end result of ranking conﬂicting forces in the phonology. My view regarding the role of iconicity in phonology can be summarized using a statement by Brentari (to appear):

Iconicity should not be thought of as either a hindrance or opposition to a phonological grammar, but rather another mechanism, on a par with ease of production or ease of perception, that contributes to inventories.

It is this approach that I take in Chapter 5 of this work, formalizing the inﬂuence of each of these mechanisms by using the constraint based Optimality Theory.

2.5 Cross-linguistic comparisons

In addition to looking at contrast with respect to the three areas above, I will also include cross-linguistic comparisons wherever possible to show that the observations made are not particular to ASL. More specifically, I examine data from two additional sign languages—Swiss German Sign Language (DSGS) and Hong Kong Sign Language (HKSL). I chose these languages for two main reasons. First, I had access to a variety of kinds of data in both languages. Specifically, I had access to phonological and/or morphological information on core and foreign forms via dictionaries or lexical databases (Tang, in press, and Boyes Braem, in progress) and access to classifier data by virtue of my work on a larger cross-linguistic classifier project.14 Secondly, I chose these languages because they appear to be sufficiently different historically from each other and from ASL to facilitate cross-linguistic comparison; HKSL shares no common linguistic ancestor with the other two languages, and ASL and DSGS are only very distantly related, if at all.

14This cross-linguistic classiﬁer project was funded by NSF grant 0112391-BCS; P.I. Diane Brentari. 23

2.6 Methodological approach

Languages are complex entities influenced by a multitude of factors. As such, if we are to have any hope of representing them theoretically, it is important that the theory be grounded in as wide a variety of areas as possible. Because I feel it is the only way to get a comprehensive view of the issues being explored (especially given the relative lack of information in the field as compared to spoken languages), I use a variety of methodologies in this work. First, in Chapter 3, I make observations based on archival data (dictionary sources and elicitation). Next, in Chapter 4, I introduce information based on the results of a psycholinguistic experiment. Finally, in Chapter 5, I attempt to provide a unified analysis of both sets of information using current phonological theory. Each type of methodology will be explained in more detail in the chapter for which it is relevant. 24

3. INTERNAL EVIDENCE: DICTIONARY AND ELICITED DATA

If, as the literature indicates, there are phonological differences in the behavior of handshapes across lexical components within languages, we should also see differences in the distribution of featural contrasts across these components. In addition, it would be useful to know if these distributional differences occur in more than one sign language. This chapter will provide internal evidence supporting the hypothesis that both the distributions and types of contrasts available in the handshape inventories of sign languages are not homogenous within or across languages. I begin by providing a description of the methodology used in the chapter. I then use two examples of featural contrast—[stacked] vs. [spread] joint configurations and the SF quantity values of three vs. two—to illustrate these cross-componential and cross-linguistic differences. Contrasts involving [stacked] and a quantity of three were chosen from among many potential candidates for investigation because they are representative of marked joint and SF properties, respectively. In addition, these properties are rare enough both within and across sign languages to show distributional differences, and neither can be totally explained by ease of articulation and ease of perception constraints. Finally, I conclude the chapter with an examination of joint and SF contrasts across classifier types within the spatial component. In this final section I address not only the differences in distribution and contrast type, but also how the contrasts are influenced by iconic relationships with real-world objects. 25

3.1 General methodology

In looking for differences in distribution and contrast, I examined a variety of data from ASL, DSGS and HKSL. Core and foreign data were taken from ASL dictionaries (Stokoe et al., 1965, and Valli, 2005) as well as the HKSL and DSGS dictionaries and lexical databases mentioned in Section 2.5 (i.e. Tang, in press, and Boyes Braem, in progress). In the case of HKSL and DSGS, information was also checked with local researchers whenever possible. In my examination of classifier forms, I used data from a larger set collected for a cross-linguistic classifier project, consisting of signers’ descriptions of 300 pictures and five short stories designed by Inge Zwitserlood (2003). These stimuli depict a wide range of participants and events, both plausible and implausible, and they were used in hopes of eliciting both common and rare classifier forms. (More detailed descriptions of specific stimulus items will be provided in later sections as necessary.) The data set also included interviews with informants conducted by Diane Brentari in which she questioned native signers (i.e. signers form Deaf families who learned their sign language from birth) about how the articulation of classifiers could (or could not) be altered morphologically using a variety of fingers and joint configurations. For all classifier data discussed here, I examined elicitations from three native signers per language. Before presenting this section of the analysis, it is important to mention that the results reported here are qualitative. Since the data come from a wide range of sources, counting the number of tokens for any given handshape on any given task is much less relevant to the analysis than is finding patterns that arise across the many kinds of data. These patterns, once identified, can then be verified using more quantitative approaches in future work. 26

3.2 Utilization of [stacked] joints

In my first illustration of contrast differences, I concentrate on the joint feature [stacked]. In this configuration, each successive selected finger of a handshape becomes increasingly flexed, beginning with the index finger (Johnson, 1990). In other words, the fingers are progressively spread apart from each other in a plane perpendicular to the palm (see Figure 3.1 for examples of ‘plain’ and [stacked] handshapes).

Figure 3.1.: Examples of plain and stacked handshapes.

To determine the distribution of the stacked configuration in these languages, I first looked at the dictionaries and lexical databases as sources for core and foreign forms. For classifiers, I concentrated on data from two picture stimuli (Zwitserlood, 2003) that I felt had the best chance of eliciting a stacked configuration due to the leg positions of the characters involved (Figure 3.2). Although the [stacked] feature is attested in four-fingered handshapes, here I limit my discussion to those handshapes with only the index and middle fingers selected (e.g. vs. ) because of their greater availability in the data. 27

Figure 3.2.: Stimuli used to elicit stacked handshapes (Zwitserlood, 2003).

In the following sections, I will ﬁrst present the distribution of [stacked] conﬁgu- rations across the lexical components in each language of the study, and then I will discuss the types of contrast (in relation to [spread]) found in each environment.

3.2.1 Language-internal distributions

ASL

In ASL, the stacked configuration is found in the foreign component in the fingerspelled letters ‘K’ and ‘P’, as well as in initialized signs using those letters (e.g. king, principal). In the core component, stacked can be seen in signs such as borrow, twice and see. The classifier data also yielded stacked handshapes; to represent the pictures in Figure 3.2, two of the three informants used the stacked configuration to represent both the boy climbing over the fence and the person hurdling. (The third informant used lexical items to portray each event.) 28

DSGS

As in ASL, a handshape is used to represent the DSGS fingerspelled letter ‘K’, and as a consequence, it is also found in initialized signs with that letter (e.g. kiosk, klasse). Examination of the lexical database (Boyes Braem, in progress) suggests that in the core, a stacked variant is sometimes used in signs with handshapes in certain orientations (I return to this issue in Section 3.5.1), however, this use seems to vary across signers. In the classifier data, the stacked configuration did occur but was rarely chosen (and was sometimes actively avoided) in the depiction of perpendicular leg positions; other strategies were utilized instead. In fact, of the three informants, no one used a stacked handshape spontaneously to represent the boy climbing over the fence—leg position was either not mentioned or a two handed form (representing one leg per hand) was used—and only one informant used stacked fingers for the hurdler, the others using the two-handed form or a non-stacked handshape with both fingers ‘jumping’ simultaneously. When asked, one of the informants who did not use a stacked handshape admitted that it was an acceptable variant, but then went on to say that she felt representing each leg on a separate hand was the better choice.

HKSL

No stacked handshapes were found in the known HKSL character signs (i.e. foreign borrowings derived from written Chinese characters), but that does not necessarily mean that stacked forms are expressly prohibited in the foreign component.1 In contrast, there were multiple signs in the dictionary using (e.g. borrow, silver, exploit), which appeared to be core forms, i.e. there was no obvious iconic relationship between the signed form and either the written characters in the gloss or a known classiﬁer. Finally, in the classiﬁer data examined, all three informants used

1It is interesting to note here that the character sign introduce (based on the written character

) could potentially use to represent the top two strokes of the written form, but instead uses

an handshape. 29 stacked handshapes spontaneously in their depictions of the two picture stimuli. In addition, is the recognized semantic classiﬁer used to represent bikes of all sorts in HKSL (Tang, in press). These classiﬁer examples indicate that the use of stacked handshapes (or at least ) is perfectly acceptable for this language in the spatial component.

Summary and cross-linguistic comparison of distribution

To summarize thus far, the stacked joint configuration in was found in nearly all segments of the lexicon cross-linguistically, although the degree to which it was used varied depending on the language. Cross-linguistic distributional differences included a lack of stacked handshapes in the foreign component of HKSL, (likely due to the relatively small set of known character signs rather than a phonological constraint), and an apparent bias against (but not prohibition of) stacked handshapes in DSGS classifiers as compared to ASL and HKSL. Further research is needed to determine whether or not these observations hold true on a greater scale and in more naturalistic settings.

3.2.2 Contrast types

Although I found stacked handshapes in both core and non-core (i.e. more peripheral) components of these languages, the types of contrast utilized often diﬀered depending on lexical aﬃliation.

ASL

When comparing the handshapes and in ASL, the feature [stacked] is distinctive in the foreign component as evidenced by the minimal pairs ‘K’ vs. ‘V’, and 30

variants of the initialized signs kitchen vs. vanilla.2 Meanwhile, this feature can be prominent in the classifier system—it is used morphologically in to represent the position of each leg when spread perpendicular to the body. If instead, the legs being represented are spread side-by-side, the spread (not stacked) configuration is used in the same fashion. Furthermore, the [stacked] feature is active throughout the ASL lexicon since is a predictable variant of in certain contexts. For example, a stacked articulation is typically seen in core signs such as borrow, figure-out and care, where the palms face inward toward the midline. This alternation also occurs for signs in which the middle finger makes contact with another part of the body (e.g. the core forms see and twice and the initialized sign verb). Finally, stacked variants can also be found in signs with an underlying palm-up orientation and/or those involving wrist rotation to a palm-up orientation (e.g. the final position of the sign fall, or the classifier construction ‘vehicle-turned-on-its-side’). The wide distribution of this active status makes sense in terms of ease of production. First, the middle finger does not have its own extensor muscle as the index finger does, and thus it is more difficult to support full extension in that finger, especially since, in this case, it is tethered via ligaments to the fully flexed ring and pinky fingers. In cases where the hand is palm-up, the flexor muscles in the fore- arm tighten making extension of the ulnar fingers more difficult.3 However, although these production-based constraints are available to and found in all segments of the ASL lexicon, they are blocked in situations where the [stacked] feature is involved in a prominent contrast. For example, in the palm-up classifier handshape used to represent a sunbathing person lying on his/her back, a stacked configuration would be rejected unless it matched the leg position of the sunbather. I return to this issue in Section 5.3.2. 2I acknowledge that the thumb positions of these handshapes are slightly different, but because thumb positions in general are highly variable, I predict that the stacked configuration is the primary contrastive feature. More research is needed in this area. 3For more information about the anatomy of the hand and its relevance for sign languages, see Ann (1993, 2006); Greftegreff (1993). 31

DSGS

For DSGS, [stacked] is also distinctive in the foreign component for and , again evidenced by the minimal pair ‘K’ vs. ‘V’. The feature did occur in the classifier data to represent the hurdling leg position, therefore it can be considered prominent in the classifier system—however, its limited use should be noted, and more data should be examined in the future to determine the extent to which this prominence is utilized. The extent to which [stacked] is active across the DSGS lexicon should also be pursued, although the data indicates that an articulation-based constraint similar to ASL’s may be in effect for some signers in at least the core component.

HKSL

Since I found no examples of character signs with stacked handshapes, the type of contrast in the foreign component of HKSL could not be determined. In the core, however, this language differs from ASL and DSGS in that [stacked] seems to be in a distinctive contrast with [spread], as seen in the minimal pair exploit (which uses ) vs. supervise ( ). Finally, because all of the HKSL informants used a stacked configuration to indicate leg position in the picture data, the contrast type in at least some classifiers (see Section 3.5.2) can be labeled prominent, as it was in the other two languages. More research will need to be done to determine if or to what extent [stacked] is also used actively in HKSL.

3.3 Utilization of three selected ﬁngers

The second illustration of distributional differences comes from an examination of what I will call three-finger handshapes (3FHSs). These are handshapes in which three fingers (not counting the thumb) are selected.4 Traditionally, 3FHSs have

4As mentioned in section 2.1, selected fingers are usually those that participate in movement internal to the handshape or make contact with another body part. In instances where signs are made in neutral space and/or when no finger group moves, however, selected status can be difficult to 32

been considered marked across sign languages due to their rare occurrence cross- linguistically and language internally, their physical complexity, and their tendency to be acquired late (e.g. Battison, 1978; Woodward, 1981; Boyes Braem, 1990). Because more marked handshapes adhere to fewer phonological constraints than unmarked handshapes, and because the lexical stratification literature (Ito and Mester, 1995a; Brentari and Padden, 2001) predicts that native constraints weaken further from the core, I expected that these handshapes would not be equally distributed across the lexicons of individual languages. Specifically, I expected that 3FHSs would occur more often in the foreign and spatial components than in the core within particular languages. I also predicted cross-linguistic differences as a result of differences in constraint rankings across languages (see Chapter 5). To determine the distribution of the 3FHSs in these languages, I looked at the dictionaries and lexical databases as sources for core and foreign forms, and at the articulatory interviews and a short story for the classifiers. More specifically, for classifiers I examined signers’ depictions of a story about a three-legged boy to see whether or not the representation of such a character would be a 3FHS, thus indicating a potential contrast between two-fingered and three-fingered handshapes in the spatial component (since the standard ‘legged-person’ classifier in all three languages use two fingers). Figure 3.3 shows a sample image from the story (Zwitserlood, 2003) as well as a set of handshapes used to express three legs by the informants.

3.3.1 Language-internal distributions

ASL

Examining the entries in the two ASL dictionaries (Stokoe et al., 1965; Valli, 2005), I found four 3FHSs in use within the core or foreign components— , , and . The ﬁrst three ( , and ) are from the ASL ﬁngerspelling alphabet and were

determine. In ambiguous instances such as these, I assumed that the most complex ﬁnger group of the handshape is the selected group, following Brentari (1998) and others. 33

Figure 3.3.: The three-legged boy (left) and handshapes used to represent him (right).

used exclusively in initialized signs, (e.g. wednesday, dormitory, museum). The fourth handshape ( ) was used in only one sign borrowed from a hearing gesture (boyscout). In other words, all signs in the ASL dictionaries using 3FHSs could be categorized as peripheral; there were no signs found that could be classified as ‘core’ by the criteria spelled out in Brentari and Padden (2001). In the ASL classifier data, 3FHSs were used, but to varying degrees depending on the informant. For depictions of the three-legged-boy, only one of the three informants used a handshape spontaneously, and that was only after starting with a two-fingered handshape (i.e. the 3FHS was not her first instinct). The other two informants did not use spontaneously, but when asked, thought it would probably be acceptable, (although, one did not allow the handshape if movement was added). In the articulatory interviews, when asked about using to depict the thickness of an object at a gradation between what was represented by two and four fingers (e.g. a medium-sized paintbrush), all of the ASL informants declared it to be ungrammatical. 34

One signer did say that and could be used for tools (e.g. fork, three-pronged rake), but only if accompanied by an extensive description of the object.

DSGS

In the DSGS lexical databases, I found six 3FHSs: , , , , and .5 Again, all 3FHSs were from components outside the core; they were either initialized (e.g. waadt (place name), digital, matura (a specific kind of test)), borrowed from other sign languages ( to in wales, in wc/toilette ‘toilet’) or in one case, a classifier handshape in the process of lexicalization ( in natel ‘cell phone’; Boyes Braem, personal communication). As with ASL, there appeared to be no 3FHSs in the core lexicon of DSGS. Also similar to ASL, use of these handshapes in the classifier data was infrequent and varied by informant. Two of the three informants used spontaneously in their retelling of the three-legged boy story; one used it spontaneously throughout, but commented that it was a difficult handshape to produce, while the other used it to introduce the character, but then proceeded to use two-finger handshapes (in both classifiers and core lexical items) to represent the boy in the rest of the story. The third DSGS informant did not use a 3FHS in the story at all, consistently using a handshape with two fingers and the thumb to represent the three-legged boy. When asked about the grammaticality of , she seemed uncertain and said she preferred the two-finger handshape. In response to interview questions about 3FHSs, the informants did not accept most three-fingered alterations of established classifiers (e.g. changing the thickness of an object, or representing three specific prongs of a fork), although one signer did offer a form changing from to to represent birds’ claws grabbing something.

5I consider the handshape to have two groups of SF: the index and thumb as ‘primary selected fingers’ and the remaining three fingers as the ‘secondary selected fingers’. See Eccarius (2002) for more information about this distinction. 35

HKSL

The HKSL dictionary (Tang, in press) yielded signs using five 3FHSs: , , , and .6 The origin of many of these signs is uncertain, but while some were outside the core—either character signs (e.g. in illegal, in jade), borrowed signs (e.g. in wc/toilet) or signs still closely related to the classifier system (e.g. in fork)—unlike in ASL and DSGS, many others seem to be established core forms (e.g. in champion, in cancer, in trendy; Tang, personal communication). As an additional point of interest, HKSL uses 3FHSs in their number system ( and are both acceptable variants of ‘3’), while ASL and DSGS do not.7 3FHSs were also more prevalent in the HKSL classifier data than they were in the other two languages. In representing the three-legged boy, all informants used spontaneously throughout the story. In response to interview questions about morphological alternations involving three fingers for known classifiers, all informants allowed 3FHS in at least some situations, although answers varied regarding which classifiers could be modified using three-fingered versions (e.g. ‘thick rope’, ‘medium sized paintbrush’, ‘cat’s paw’) and with what joint configurations (e.g. extended, bent, crossed).

Summary and cross-linguistic comparison of distribution

To summarize, the data indicate that, within languages, the distribution of 3FHSs was not homogeneous; more (if not all) of these handshapes tend to be found in the foreign and classiﬁer components of the lexicon rather than in the core (see Figure 3.4). The data also indicates that there are cross-linguistic diﬀerences regarding the acceptability of these handshapes in the various parts of the lexicon. For example,

6The handshape has two possible SF groups. Here, I refer to the form with the three extended fingers selected. 7The status of numbers in the organization of the lexicon has not yet been determined and is a topic for future research. 36 based on the available data, HKSL appears to be more accepting of 3FHSs in core signs than DSGS and ASL. These handshapes are still marked in HKSL (based on dictionary frequency), but they do occur in at least a few core lexical signs. I also found a greater variety of 3FHSs as classifiers (i.e. variation in allowable joint configurations and in the specific fingers selected) in HKSL than in the other languages, as well as a higher degree of certainty about using them (e.g. the three-legged boy).

Figure 3.4.: Examples of three-ﬁnger handshapes (3FHSs) in the foreign, core and classiﬁer components for ASL, DSGS and HKSL. 37

3.3.2 Contrast types

Additionally, across the substrata, there are different types of contrasts involved in the selection of three fingers as opposed to two or four.8 In the foreign substrata of ASL and DSGS, the choice of three SF vs. two is distinctive—changing the number of fingers selected (e.g. to ) results in a different and unrelated meaning (‘W’ to ‘V’). This is probably also true of character signs in HKSL, however, using the available data, no minimal pairs were found involving 3FHSs in this segment of the lexicon to support this supposition.9 In classifiers, the use of three fingers could be considered prominent in all three languages, but the degree to which that prominence is/can be utilized varies cross- linguistically. In ASL and DSGS, the grammaticality of these handshapes varied across informants, but when they were deemed acceptable, the 3FHSs were only used to morphologically alter the number of subparts belonging to the whole (e.g. three legs of a boy, three prongs of a tool, three claws of a bird’s foot). In HKSL, this alternation involving the quantity of subparts was accepted and confidently utilized by all informants, and a wider variety of joint configurations was available to them (e.g. [crossed] in ‘thick rope’, bent in ‘cat’s paw’), although the acceptability of specific examples varied across informants. Since no 3FHSs were found in the core lexical components of ASL or DSGS, discussion of contrast type for these handshapes in the core is a moot point. Furthermore, although these handshapes did exist in the HKSL core, due to their rarity, no minimal pairs could be found to test for contrast. However, it is interesting to note that there are instances where contrasts involving three SF seem to be disappearing over time. For example, the ASL signs doctor and thousand were originally borrowed from the initialized LSF signs médecin and mille using handshapes, but there seems

8In Brentari (1998), two and three selected fingers are represented by a dependancy relationship between the quantity features [one] and [all]: two = [one]/[all] and three = [all]/[one]. 9While I could not find true minimal pairs in character signs for SF quantity involving 3FHSs, there are pairs that differ in both quantity and point of reference (for an example, see Figure 5.5 in Section 5.3.1). 38

to be a diachronic move toward the use of four-selected fingers instead as these signs lose their foreign status (see Section 5.4.1 for an analysis). Similarly in HKSL, the dictionary lists three- and two-fingered variants for one instance of a borrowed sign ( vs. in boyscout, originally from a hearing gesture), perhaps indicating that, although allowable in the core, in some cases three-fingered forms are still less desir- able than other alternatives. In fact, given the lack of perceptual salience (especially at a distance) between unspread 3FHSs and their two- or four-fingered counterparts (e.g. vs. or ), these changes should not be wholly unexpected. However, more research is needed to know if this is indeed the case.

3.4 Summary of [stacked] and 3FHS contrast diﬀerences

To summarize these two sections on distribution, we see that there are discernible differences both across languages and within languages in the way that sub-lexical properties of handshapes are used. Regarding the [stacked] joints feature, the data show that HKSL is the only one of the three languages that uses this feature dis- tinctively in the core (there is at least one minimal pair), and that it is acceptable and prominent in the spatial lexicon to some extent in all three sign languages. For 3FHSs, this analysis shows that there are differences in the degree of acceptability across the three sign languages: 3FHSs are more acceptable in HKSL than in either DSGS or ASL as evidenced by the existence of core lexical items with 3FHSs in HKSL but not in the other two sign languages. My analysis suggests that for the two specific properties studied thus far, DSGS is the most conservative of the three sign languages, especially in classifiers—given its preference for two-finger variants over 3FHSs and its reluctance to use [stacked] when another option is present—suggesting that iconicity is less of a motivation in this language than in the others in the spatial component. Finally, the kinds of contrast described thus far are similar across the three sign languages—prominent in the spatial lexicon, potentially distinctive in 39 the foreign and core lexicons, and potentially active in all three (although sometimes blocked under certain conditions).

3.5 Handshape contrast across classiﬁer types

Distribution and type differences of feature contrasts not only occur among the major lexical components as illustrated above, but they can also occur across smaller segments of the lexicon. In this section, I propose a new taxonomy for classifier types based on observed relationships between their morphophonemic behavior and the properties of the real-world objects they represent. I then make more general observations on the distribution of various kinds of features and contrasts across the various classifier types. Finally, I suggest that there are asymmetries in the phonological behavior of handshape features across the different parts of the spatial component. The observations here are based primarily on ASL data (leaving most of the cross-linguistic comparisons as an area for future work), and the data come from descriptions of the Zwitserlood pictures and from the interviews eliciting grammaticality judgments about which feature classes can be altered in members of the four types.

3.5.1 Classiﬁer types

In the literature categorizing classifiers, most researchers have been primarily interested in categorizations of classifier handshapes in terms of how they interact with movement types, often at the morphosyntactic level (e.g. Engberg-Pedersen, 1993; Schick, 1987; Liddell and Johnson, 1987; Zwitserlood, 2003; see Schembri, 2003 for a more complete review). With the exception of a few (e.g. Supalla, 1986; Boyes Braem, 1981; McDonald, 1982), if these researchers discussed the phonological details of these handshapes, it was usually only a small part of their overall analysis. In contrast, I will only be focusing on the morphophonemic behaviors of the handshapes themselves (and only static ones at that), i.e. I will focus on how various handshape features 40 are utilized (or not utilized) in representing particular aspects of the physical world. I should emphasize here that this analysis is not meant to replace the earlier morphosyntactic categorizations. Instead, it is my hope that this analysis will expand the work already done by contributing more details at the morpho-phonological level. As we have seen in previous sections, classifiers have the potential to alter their morphological composition by changing either their SF combination or their joint specification. However, this potential is utilized to varying degrees throughout the spatial component. The classifier types proposed here are based not only on the degree to which morphological alteration is possible in these handshapes, but also on the specific kinds of iconic representations those features represent. In this section, I first present an overview of the new categorization, and then I describe the ways in which each classifier type can be used to represent iconic relationships such as size and shape. In this work I propose four general types of classifiers based on the interaction of morphophonemic alterations (linked closely to contrast type) with iconic representation. Again, I am only looking at the types of information that can be represented by the various aspects/features of the handshape, and not at their semantic or syntactic behavior, although there is likely to be overlap between these categorizations and previously established typologies. When such an overlap seems probable, established names are used (in part or in whole) for given groups. The four types are as follows:

Semantic classiﬁers (SCLs): classiﬁer handshapes with no morphological/prominent indicators of shape or size (although they may have originated as such) (e.g. for vehicles).

Whole classiﬁers: classiﬁer handshapes where the visual properties of the hand itself become a representation of the physical characteristics of its referent (e.g. for ‘paper’, for ‘pencil’, for ‘two-legged being’). 41

Descriptive perimeter classiﬁers (DCLs): classiﬁer handshapes (typically involving the thumb) where the shape of the hand directly represents the size and/or shape of the object via its perimeter only (e.g. for ‘soda can’).

Handling perimeter classiﬁers (HCLs): classiﬁer handshapes (also typically involving the thumb) where the shape of the hand indirectly represents the size and/or shape of the object via its perimeter by representing how the hand interacts with that perimeter (e.g. for ‘hand-holding-a-book’).

SCLs are no different than Supalla’s (1978, 1982, 1986) ‘semantic’ category for classifiers; they represent the semantic category of the referent, but no information about their physical characteristics. In WCLs, the shape, size or configuration of the hand as a whole (or more specifically, of the SF) represents certain physical characteristics of the object being represented. This group includes examples from a number of morphosyntactic types previously seen in the literature—for instance, it can contain some examples of Engberg-Pedersen’s (1993) Whole Entity stems and Limb stems, as well as some of Supalla’s (e.g. 1986) static Size-and-Shape-Specifiers (SASSes) and most of his Body Part and Tool classifiers. Besides representing size and shape, these classifiers can (in some cases) utilize the quantity of unspread SF to indicate the width of an object (e.g. vs. to indicate the relative width of a knife blade), while spread SF indicate a given number of subparts of the whole (e.g. for legs or prongs). If the latter, these parts are often able to display a certain degree of independence from each other with regard to phonetic joint position, leading to prominent uses of [stacked] and [crossed]. (More details are given below). Because we still know so little about the behavior of individual classifier handshapes, this group is not as well defined a category as it could be, and the line between WCLs and SCLs can sometimes blur. For example, in most cases is considered to be a semantic classifier in ASL for upright beings—its obvious relationship to the overall shape of an upright person considered an historical artifact. However, some (but not all) ASL signers can 42

contrast this straight handshape with a bent one ( ) to indicate a person with a hunched back (i.e. a being with a curved shape). Individual classifiers, their possible modifications, and the movements and locations with which they can combine must be studied further to determine whether or not this group should be divided further and where lines should be drawn. For this work, I concentrate on examples where the relationship to shape is what determines the classifier’s use (e.g. for relatively round compact objects vs. for flat, rectangular objects), and/or those that can be altered morphemically by changing a given feature or group of features (e.g. for two-legged beings vs. for three-legged beings). As I will show below, the final two of these four types—DCLs and HCLs—behave similarly in many respects concerning how size and shape are represented, many of their differences being based on the order in which certain iconic relationships can be combined. I refer to both in a general way as perimeter classifiers, because their representation is not of the object itself (as with WCLs), but of part or all of the area surrounding the object. DCLs include examples from most size and shape indicating categories proposed by others (including Corazza’s, 1990, similarly named Descrip- tive classifiers), but I make no assumptions about their allowable combinations with movement types as most of these researchers do. Conceptually, these handshapes either represent the perimeter directly (DCLs, e.g. for ‘coin’) or indirectly (HCLs, e.g. for ‘hand-holding-a-letter’), and this difference allows additional iconic relationships and alterations in HCLs not available for DCLs, as shown below.

3.5.2 Iconic representations via phonological features

I now discuss how the last three of the four types of classifiers (WCLs, DCLs and HCLs) iconically represent physical dimensions like size and shape using different parts of the feature geometry (repeated in Figure 3.5 for convenience). Many similar observations about size and shape have been made by other researchers (e.g. Supalla, 1986; Boyes Braem, 1981; McDonald, 1982) with varying degrees of detail, but to 43 my knowledge, only Supalla has made explicit connections between the shape and/or size morphemes themselves and the specific parts of the hand used to represent them. My observations do not conflict with Supalla’s, but they are expressed in a slightly different way, using structural aspects based on the feature geometry instead of assigning morphemes to individual fingers as he did (see Supalla, 1978, 1982 and 1986 for details).

Figure 3.5.: The phonological structure for handshape (Brentari, 1998).

Size

Here I distinguish between three kinds of size representation (Figure 3.6), each making reference to aspects of both the referring object and the handshape: 1) size of the whole entity, 2) size between outside edges—i.e thickness or depth—which I call ‘surface’ size, and 3) size between reference points on the perimeter (henceforth RP size). Each will be exempliﬁed below. 44

WCLs are able to express the first two types of size (i.e. whole object and surface size), although instances are rare and seem to be subject to form-specific limitations. The size of the object as a whole can be represented by the choice of finger (at the Fingers node), for example, use of the pinky for small thin objects (a bee’s stinger) vs. the index finger for larger thin objects (a dog’s tail). Relative surface size, or thickness, of an object (when allowed, e.g. a knife blade) is represented in WCLs by changing the quantity of fingers ( for a paring knife vs. for a cleaver), although in ASL this alteration is limited to adjacent fingers with a radial (index side) point of reference. DCLs and HCLs can express size via the last two kinds of representation (i.e. surface and RP size). Surface size (i.e. depth) is represented as it is in WCLs by changing the quantity of adjacent (radial) SF (e.g. ‘thin disc’, vs. ‘hockey puck’, vs. ‘soda can’, ). RP size is represented via the distance between finger and thumb tips, which is largely controlled by varying degrees of flexion in the base joint. (I return to this point below.)

Figure 3.6.: Size representations. 45

Shape

Shape can also be represented in three ways, making reference to roughly the same dimensions as those used for size, namely: 1) shape of the object as a whole, 2) continuous shape of the surface perimeter, and 3) shape between reference points on the outer perimeter (see Figures 3.7 — 3.9 for illustrations). WCLs can represent shape in only one of these three ways, referencing the object’s shape as a whole. In these representations, the entire joint configuration (flexion starting at the Joints node) is used to approximate spherical, curved or hooked, and flat shapes in the closest way possible given the physical dimensions of the hand (Figure 3.7). Examples include representation of curved objects like spoons, flat objects like pieces of paper, and spherical objects like heads.

Figure 3.7.: WCLs representing whole object shape.

The other two kinds of shape representations are used by DCLs and HCLs to represent shape in various dimensions (or, more specifically for HCLs, the shape of the part being handled; cf. Wallin, 1996).10 Surface shape is represented using primarily flexion at the nonbase joint, and can be curved (flexed) or flat (extended). These shapes can then be combined with varying degrees of base joint flexion (i.e. RP size)

10See Wallin (1996) for a more general discussion of dimensional representations including both handshape and movement. 46 to create similarly shaped objects of diﬀerent sizes. Examples of such combinations can be seen in Figure 3.8.11

Figure 3.8.: Diﬀerent sizes of similarly shaped objects via degree of base and nonbase joint ﬂexion.

RP shapes, however, are representations of shape via the arrangement of the finger and thumb tips. These also seem to be restricted to flat or round relationships, as illustrated in the top half of Figure 3.9. These shapes can then be combined with surface shapes to create three-dimensional representations of object shapes, as seen at the bottom of Figure 3.9. This combination is not required, however, as is evidenced by some HCLs in which nonbase flexion is not expressly specified (e.g. moving checkers on a checkerboard). As with surface shape, RP shape can also be combined with RP size (represented by the arrows in Figure 3.9), controlling the understood size of the object by the relative distance between points via base joint flexion.

11The joint combination marked by the blank space in Figure 3.8 could theoretically be used to represent triangles, but in my data, signers preferred tracing triangular shapes using movement instead of representing them with DCLs. This is another area for future research. 47

Here I should note that I have not yet found explicit mention in the literature of the rounding of the fingertips used to represent RP shape—most researchers explain the difference between cylindrical and spherical shapes by referring only to the spreading of the fingers. As of yet it is unclear how this phenomenon should be represented by the feature geometry; it could be subsumed under Brentari’s [crossed] and [spread] features, or a new feature may be warranted. For now, I refer to this configuration as ‘bunched’.

Figure 3.9.: Fingertip arrangements depicting round and ﬂat RP shapes (top) and perimeter classiﬁers combining these RP shapes with a curved surface shape (bottom).

Size and shape combination diﬀerences

As mentioned, both DCLs and HCLs can utilize combinations of RP and surface size and shape representations; however, the ways in which these aspects can combine in each classifier type varies. DCLs can represent objects one-dimensionally 48 by only representing their size via base joint flexion (RP size); two-dimensionally by using a combination of base and nonbase flexion (RP size + Surface shape); or three-dimensionally using base and nonbase flexion, SF quantity, and fingertip arrangement (RP size + Surface shape + Surface size + RP shape). HCLs can also represent what is being handled one-dimensionally using only RP size (base joints) and three-dimensionally using all RP and Surface aspects combined, just like DCLs. This similarity can lead to ambiguous interpretations in both dimensions. However, to represent two dimensions in HCLs (e.g. moving a checker) only base joint flexion and fingertip arrangement (RP size + RP shape) are combined, resulting in an intermediate form different than that of DCLs. (See Figure 3.10 for an illustration of these combinations.) The differences in combination order are likely due to the relationship between HCLs and precision vs. powergrip types as discussed in the next section.

Other types of iconic representation

Size and shape are not the only kinds of relationships represented by these classiﬁer types, although they are arguably the most commonly recognized. Other relationships include the representation of subparts in WCLs and representations of control in HCLs. The use of [spread] in WCLs yields an interpretation of multiple subparts of the whole (e.g. legs of a person, prongs of an implement) rather than surface size as unspread ﬁngers do. This, like surface size, is represented via the quantity of SF.12 Furthermore, the arrangement of these parts can be represented using [spread] or [stacked] (as seen with leg position in Section 3.2) or [crossed] (e.g. for braided segments of hair). HCLs, too, utilize additional iconic relationships, this time based on their representation of the hand’s interaction with objects rather than the perimeter of the

12There is some indication that the representation of parts happens at the Fingers node instead, since SF may not be restricted to adjacent ﬁngers in this case (e.g. use of for goal posts), but these examples are rare and need to be studied further. 49

Figure 3.10.: DCL and HCL orderings for size and shape combinations. 50 object directly. Because of this relationship, aspects of kinematic control can be represented. These are best understood by briefly looking at the nonlinguistic literature on grip types.13 There are many nonlinguistic taxonomies of grip type that may prove useful in the future study of HCLs (e.g. Kamakura et al., 1980; Cutkosky, 1989; MacKenzie and Iberall, 1994), but because this is only a preliminary examination of the topic, I will only discuss the most straightforward of these, that of Napier (1956). While acknowledging that multiple factors influence the grip chosen for a particular object (e.g. shape, size, weight, texture, etc.), Napier believed that the most significant factor was stability of the object in relation to the wielder’s intended task. He thus divided prehension into two main types (Figure 3.11), which he describes as follows:

1) The object may be held in a clamp formed by the partly flexed fingers and the palm, counter pressure being applied by the thumb lying more or less in the plane of the palm. This is referred to as the power grip... 2) The object may be pinched between the flexor aspects of the finger and the opposing thumb. This is called the precision grip. (903)

Put another way, the primary joint feature distinguishing between these two grip types—what I am considering [stacked]—directly contributes to the power requirements with regard to direction of force on and support of the object. Anatomically speaking, by stacking the fingers in power grips, the fleshy part of the palm under the pinky rises, providing additional support by cradling the object between it and the fleshy eminence at the base of the thumb. The size and shape of the object itself then determine the degree of finger flexion and the area of the palm involved. Conversely, in precision grips, the degree of flexion and spreading in the base joints depends primarily (instead of secondarily) on the dimensions of the object being held. In this grip type, instead of being in a stacked position, the fingers are typically distributed

13Boyes Braem (1981) and Supalla (1986) also include observations of control diﬀerences in their handshape analyses, but neither go into much detail regarding handshape internal features. 51 around the rotational axis of the object (as seen on the right in Figure 3.11).14 Napier also distinguishes between the thumb positions of the power grip and the precision grip—in most power grips, the thumb is largely unopposed, lending support in opposition to the ulnar side of the hand, and in precision grips it is opposed to bring it in more direct opposition to the ﬁngertips.

Figure 3.11.: Power vs. precision grips (based on Napier 1956).

Napier stresses that grips should not be categorized solely by the physical properties of the object (as others have done) noting that there are many ways of manipulating the same object depending largely on the task at hand. For example, he points out that in unscrewing a tightly sealed jar-lid, a power grip is used until the lid has been loosened, at which point a precision grip is used to ﬁnish the removal process. Napier does not, however, state that the physical properties of the object play no part in the choice of grip. In fact, in explaining the inﬂuence of size of prehension, he describes a continuum of sorts based on which hand posture would provide the most stability—at the extremes, very small and very large items are most stable when

14Napier also comments that as the size of the object decreases, the fingers involved shift toward the thumb and index finger because they are better suited to fine motor control, which parallels SF choice in HCLs. 52

using a precision grip (one-handed for small items, two-handed for large), while in the middle, moderately sized objects can be stabilized using either a precision or a power grip, depending on the task.15 If such a systematic use of [stacked] and thumb opposition also occurs in sign languages, these two features could be considered prominently contrastive in HCLs as indicators of control type. More research is needed to ﬁnd out if such a contrast exists.16

3.5.3 Feature asymmetries across classiﬁer types

Now that I have explained the various types of classifiers and how they represent iconic relationships using phonological features and structures, I return to the issue of contrast differences across the spatial component. In this section, I show that there are asymmetries in the phonological behavior of handshape features throughout the spatial component of the ASL lexicon based on the classifier types described above. (Example handshapes from all types illustrating these points can be seen in Figure 3.12.)

Distribution of joints and selected fingers in different kinds of classifiers

Recall that SCL handshapes represent an object as a whole according to its semantic class. As such, their representation tends to be more abstract than other classiﬁer types, although historically, connections to iconic relationships can still be made in some cases (e.g. for an airplane). Within this classiﬁer type, I have found quite a variety of possible SF combinations (especially considering that SCLs constitute a very small number of forms) ranging from unmarked combinations ( )

15Napier also points out that the need for power and the need for precision are not mutually exclusive; this can (and often does) result in combination grips (e.g. the grip used in tying shoelaces). The relationship between these grip types and HCLs handshapes still needs to be studied. 16Some examples suggestive of a stacked vs. nonstacked contrast did occur in my data, (e.g. the HCLs for picking up an apple vs. eating an apple), but these examples were complicated by lexical influences from the core, as well as the orientation changes discussed in Section 3.2. 53 to more marked ones ( ). In contrast to this variety in the SF, I found only a small number of distinctive joint configurations used in these forms (extended, bent, and fully flexed), with the vast majority of SCLs having extended fingers. WCLs, on the other hand, make use of more iconicity than do SCLs, representing some physical attributes of the object being described. Like SCLs, they have an asymmetry between the distribution of their joint and SF possibilities, albeit not as pronounced. I found only three joint configurations based on flexion—extended, fully flexed, and a curved/bent configuration varying with finger spread—in addition to the more minor configurations of crossed, spread and stacked (which can only be combined with extended or partially extended base joints). In contrast, a relatively large number of SF combinations were used in WCLs (see Figure 3.12 for examples and Section 5.3.1 for a more complete list). DCLs (in my phonologically based interpretation) do not represent the object, per se, but instead represent the perimeter of that object; thus, altering the joints and/or SF of the handshape alters the dimensions of that perimeter, as shown in Section 3.5.2. Many in the past have assumed that classifier constructions were continuous or gradient representations of objects and motion events (e.g. DeMatteo, 1977). How- ever, Emmorey and Herzig (2003) determined that instances of this type of classifier handshape were categorical when produced by signers naïve to a specific range of sizes—although perceptually, signers could also be sensitive to gradient differences in handshape. Based on their results, they hypothesize that: 54

[W]ithin a contrast set, signers know that handshape size can be manipulated to indicate gradient variations in size. This gradient variation may be thought of as a gestural overlay on a morphemic representation. As noted, the analogy may be to English speakers using variations in vowel length to indicate gradations in duration or length. (244)

This ability to use DCLs either gradiently or categorically makes determining specific contrasts challenging, to say the least. However, in the data thus far I have found examples of numerous joint configurations that I feel have the potential to be categorical contrasts in DCLs (see Figure 3.12).17 Further analysis will be carried out in future work to determine more exactly the categorical boundaries (assuming they exist) between these configurations. The SF possibilities for DCLs on the other hand, are much more restricted (although admittedly the potential for direct gradient representation is lessened in this case by the limited number of fingers available). Thus far I have observed only the use of the index finger, middle finger, index and middle fingers together, and all fingers (Figure 3.12).18 Unlike SCLs and WCLs, in all possibilities including more than one finger, SF combinations in DCLs were restricted to those including adjacent fingers. In contrast to DCLs, I consider HCLs to be a representation of the shape of the hand manipulating an object rather than the dimensions of the object itself. There is some research that indicates that HCLs, at least in terms of their handshapes, are highly gestural in nature, (e.g. Slobin et al., 2003). I agree that of all of the classifier types, HCLs seem to be the closest to the gesture system, but I maintain that they are not completely void of linguistic systematicity. For example, in my data, ASL signers did not permit HCLs with three fingers, despite the fact that such grips are

17This is supported by an increased number of joint contrasts found for DCL forms in the experiment in Chapter 4 as compared to the core and foreign forms. 18The difference between use of the index finger and the middle finger with the thumb to represent thin disc-shaped objects (e.g. ‘plate’), seems to be a stylistic difference rather than a prominent contrast, although one informant did comment that she herself often used the middle finger version to indicate a piece of something (e.g. ‘slice of pizza’) and the index finger handshape to indicate a whole object (e.g. ‘pizza’). 55 used in the real world.19 Even so, I acknowledge that the close relationship between HCLs and gestures, and the resulting complexity of the handshapes themselves, make determining specific contrasts extremely difficult. As with the DCLs, however, there are numerous potential joint contrasts based on differences in handling objects of various shapes and sizes as shown in Section 3.5.2 (als see Figure 3.12), especially as compared to the small number of observed SF combinations; only one, two, and four finger handshapes have been found, all of which include adjacent fingers.

Figure 3.12.: Example handshapes from the four proposed classiﬁer types.

19For an example, see the ‘Thumb-3 Finger’ grasp in Cutkosky’s (1989) taxonomy of grasp types used in manufacturing tasks. 56

Contrast diﬀerences across classiﬁer types

As we have seen in previous sections, the morphological composition of classifiers can be altered by changing either their SF combination or their joint specification. However, this potential is utilized to varying degrees throughout the spatial component. For SCLs, changing the number or combination of SF involves distinctive contrasts, since it results in a separate or ungrammatical forms. For example, if you increase the number of SF (not counting the thumb) in a ‘vehicle’ classifier from two ( ) to four ( ), the resulting handshape can not be used to represent a larger car; it is ungrammatical in this context. (A change in size of a vehicle can instead be represented by using lexical items in addition to adding adjectival nonmanuals to the production of the classifier construction, e.g. puffed cheek for ‘big-truck-driving- past’.) Furthermore, changing which two fingers are selected—e.g. the index and middle fingers for a ‘vehicle’ classifier ( , e.g. car) to the index and pinky ( )— results in a different SCL (in this case, ‘airplane’). Regarding joint specification, only one alteration, from extended to bent, seems to be allowed in SCLs for some (but not all) ASL signers, and that alteration can only be made to the classifier handshape after the canonical form has already been established in the utterance (e.g. the change to after the introduction of the vehicle classifier to indicate a crashed car; first mentioned in Supalla, 1982). This process is more similar to the sequential morphology found (albeit rarely) in the core lexicon than it is to the simultaneous morphology most often found in classifier constructions (Aronoff et al., 2005), which means that contrasts in the joint configuration of the classifier itself (as opposed to the additional affix) could also be considered distinctive. In WCLs, determining the contrast type is not a straightforward task. As shown in Section 3.5.2, shape, size and both number and arrangement of subparts are represented in WCLs in systematic ways at the Joints and Finger nodes (e.g. thickness using different SF combinations) which I take to indicate prominent contrasts for 57 both kinds of features. However, these alternations seem to be highly restricted by individual form, indicating that some feature contrasts for particular classifiers may be distinctive instead. More research is needed to clarify which features are prominent and in what specific classifiers. The wide variety of joint configurations available to DCLs mentioned in the previous section represent (potentially) prominent contrasts—i.e. changing the joint configuration morphologically changes the shape and/or size of the perimeter of the object being described. The use of different SF combinations to represent the surface size of an object can also be considered prominent in most cases, since a change in SF corresponds to a change in object width. As shown in Section 3.5.2, HCLs are similar to DCLs in that changes to the joints and/or SF of a given HCL can result in a perceived change in an object’s shape or size, albeit in this case by virtue of the way various objects are routinely held (e.g. for handling an envelope vs. for handling a book). Similarly, changes in SF combinations (in conjunction with varying joints) indicate object width/surface size (e.g. different sizes of light bulbs) by showing how each is manipulated. Consequently, these changes indicate prominence in both joint and SF contrasts, this time at the level of individual features or terminal nodes of the phonological structure. In summary, the four types of classifiers discussed in this section vary in the distribution of joint and SF contrasts as well as in the type of contrasts possible. These four subtypes constitute a continuum ranging from the more core-like SCLs to the more gesture-like HCLs (Figure 3.13). As you move away from the SCLs on this continuum, the number of SF contrasts found in each subtype decreases while the number of joint contrasts increases.20 Also, the type of contrast within these features

20This generalization is supported by preliminary evidence from an ongoing project studying the handshapes of signers and gesturers. Only signers exhibit the type of handshape complexity described here (both in ASL and Italian SL), and particularly, SF complexity in classiﬁer constructions is most evident in the types of classiﬁers that I am calling SCLs and WCLs (i.e. those in which the hand as a whole represents the object). 58 ranges from distinctive in the SCLs to prominent in the DCLs and HCLs, with WCLs being a middle ground between the two.21

Figure 3.13.: Continuum for joint and selected ﬁnger possibilities in the four types of classiﬁer handshapes.

This balance between SF and joints makes sense in light of articulatory and perceptual markedness. For example, the more varied the joint contrasts required for iconic representations of size and shape, the more muscular control you need. Consequently, components with high numbers of joint contrasts tend to have either all of the fingers selected, or they utilize those with the support of the anatomically independent index finger. If larger numbers of joints are not required, however, perceptual salience wins out and the forms are free to use a more diverse set of finger combinations (usually in an extended configuration) including visually salient nonadjacent combinations (e.g. .) The balancing of these kinds of pressures in the determination of forms is the topic of the analysis in Chapter 5. 21The idea that a continuum exists between frozen forms and classifier constructions is far from new. For example, Supalla (1978) discusses (in a very general way) a continuum ranging from ‘frozen’ to ‘novel’ forms, with forms in the middle that can be analyzed as one or the other depending on how much information is being attended to or is needed in the conversation. 59

4. EXTERNAL EVIDENCE: EXPERIMENTAL DATA

So far, I have presented evidence that the distribution of certain contrastive structures can differ across the lexicon of a given language (e.g. 3FHS) as well as evidence suggesting that a single feature can vary in terms of contrast type depending on the part of the lexicon in which a form occurs (e.g. [stacked]). I now ask: are these lexical differences artifacts of the same phonological gaps that make finding minimal pairs in sign languages so difficult, or are the boundaries and phonetic targets of handshape contrasts actually different cross-componentially in the mind of the perceiver? In Chapter 3, I supported the idea that contrasts differ with evidence from language- internal sources, namely dictionaries and the classifier project data for the three languages under discussion. In this chapter I explore contrast variation in one of these languages, ASL, from a language-external point of view via a psycholinguistic experiment.1 The goal for this experiment was to see whether or not signers respond the same way to certain sets of handshapes regardless of their lexical affiliation. More specifically, for a given range of handshapes common to all ASL lexical components, are 1) the number of contrasts, and 2) the phonetic targets for those categories consistent across lexical components? The experiment itself consisted of two tasks. The first attempted to locate phonological contrasts by looking for differences in meaning based solely on handshape variation, and the second tried to map (at a very basic level) the phonetic targets within those contrasts by determining how good a fit each variant was based on its meaning.

1I leave exploration of this kind in DSGS and HKSL to future research. 60

4.1 Background

The first step in testing my hypothesis experimentally was deciding on an appropriate design. While cross-linguistic comparisons of phonemic contrasts are often performed (especially in connection to L2 acquisition research), to my knowledge, no one to this point has attempted to map contrasts within different segments of the same language. I therefore had no experimental precedent to follow. Often, when people want to identify the boundaries of a phonemic contrast, they do so by means of a categorical perception (CP) experiment (see McQueen, 1996 for a review). However, testing even a single contrast using a CP experiment is very time intensive due both to the large number of test items and to the number of repetitions required for adequate results; since I wanted to compare numerous potential contrasts, each at a very basic level, a CP experiment was not a good match for my goals. Alternatively, some researchers use an identification task in combination with a goodness of fit task when investigating contrast (e.g. Guion et al., 2000). By using these methods they attempt not only to determine the category boundaries, but also to map the locations within the category that represent the best exemplar (i.e. prototypical example) of the segment. Like CP experiments, these experiments also tend to be time intensive due to the extensive range of phonetic variants tested per contrast; nevertheless, it was this type of design that I decided to use as a model for my experiment, albeit on a smaller scale. One possible hypothesis for this experiment is that contrasts are the same throughout a given lexicon—i.e. if a contrast is present in one part of the lexicon, it should also be present in all other parts—and as a consequence, signers’ responses to tasks involving the same sets of handshapes across lexical components should also be the same. Based both on the constraint asymmetries discussed by Ito and Mester (1995a) and Brentari and Padden (2001), and on the differences in contrast discussed in Chapter 3, I predicted that the results of this experiment would disprove this ‘uniform lexicon 61

hypothesis’, that is, that participants’ responses to each task would differ between one or more parts of the lexicon. However, the issue is not as straightforward as simply determining whether or not any differences exist. As mentioned in Section 2.2, sign languages tend to have very few minimal pairs, even for well established contrasts. In other words, minimal (or morphological) pairs available for one component of the lexicon can not always be found in other components for a given handshape range. For example, there is a known meaning distinction in ASL between and in perimeter classifiers used to describe round and flat shaped objects (e.g. the description of a cardboard tube vs. a sheet of cardboard). However, I could find no such minimal pairs in initialized signs (hardly surprising since there is no letter represented by ) or in core lexical items. This deficit could be the result of the very contrast differences I hypothesize, or it could simply be an accidental (albeit large) phonological gap. This imbalance in the amount of lexical competition present throughout the lexicon has the potential to affect signers’ responses in identification tasks. For instance, in the spoken language literature it has been shown that contextual factors—for instance higher-order linguistic factors such as lexical vs. non-lexical status—can affect the location of phonemic category boundaries.2 In a series of perceptual rating tasks judging changes in voice onset time between /p/ and /b/, Allen and Miller (2001) found that the location of category boundaries shifted as a result of lexical status (i.e. whether or not the item is a word or a non-word). For example, the location of the category boundary between /b/ and /p/ in beef vs. peef would be different than in bat vs. pat, because people are more likely to accept a greater range of phonetic variation for /b/ in situations where the /b/ word is the only grammatical option. However, Allen and Miller also found that, despite the change in category boundary, the mapping of ‘best exemplars’ within those category boundaries remained comparatively stable. 2See Allen and Miller, 2001 and the references therein. 62

Assuming this also holds true for sign languages, it would suggest that for any given contrast, the location of category boundaries between handshapes could change dependent on whether or not meaningful minimal pairs were involved. The implica- tion here is that identification tasks alone are insufficient for determining the existence of a contrast in cases where the lexical status of the test items is not the same. How- ever, since Allen and Miller’s findings also suggest that the best exemplar for each handshape category within a particular range should remain fairly stable, a goodness rating task could be used to further identify the existence and location of contrast. For example, a meaningful contrast is known to exist between the curved-closed joint configuration in and the flat-closed configuration in for perimeter classifiers. Let us assume for the moment that the best exemplars, or phonetic targets, for these handshapes are at and as well. If the uniform lexicon hypothesis is correct, and this contrast persists throughout the lexicon of ASL, I would expect the same handshape preferences (i.e. and ) to be found for items from other lexical components, regardless of whether or not there was a minimal pair available. In other words, for a core sign without a corresponding minimal pair (e.g. TEACH), the uniform lexicon hypothesis would predict either or to be the handshape preferred by signers, matching the best exemplars from the classifier contrast. Conversely, a phonetic target different from those of the classifiers would be an indication that the contrast is not homogenous cross-componentially. Based on the contrast differences already observed, I predicted this latter result would occur.

4.2 Methodology

4.2.1 Stimuli

Three sets of handshapes were tested, referred to here as the O, F, and C groups. Each set constituted a kind of ‘mini-continuum’ of three handshapes—two endpoints, each corresponding to a joint configuration in Brentari’s Prosodic Model (1998) (curved-closed vs. flat-closed for O and F, and curved-open vs. extended for C) 63 and a midpoint handshape with a joint configuration halfway between the other two. Approximations of the handshapes in these three groups are given in Figure 4.1.

Figure 4.1.: Handshape groups tested.

The stimulus items themselves consisted of signs using handshapes from the three handshape ranges, chosen to represent the three main lexical components of ASL. Specifically, initialized signs were used to represent the foreign component, and perimeter classifier constructions (DCLs for the O and C ranges, and both DCLs and HCLs for F) were used to represent the spatial component. Meanwhile, the signs used to represent the core were chosen to be as abstract as possible (i.e. obvious iconic relationships were avoided) in attempts to avoid signs on the lexical boundary with classifiers. As much as possible, items were balanced with regards to place of articulation, movement, orientation, and sign type (one handed, symmetrical two-handed, 64

etc.), however, complete balance was not possible due to the lack of forms that met the experimental criteria. Every sign for a particular handshape set was produced with all three handshapes in that set. For example, the core sign teacher from the O group, was produced once with an , once with a , and once with the midpoint handshape, . The same was done for the initialized signs (e.g. opinion) and classiﬁer constructions (e.g. ‘long-thin-object’). Examples stimulus items are provided in Figure 4.2.

Figure 4.2.: Examples of stimulus items for the O handshape group. 65

Each stimulus item was signed by a Deaf signer and recorded with a digital video- camera. Since this study was only a preliminary investigation of contrast variation, I chose to use an actual signer instead of animation. I felt justified in this decision because, unlike with finer grained tasks such as those in CP experiments, the phonetic spacing between these handshape variants was quite large. In this situation, the disadvantages of using artificial stimuli (i.e. the unnaturalness of the context in which the handshape was presented) outweighed those of using a human signer (i.e. the reduced amount of control in the articulation of each joint configuration). To ensure that the handshapes were as consistent as possible, the signer producing the stimuli spent time practicing each handshape set (endpoints and midpoint) before the actual taping. After the taping, I visually checked the stimulus items for obvious differences, and any found lacking were redone. Each item was signed with neutral (but not ‘flat’) nonmanuals in attempts to produce a natural looking sign without including additional cues that might distract from the manual features. To summarize thus far, four variables were used in creating the stimuli, each with three possibilities:

• 3 short handshape continua (C, O, and F; to be analyzed separately from each other)

• 3 handshape variants per continuum (two endpoints and a midpoint)

• 3 lexical categories (classiﬁer, core, and initialized)

• 3 ‘signs’ (i.e. combinations of phonological parameters) per lexical category

The combination of these variables yielded 27 test items (i.e. videos) per handshape continuum, or 81 test items in the experiment as a whole. In addition, 54 distractor items were added in attempts to prevent participants from formulating a response strategy based solely on handshape variant. These distractors included approximately the same number of signs from each lexical group, used a variety of handshapes—some of which would be considered well formed and others not (e.g. the 66 initialized sign museum was signed using the four-ﬁngered handshape instead of the usual fattened version of the ﬁngerspelled ‘M’, )—and were often minimal pairs of the stimulus items at the level of the whole handshape (e.g. the test item chicago vs. the distractor texas).3 The test stimuli plus distractors totaled 135 stimulus items.

4.2.2 Design

As previously mentioned, this experiment consisted of two tasks. In the ﬁrst task— which I will refer to as the Meaning-Choice Task—a meaning was elicited for each sign to ascertain the participant’s semantic association with each handshape variant, thereby determining (and/or verifying) the existence of any meaningful contrasts within the range. This determination consisted primarily of a binary choice of ‘A’ or ‘B’ based on the expected meaning(s) at either end of the continuum, but it also included the option to write in a separate answer. Because of minimal pair problem discussed in Section 4.1, for some items I expected only one answer per continuum, while in others I expected two. In the second task—the Form-Rating Task—participants performed a goodness rating task (ratings 1-5) for the handshape in each stimulus item based on their chosen meaning from the Meaning-Choice Task. This task was included to determine how good a ﬁt the test handshape was to their best exemplar for that meaning. These ratings were then used in cases where the Meaning-Choice Task responses yielded no contrast, allowing comparisons with the best exemplars of previously established contrasts. The items themselves (including distractors) were presented via PowerPoint slides. For each item, a pair of slides was created to be shown in order, the Meaning-Choice

3The English translations of these handshape-level minimal pairs were often used as an option in the Meaning-Choice task described below. 67

Task ﬁrst and then the Form-Rating Task, as illustrated in Figure 4.3.4 In cases where a minimal pair (or near minimal pair) was available within the test range for a given item (e.g. chicago and never), both members of the pair were provided as choices. In situations where there was only one possible meaning for a given range of handshapes, (e.g. teach), the second meaning choice provided was, whenever possible, a form that diﬀered from the item in handshape alone.5 The stimulus items and their meaning pairs for the Meaning-Choice Task are listed in Appendix A.

Figure 4.3.: Example presentation slides for each task.

4For the Meaning-Choice Task, choices were provided in English since ASL does not have a widely accepted written system of its own, and because the American Deaf community is (to varying degrees) bilingual in ASL and (written) English. 5One pair, strange and overlook, diﬀered in orientation in addition to handshape, since the handshape of the latter form was very close to HS 3 of the C handshape range. 68

4.2.3 Participants

Twelve Deaf ASL signers participated.6 Of these, 6 had Deaf families and learned ASL from birth, while the remaining 6 learned ASL by or before the age of 5. At the time of their participation, all 12 attended school or worked at Gallaudet University in Washington DC (where the data was collected), however, they grew up in a variety of locations across the country including the East and West Coasts and the Midwest. Ages of the participants ranged from 19-56 years, (7 below 30 years and 5 above), and all participants had at least some college education, with 4 having obtained post-baccalaureate degrees. Participants were paid for their participation.

4.2.4 Procedure

The 135 pairs of stimuli and distractor slides were randomized five times resulting in five lists, and each list was divided into two blocks (67 and 68 items respectively) with a break in between. The order of these blocks was then reversed resulting in ten possible presentation orders.7 In response to the first slide of a particular pair, participants were instructed (in both ASL and written English) to pick the best English meaning for each sign from the choices given. If they felt that the sign meant something different, they were asked to provide a better meaning. Since classifier constructions can represent multiple items and have no good English equivalent out of context, they were told the following: “Sometimes, you will see signs that can be used to describe many possible objects or situations. When this happens, see which English word or phrase makes the most sense. If neither choice is good, think of an example yourself and write it in the blank.” For the second slide, they were instructed to rate how ‘good’ the sign was based on what the handshape looked like.

6Three additional participants were excluded—one turned out not to meet the qualiﬁcations, and two either did not understand the task, or did not rate meanings critically enough to be useful (e.g. rated almost all signs as ‘5’). 7While the presentation order of the slides was randomized, the order of the meaning choices on each slide was not. 69

In the instructions, they were speciﬁcally asked to focus on handshape and not to include factors such as nonmanual markers or movement in their rating decisions. Participants circled (or wrote in) their responses to each task on a prepared answer form. The experiment as a whole lasted about 45 minutes.

4.3 Results

Because the test items differed in the number of ‘correct’ meanings possible for a given handshape range, only the responses for one meaning—which I will call the Targeted Meaning—were used in statistical comparisons. In cases where there was no minimal pair, the Targeted Meaning is defined as the only ‘correct’ meaning for the signed form. Where a minimal pair or morphologically-related pair did exist (i.e. if both choices provided were logical answers given the handshape range being tested), the Targeted Meaning was the choice expected for the rounder of the two endpoints.8 For example, if participants were given the two choices ‘Cafeteria’ and ‘Bachelor’ for a particular stimulus item, ‘Cafeteria’ would be the Targeted Meaning (regardless of which handshape was used in the stimulus item itself) because the normally associated with that sign is rounder than the associated with the sign bachelor. Likewise, if the two choices given for a classifier form were ‘cardboard tube’ and ‘sheet of cardboard’, the former meaning would be the Targeted Meaning because the handshape typically used to represent the shape of a cardboard tube is rounder.

4.3.1 O handshape group

To compare the number of the Targeted vs. non-Targeted Meaning responses for the O group from the Meaning-Choice Task (Figure 4.4(a)), I used a Fisher’s Exact test. This test showed that the classifier group was significantly different than

8These expectations were formed based solely on prior knowledge of ASL vocabulary and not on the experimental results themselves. 70

100 4 4 4 u u u e 80

60 e

Classiﬁers 20 Core 4e Initialized u 0

Percentage of Targeted Meaning Responses e HS1 HS2 HS3 Handshape Variants

(a) Percentage of Targeted Meaning responses for all lexical groups in the Meaning- Choice Task. 5 4 4 4 4u u

3 u

2 Average Rating

1 Core 4 Initialized u 0 HS1 HS2 HS3 Handshape Variants

(b) Average ratings of Targeted Meaning responses for Core vs. Initialized signs in the Form-Rating Task. Error bars represent the standard error of each average rating.

Figure 4.4.: O handshape group results. 71

both the core and the initialized groups at HS variants 2 and 3 (p < .01 for each comparison, α = .05). The core and initialized groups were not significantly different from each other in number of Targeted Meaning responses since, unlike the classifiers, there was only one correct answer for these items. I then performed a mixed model analysis of variance (ANOVA) to determine whether or not differences existed between the core and initialized groups. In this analysis, the outcome variable was the average rating for the Targeted Meaning (from the Form-Rating Task), and the fixed effects of primary interest were handshape variant and lexical group. (Adjustment for variability between participants was made by including it as a random effect in the model.) To adjust for three comparisons (i.e. HS 1 core vs. HS 1 initialized, HS 2 core vs. HS 2 initialized, and HS 3 core vs. HS 3 initialized), I used a Bonferroni adjustment with α = .05/3 = .0167. Using the mixed model analysis, the average ratings of core vs. initialized forms (Figure 4.4(b)9) were found to be significantly different at HS 3 (t = 3.74, df = 197, p <.01).

4.3.2 F handshape group

For the F group, the Fisher’s Exact test showed that the number of Targeted vs. non-Targeted Meaning responses for the classifier forms was again significantly different than the core and initialized groups at HS variants 2 and 3 (p < .01, α = .05) (Figure 4.5(a)). As with the O handshapes, the core and initialized groups were not significantly different from each other, resulting in a second comparison of these groups using the mixed model ANOVA of average ratings (Figure 4.5(b)). Despite this group of handshapes having the same joint configurations in the selected fingers as the O group, the average ratings for each HS variant in the F group were not significantly different for any HS variant at the level α = .0167.

9The mixed Model takes the number of data points into consideration in its calculation of diﬀerence. To better represent the relationship between the number of Targeted Meaning responses (which can vary in cases where more than one meaning was possible) and the average ratings for those responses in the corresponding graphs, non-Targeted Meaning responses were assigned a rating of zero for graphing purposes only. 72

100 4 4 4 u u e u 80

60 e 40

Classiﬁers 20 Core 4e Initialized e u 0 Percentage of Targeted Meaning Responses HS1 HS2 HS3 Handshape Variants

(a) Percentage of Targeted Meaning responses for all lexical groups in the Meaning- Choice Task. 5 4 4 4 4 u u u 3

2 Average Rating

1 Core 4 Initialized u 0 HS1 HS2 HS3 Handshape Variants

(b) Average ratings of Targeted Meaning responses for Core vs. Initialized signs in the Form-Rating Task. Error bars represent the standard error of each average rating.

Figure 4.5.: F handshape group results. 73

4.3.3 C handshape group

Because items in the C group varied with respect to how many ‘right’ answers were possible per item across the handshape range, a statistical comparison was only made for items with more than one meaning. The resulting Fisher’s Exact test showed that the number of Targeted vs. non-Targeted Meaning responses for classifier and core forms was significantly different at HS 1 (p = .048, α = .05) and significantly different than the initialized forms at HS variants 1 and 2 (p < .01 for each).10 (The difference at HS 1 in this case was due in part to write-in answers indicating a third meaning associated with that end of the handshape range.) As with the other two handshape groups, the responses of the core and initialized groups were not significantly different using the Fisher’s Exact test, requiring the additional mixed model ANOVA. The average ratings for core vs. initialized forms in the C group (again, comparing only those forms with minimal pairs) were found to be significantly different for HS 1 (p = .013, α = .0167) but not for the other two handshape variants.

4.4 Discussion

In comparing the results of the Meaning-Choice and Form-Rating Tasks, significant differences were found between the lexical groups. The results of the Meaning- Choice Task results show that the classifiers were perceived differently from core and initialized forms in all three handshape ranges by virtue of having more possible meanings per range. For the O handshapes, there were meaningful differences between the two endpoints in the DCLs used—HS 1 represented round objects and HS 3 represented flat objects—while all handshape variants in the core and initialized forms elicited the same meaning per lexical item. Likewise, for the F handshapes, classifiers were the only group to elicit two separate meanings at the endpoints, this

10Even though the percentage of Targeted Meaning responses for both the core and initialized items were at 100%, p-values are diﬀerent because the core group only included one item with a minimal pair. 74

100 4 u4 u 80 e e 60

Classiﬁers 20 Core 4e u Initialized u 0 4

Percentage of Targeted Meaning Responses e HS1 HS2 HS3 Handshape Variants

(a) Percentage of Targeted Meaning responses for all lexical groups in the Meaning- Choice Task. 5

4u 4 4u

2 Average Rating

1 Core 4 Initialized u 0 4u HS1 HS2 HS3 Handshape Variants

(b) Average ratings of Targeted Meaning responses for Core vs. Initialized signs in the Form-Rating Task. Error bars represent the standard error of each average rating.

Figure 4.6.: C handshape group results (those with minimal pairs only). 75 time round DCLs at HS 1 vs. precision HCLs at HS 3.11 Finally, the Meaning-Choice Task results for the C handshapes showed a contrast between extended (HS 3) and not extended (HS 1 and 2) joint configurations in all lexical groups. Additionally, a higher degree of uncertainty (lower ratings and/or question marks) along with write- in answers (e.g. ‘binoculars’ instead of the provided choices ‘look through round window’ or ‘look through square window’) indicate the potential for a third category in the C classifiers at the rounder endpoint (HS 1). This additional contrast is supported by Supalla’s (1986) observation of a four-way distinction between round shape depicting classifiers, two of which were joint configurations similar to HS 1 and 2 of the C range.12 While future research is needed to know for sure, Supalla’s observation supports the idea that there is a prominent contrast between these two curved-open joint configurations. The most likely explanation for the results thus far lies in the kind of contrast utilized by the various components—core and initialized signs, as I have shown, utilize distinctive contrasts, while perimeter classifiers primarily utilize prominent ones. This prominence could explain why the classifier responses to the Meaning-Choice Task were significantly different from the other two groups in all cases, despite the handshapes being common to all three lexical substrata. A comparison of the ratings for core and initialized signs from the Form-Rating Task also yielded slight, but significant, differences. The results of the O range suggest that, while signers found all three handshapes acceptable for the signs in question, (all average ratings were above 3.4), (HS 3) is less acceptable in the

11Although a contrast was found in the F classifiers, the choice of the round descriptive classifier for HS 1 was stronger than that that of the handling classifier for HS 3 (97% vs. 72% respectively). This is likely due to the lack of shape iconicity in the handling form (Section 3.5.2). The results do, however, show a contrast, which could be due to competition in the lexicon itself (i.e. the handling form becomes ‘flatter’ to distinguish itself from the rounder form) or could be due to the competitive nature of the binary choice of the experiment (if a variant was not an acceptable enough ‘round object’, it was judged by default to be the handling classifier which has no specific joint specification). Further experimentation is needed to determine which is the case.

12The other two joint conﬁgurations mentioned by Supalla were for small round objects, and a ‘squeezed’ version, , depicting very small round objects. 76

initialized forms than in the core forms. Similarly, the results for the C handshapes suggest that, of the two non-extended handshape variants, HS 1 (i.e. the one with a closer resemblance to the written letter C) is more acceptable in the initialized forms than it is in the core forms. In contrast, no significant difference was found in the average ratings of the F handshapes. The fact that these differences were found for O and C rather than in F handshapes may stem from an iconic relationship between and and the roundness of the written letters ‘o/O’ and ‘c/C’ that is not present for ‘f/F’. It is worth noting, however, that although there are not enough data points to run convincing statistical tests, the average rating scores for get-an- ‘F’ (the form with the most direct semantic tie to the written letter it represents) also showed a preference for the rounder handshape, which is the handshape found on most ASL fingerspelling charts. This indicates that perhaps there is an iconic relationship at work in the initialized forms—either to the written form or to the canonical fingerspelled handshape—but that this relationship is much weaker than the shape iconicity in the classifier handshapes because of its lack of prominence. One reason why the difference between core and initialized preferences was not more pronounced might be that the initialized signs were seen as less foreign by virtue of their common handshape—i.e. the shared status of O, F and C handshapes across the lexicon may allow these forms to gain core status more easily in the mind of the user. (This may be especially true in the case of initialized F signs, since these handshapes lack the aforementioned resemblance to the written counterpart.) Unless the connection to an English gloss is explicitly made, it is possible that the signer would not consider some of these signs ‘foreign’ unless there was some other sociolinguistic reason for them to remain so (e.g. use in only an academic register). In future experiments of this kind, it might be useful to gain lexicality judgments of potential stimulus items from a separate group of participants to better determine strata membership before perceptual data is obtained. To summarize thus far, the experimental results indicate that a larger number of handshape contrasts exist in forms where joint configurations have a strong iconic 77

Figure 4.7.: Summary of contrasts across lexical components.

relationship to shape (i.e. classifiers, where contrasts are prominent) than where they do not (i.e. core and initialized forms, where contrasts are distinctive). Figure 4.7 illustrates these contrasts. The results also indicate that the phonetic preferences within the phonemic boundaries of these handshapes differ cross-componentially— slightly between core and initialized signs, and more substantially between classifiers and core items (since the best exemplars for core forms seem to be midpoints instead of endpoints).

4.4.1 The minimal pair problem revisited

We now return to the problem of asymmetry with regard to minimal pairs across the stimuli. How do we know that the differences found here are the result of lexical 78 affiliation and are not based in the presence or absence of lexical competition? First, we can compare the average ratings of core items without minimal pairs across the three handshape groups, O, F and C. If the differences are due only to a lack of competition, I would expect the items without competitors to be rated in similar ways. Figure 4.8 shows that this is not the case. The items from the C range, where a contrast was found between the midpoint and the extended endpoint in all cases, were rated much lower at HS 3 than at any of the handshape variants in the O and F groups, indicating that this extended variant was not (or, at least, was much less) acceptable for those forms.

5 × ?5× ?5 4 ?5

2 × Average Rating C × 1 F 5 O ? 0 HS1 HS2 HS3 Handshape Variants

Figure 4.8.: Average ratings in Form-Rating Task for core items without minimal pairs for all three handshape groups.

Let us now return to the example prediction from Section 4.1. The results of the Meaning-Choice Task for the O group verified a morphological contrast between the joint configurations in and for perimeter classifiers representing round and flat objects, and the extremely high and low percentages of Targeted Meaning responses 79 at each endpoint for that task indicate that they were the preferred handshapes in the contrast (within the limited range tested). The uniform lexicon hypothesis predicted that the endpoints should likewise be preferred by signers in all other lexical components. However, as already shown, that is not what the results indicate; for core signs, for instance, signers preferred the midpoint handshape . This lends further support to the idea that the contrast differences are due to lexical affiliation.

4.4.2 Predictions for language change

What about forms on the border between lexical components, for instance, the core forms in ASL that were lexicalized from the classifier system, but that still appear to maintain an iconic connection to shape (e.g. those studied by Boyes Braem, 1981)? The category boundaries and preferences of such forms are as yet unknown, but predictions can be be made based both on the current set of results, and evidence from dictionary sources. As Frishberg (1975) illustrates, signs tend to become less iconic and more abstract over time. If meaningful contrasts between the more deliberate joint specifications (i.e. the endpoints of the current handshape sets) are indeed tied exclusively to the morphological representations of shape in the classifier system as these results suggest, then we would expect such signs to lose these contrasts as they are grammaticized over time. If this is the case, there should be few, if any frozen lexical items with these handshapes that have an obvious connection to the perimeter shape of the referent. In fact, core signs with transparent ties to shape in these three handshape ranges do seem to be quite rare. For example, the vast majority of ‘O’ signs from DASL (Stokoe et al., 1965) (which include handshapes ranging from to ) likely have their origins in handling classifiers and no longer appear to have shape associated with them (if they ever did). Further, with very few exceptions, the ‘O’ signs that do not seem to be derived from HCLs and have on the non-dominant hand, have a point of contact on the inside (i.e. palmar side) of the handshape (e.g. vote, 80

gasoline). In these cases, the rounder variant may be preferred only because it allows for the widest contact area and not because of a shape correspondence with the original referent. Of the signs that do have an obvious connection to a round shape, some seem to be borrowed from the writing system, (e.g. nothing from zero, and percent from the circles in ‘%’); if tested, I would expect these signs to behave similarly to the initialized O-signs in the experiment, that is, they would prefer the and the over , with all being acceptable to a large extent.13 So far, I have found very few actual instances of frozen signs in which the O handshape may still be indicative of object shape (e.g. owl and octopus, which could overlap with initialized signs). Similarly, there were very few ‘F’ signs in DASL that had an obvious connection to round objects (e.g. hole, diploma, wristwatch). Of the signs that seem to be iconically motivated in some way, most were either initialized or appeared to have their origins in precision handling classifiers (and thus had no inherent joint specification). In the ‘C’ signs, a transparent relationship to shape was more common than in the other handshape groups (e.g. drink, sun), but as we saw from the experiment, a contrast does appear to exist between curved and extended joint configurations in core forms as well as in initialized signs and classifiers. It is unknown, however, which handshape variant (HS 1, or HS 2, ) is preferred for these signs. Based on my analysis so far, I would expect the rounder handshape ( ) to be preferred rarely if at all, and only in signs that can be deconstructed back into perimeter classifier forms indicating shape.14

13Further investigation is needed to know how these signs tend to be articulated, but in pilot studies from this experiment, for the signs none and know-nothing (both presumably derived from ‘zero’)

participants preferred followed closely by , with lower (but not low) ratings for . 14This rarity of purposefully round joint conﬁgurations in the core (as opposed to the more relaxed midpoint handshapes) may not be restricted to ASL. van der Kooij (2002) mentions that all seven signs in her NGT lexical data using a static O handshape were either motivated by a number or letter or by the shape of the referent. She goes on the mention that one-handed signs using C handshapes were usually indicative of holding a round object (e.g. a cup) or were used in initialized C signs. 81

More research is needed in this area to find out 1) how many core forms are specified for the rounder joint configurations like the HS 1 variants of the experimental handshape ranges, and 2) how much handshape variation is acceptable for those signs. Based on the differences in ratings for initialized signs, however, I would expect grammaticalized perimeter classifiers to gradually lose their prominent joint contrasts, maintaining (at least for a while) the original variant as a best exemplar of the new category (especially if the frozen form still closely resembles a productive classifier construction). Ultimately, I would expect a move toward the easier to produce midpoint handshapes in all cases, or, put in terms of OT, once signs enter the core lexicon, markedness constraints based on articulation (in this case, constraints against the more difficult, deliberately round or flat variants) become higher ranked than iconicity-based constraints. This analysis will be further developed in Chapter 5.

4.4.3 Future experiments

As mentioned, this experiment was only meant to be a preliminary mapping of handshape contrasts across the lexicon—essentially it was an attempt to figure out whether more exploration of these issues was warranted. Now that some indications of differences have been found, many more detailed experiments will be needed to confirm these results and gain a more detailed understanding of contrast differences in these forms. For example, CP experiments should be run testing more extensive and more precisely controlled handshape continua. In future studies, since larger ranges of handshapes will be tested, some sort of precise measurement and/or artificial manip- ulation will need to be utilized to ensure that the joint configurations are completely consistent across items. In addition, more attention should be paid to the specific lexical affiliation of stimulus items—different kinds of classifier handshapes (e.g. DCL vs. HCL) should be acknowledged, and initialized items should be judged for ‘En- glishness’ by a separate group of participants to determine how foreign they seem to 82 the actual signing community. More detailed goodness rating tasks based on perception as well as production tasks measuring variation in articulation should also be performed to determine more accurately where the best exemplars lie for particular items. Furthermore, frequency effects should be taken into consideration to determine whether or not preferences truly are related to lexical status (cf. Exemplar Theory; e.g. Pierrehumbert, 2001). The use of reaction times in priming experiments may also prove informative in determining whether or not certain variants tend to prime the wrong type of sign, leading to potential confusions. For example, would signing TEACH with a round handshape (instead of the (seemingly) preferred ) prime descriptive perimeter classifiers for round objects to a greater extent than signs within the same substrata using good exemplars for handshape (e.g. EAT with a )? There is much still to be discovered in this area.

4.4.4 Theoretical implications

On a final note, these results could have implications for theoretical models. For example, the differences between core and initialized forms in the Form-Rating Task but not the Meaning-Choice Task could be represented using a theoretical analysis proposed by Avery and Idsardi (2001) for spoken languages. In this analysis, contrast occurs at the phonological level, while the inclusion of the phonetic details for the pronunciation itself (termed completion by Avery and Idsardi) only occurs after the phonology has been mapped to the phonetics. Under this approach, identical phonological contrasts across multiple languages can then differ in their phonetic details based either on universal or language-particular factors, resulting in a more economic and less redundant representation. Perhaps a similar analysis could be used to represent language-internal phonetic differences as well. Results such as these could also aid in the simplification of sign language feature geometries such as Brentari (1998). For example, signers’ responses indicate a lack of contrast between any of the O handshapes in core forms despite the differences 83 in joint specification. Although not explicitly tested here for contrasts with handshapes outside of the given range (e.g. vs. ), the only configuration-related characteristic held in common between all three O handshapes is contact between the fingers and the thumb, implying that it may be the only phonologically relevant feature.15 This further suggests that the Joints structure for the representation of core forms could be simplified, replacing part or all of the structure with a feature like [contact].16 The details of such theoretical revisions, however, will depend upon more studies being done on contrast differences throughout the lexicon.

15This is easily tested: when contact is taken away, these signs are no longer grammatical or they change their meaning. 16The initialized forms also indicated a lack of contrast in the O group, but the more marked nature of foreign handshapes in general argues against theoretical simpliﬁcation for these forms. 84

5. OT ANALYSIS OF CONTRAST

This chapter addresses the second hypothesis of this work, that is, that the differences in handshape contrasts shown in Chapters 3 and 4 can be explained as a confluence of pressures on language using Optimality Theory. I begin with a basic introduction to OT and its uses—both generally and with respect to contrast, lexical stratification, and iconicity. I then propose constraints based on articulatory and perceptual Markedness and Faithfulness to language external factors (i.e. iconicity and foreign borrowings). Finally, using these constraints, I explain how OT can be used to explain many of the contrast differences described in earlier chapters.

5.1 Background

5.1.1 OT in general

Optimality Theory (Prince and Smolensky, 1993) as a theory considers language to be a system of conﬂicting forces expressed by means of linguistic constraints. The two main forces in this approach are Markedness, which assumes systems prefer unmarked structures, and Faithfulness, which assumes systems prefer to maintain lexical contrasts. Markedness is typically grounded in factors like articulation and perception, while Faithfulness is primarily concerned with making sure that the system has enough contrasts available to convey the necessary diﬀerences in meaning. The optimal candidate output form is then determined according to language particular rankings of these constraints.1

1See Kager, 1999, and McCarthy, 2002, for more information on OT. 85

5.1.2 OT and contrast

Over the past fifteen years, OT has been used to represent many aspects of spoken language phonology. In his dissertation, Edward Flemming (1995; reprinted 2002) used this theory in a new way to try to represent the different inventories of phonological contrasts found across spoken languages. He called this new approach Dispersion Theory.2 Here, I will follow recent work in Dispersion Theory (e.g. Flemming, 2002; Padgett, 1997, 2003; Ito and Mester, 2003b) in accounting for contrast differences in ASL, DSGS, and HKSL. Flemming summarizes the core of Dispersion Theory by claiming that languages have three goals in selecting their phonemic contrasts (2002:4):

• Maximize the number of contrasts

• Maximize the distinctiveness of contrasts

• Minimize articulatory eﬀort

The first of these goals, maximize contrasts, is based on the fact that a large number of phonemic contrasts allows for a greater number of possible linguistic combinations. The second goal, maximize distinctiveness, observes that the more perceptually distinct phonemes are from each other, the easier they are to process by the recipient. In other words, languages prefer to keep perceptual confusion to a minimum. Finally, the last goal, minimize articulatory effort, speaks to the human tendency to favor motor behaviors (linguistic and non-linguistic) that require the least anatomical effort. Dispersion Theory uses these goals as a basis for OT constraints and assumes that cross-linguistic variation stems from differences in how languages choose to mediate between them, i.e. variation can be represented via different rankings of the constraints. Flemming’s major departures from traditional OT were that 1) his constraints were based largely on auditory (i.e. perceptual) distinctiveness as opposed to

2Flemming’s work was based largely on the work of Lindblom (e.g. 1986; see Flemming for further references.) 86 primarily articulatory factors, and 2) his constraints addressed the well-formedness of contrasts rather than of segments or words in a vacuum (i.e. the candidates themselves consist of sets of contrasts and not single words). More details about the mechanisms used by Dispersion Theory will be given in the sections that follow. Flemming’s work (as well as the work of those who followed him) focused on auditory distinctiveness and articulation of speech sounds, but there is no reason why the goals of his theory should not work equally as well for describing the type of mechanisms involved in phonemic contrasts in sign languages, particularly those that have been described in previous chapters. Following the tenets of Dispersion Theory, I base the constraints of my OT analysis on perceptual and articulatory pressures present in visual languages (grounded wherever possible in anatomical observations—e.g. Ann, 1993; Greftegreﬀ, 1993—and visual perception experiments—e.g. Lane et al., 1976; Stungis, 1981), combined with Maximize Contrast constraints similar to those used by Flemming for spoken languages. Cross-linguistic and cross-componential variation in handshape contrasts will be represented by altering the rankings of these constraints.

5.1.3 OT and lexical stratiﬁcation

Using OT to represent cross-linguistic differences via constraint rankings is not unusual, but what about the cross-componential differences shown in Chapters 3 and 4? Both Ito and Mester (e.g. 1995b, 1995a, 2001, 2003a) and Brentari and Padden (2001) use OT in their analysis of lexical stratification, the former to represent a variety of phonological behaviors in spoken languages, and the latter to support further stratification of the foreign component of ASL based on syllable structure and prosody constraints. I intend to add to this literature by using OT to represent some of the language internal differences in handshape contrast across the three languages of this study. 87

Ito and Mester (1995a) summarize their approach towards the representation of cross-componential behavior in the lexicon as follows:

[I]n a substantial class of cases constraint reranking can be limited to a reranking of Faithfulness constraints, within an otherwise invariant ranking order of constraints. It is this limit on reranking that gives rise to the core-periphery structure, appearing observationally as a gradual “weaken- ing” of the “force” of various wellformedness constraints. (206)

By reranking only Faithfulness constraints (in this case, Faithfulness to foreign borrowings) and leaving the ranking of all other constraints untouched, Ito and Mester have successfully analyzed examples of lexical stratification from multiple languages, including Japanese (1995a, 1995b), Russian (1995a), German and Jamaican Creole (2001). In this work, I show that Ito and Mester’s approach is also useful for analyzing sign languages. By establishing fixed rankings for language internal Markedness (perceptual and articulatory) and Max Contrast Faithfulness constraints, and only re-ranking Faithfulness constraints referencing external forms (i.e. borrowings from other sources), I attempt to represent the differences in handshape contrasts between lexical components both within and across the three sign languages under discussion.

5.1.4 OT and iconicity?

As mentioned in Chapter 2, researchers have long debated the role (or lack thereof) that iconicity plays in the phonology of sign languages. Most of these arguments treat the issue as an ‘all or nothing’ situation—either iconicity is an external force that nullifies attempts at phonological formalization for the forms it affects (e.g. Sandler, 1996), or it is largely irrelevant and alterations can be explained without making direct reference to it (e.g. Supalla, 1982). The few who have tried to incorporate iconicity into some sort of formal representation (e.g. Boyes Braem, 1981; van der Hulst and van der Kooij, 2006) have typically done so by relegating iconic features 88 and more purely phonological features to separate levels of the representation. OT affords us an opportunity to reach a middle ground in this debate, acknowledging that iconicity must play a part as a motivating force in these forms, but treating it as only one of the conflicting forces at play in an already established architecture. I argue here that the desire to be faithful to certain aspects of the real-world adds an additional pressure on the phonological system, but that this pressure—like that of articulatory and perceptual Markedness—is not inviolable. This analysis does not address why certain kinds of iconicity are chosen over others to represent real-world objects or concepts (cf. Taub, 2001); it instead considers the impact that an iconic relationship has on the phonological form. But how is this to be done? Perlmutter (2006), in a critique of van der Hulst and van der Kooij (2006), disagrees that iconic forms should be explained away by phonetic implementation rules at the semantic level (Section 2.4), and says instead that such features are phonological. Among the alternatives he suggests is an idea he calls the Loan Hypothesis:

The Loan Hypothesis: Iconicity is a form of borrowing. Iconic signs in sign languages are loan words. (320)

This hypothesis predicts that, phonologically, iconic signs are like loan words in spoken languages in two ways: 1) they can violate the patterns/constraints of the native vocabulary, and 2) they are expected to conform to more of those native patterns the longer they are in the language. As support of this hypothesis, he cites Frishberg’s (1975) observation that iconic signs become less iconic and conform more to ASL’s phonological constraints over time, much as loan words tend to do the longer they are in a language.3

3Perlmutter further maintains that signs that maintain their iconicity (i.e. do not eventually conform) either: 1) should be analyzed as a separate stratum of the lexicon following the work of Ito and Mester (e.g. 1995a and Brentari and Padden, 2001) or 2) they can be considered outside the phonological system entirely. I adopt the former of these two alternatives in the current analysis (see Section 5.4.3). 89

As a continuation of Perlmutter’s idea, I argue here that the phonology is not directly accessing the semantics, per se (as others have claimed), but instead that it is actually borrowing from physical aspects of the visual form itself, articulating comparable visual structures with the hands, and then filtering the resulting forms through the linguistic constraints of the system. One potential problem with using the Loan Hypothesis in this way is that in traditional OT, the input (to which an output form is faithful) is usually thought of as a linguistic construct. More recently, the concept of ‘input’ in OT has become a controversial and debated topic. It was originally conceived as a lexical representation, similar to the underlying representations used in work such as Chomsky and Halle (1968)—although in OT, Richness of the Base (part of the basic architecture; see Prince and Smolensky, 1993; Kager, 1999) implies no restrictions on the input. Prince and Smolensky (1993) say that Richness of the Base “holds that all inputs are possible in all languages” (209)—i.e. only the output is restricted by language particular constraints. Usually this richness is interpreted in terms of linguistic features (e.g. no restrictions with respect to nasality/orality contrasts in vowels; Kager, 1999) or lexical properties relating to various segments of the grammar (semantic features, syntactic features, etc.). However, more recently researchers using OT have questioned the need for inputs altogether (e.g. Flemming, 2002 and work cited therein), while others treat the input as a more general type of comparison form. This latter approach (explained in further detail in Section 5.2.2) is used for historical comparisons where the more contemporary form (the output) is determined in part by a comparison to an historical counterpart (the input) (Padgett, 2003). But could this richness be extended even further? What if, in this capacity as a comparison form, the acoustic properties of a bell (as in Taub’s example from Chapter 2) were used as input? English-specific rankings between articulatory and perceptual Markedness constraints and Faithfulness constraints referencing the acoustic properties of the sound would result in the optimal output /dIN/. Likewise, in sign languages, what if the round shape of a ball was treated as part of the input, 90 resulting in an output with a curved joint configuration. (I return to this issue in Section 5.2.2.) More work must be done before such an alteration can be officially considered for the theory, but this work starts down a similar path. Utilizing the basic architecture of OT, I propose that iconic forms are the result of highly ranked Faithfulness constraints referencing aspects of the original, real-world objects despite their non-linguistic nature. In this way, this analysis treats forms entering the lexicon from outside sources (be they foreign linguistic inputs or perceptual relationships to physical objects) in a similar fashion. The distinction between attributing iconicity to semantic motivation and considering it to be borrowed in part from the physical characteristics of the real world is a subtle one, but I feel that it is an important distinction to make. By placing the emphasis on parallel visual dimensions associated with both the hand and the object being represented via Faithfulness constraints instead of on a particular semantic meaning, the theory is able to incorporate these elements directly into the phonology instead of depending on other parts of the grammar. However, since the ‘input’ in this case (i.e. the real-world dimensions) is not of a linguistic nature, determining what Faithfulness implies for the handshape is not a straightforward task. It is a bit like comparing an apple to an orange in a strict diet; both certainly qualify as a fruit, but the details of that qualification can be difficult to define. In the same way, the geometrical ‘roundness’ of a coin and the ‘roundness’ of the shape created by the fingers of an (‘F’) handshape are, in actuality, fairly different when you examine them in detail, as seen in Figure 5.1. This mismatch is even more apparent when comparing real-world degrees of thickness with the limited number of finger combinations used to represent them. For example, the thickness of the index finger and thumb is hardly comparable the actual thickness of a coin, even if the size difference of the two forms is taken into consideration. So how is ‘Faithfulness’ to be determined in this relationship between referent and output? To follow Perlmutter’s Loan Hypothesis, we need to find visual aspects that the two types of forms (i.e. objects and hands) have in common. In Section 5.2 I at- 91

Figure 5.1.: Diﬀerences in the conceptualization of ‘round’ within an F handshape (left) and in the shape of a coin (right).

tempt to do just that, formulating constraints based on shared physical characteristics in conjunction with perceptual principles.4 I should note here that I do not believe that all instances where handshapes share a physical resemblance to their referent should be treated equally with regard to iconic constraints. I instead concentrate on forms in which the iconic relationships have morphological status, that is, the relationships resulting in the kinds of prominent contrasts discussed in Chapter 3 and 4. Of course, not all prominent feature contrasts are direct iconic borrowings (e.g. aspectual movements, the ‘crashing’ of SCLs discussed in Section 3.5.3 ), but I propose that all iconic borrowings are prominent contrasts. Conversely, I argue that distinctive contrasts, even if they were, historically, iconically motivated, are no longer determined via highly ranked iconic Faithfulness constraints. 4Ultimately, I hope that experimental data will be available for use in these sorts of comparisons (e.g. measurements of degrees of joint ﬂexion used to represent speciﬁc object shapes and sizes), but for now, we will have to make do with approximations and general observations. 92

5.1.5 Challenges to the analysis (and disclaimer)

There are two major difficulties in using this theory with sign languages as opposed to spoken languages. First, Flemming bases much of his analysis on the relative acoustic dimensions of auditory contrasts (e.g. formant frequency values), but the visual equivalents for sign languages are not yet well understood; little is known about the visual mechanisms involved with perception with regard to sign articulation. For instance, most of the non-linguistic literature on shape perception deals with the perception of geometrical shapes or novel objects, and not the shape of the hand. Further, the small amount of research that has been done on the perception of sign language handshapes (e.g. Lane et al., 1976; Stungis, 1981) has been limited to forms found in fingerspelling and/or the core of ASL, leaving out many of the additional contrasts found in the classifier component and in other languages. I must therefore rely on these works in addition to the work conducted here on perceived differences in contrast (presented in Chapter 4), limited though it may be. On the articulation side, too, we have very little information on handshape. Only a relatively small amount of work has been done comparing the anatomical structure of the hand with linguistic handshape distribution in sign languages (e.g. Mandel, 1979, 1981; Boyes Braem, 1990; Ann, 1993, 2006; Greftegreff, 1993), and all of the work I have found thus far is based on observational and dictionary data rather than on physiological measurements during sign production. The second major difficulty is our lack of information about the nature of the contrasts themselves. While a great deal of information is available to many spoken language linguists regarding the inventories and contrasting segments of the languages they study via experimentation, historical texts and extensive fieldwork, sign language linguists have only just begun to explore the phonological and phonetic boundaries of segments in their languages with any detail. Until more of these experiments have been undertaken, we will continue to have an incomplete picture of the kinds of articulatory and perceptual factors in conflict 93

within the phonology. As a consequence of these uncertainties, I will base my analysis on both the work presented in the previous two chapters and general observations, with the expectation that expansions will be made as more information becomes available. In this work, I do not claim to present ﬁnal answers to the problem of diﬀering contrasts in sign language handshapes, but rather, I present an approach that I hope will be useful in the ongoing pursuit of these answers.

5.2 Constraints

In general, my OT approach uses four main types of constraints: articulatory and perceptual Markedness constraints, and featural and iconic Faithfulness constraints. I explain each in turn. (See Appendix B for a full list of the proposed constraints.)

5.2.1 Markedness constraints

In describing the signiﬁcance of Markedness constraints in his theory, Flemming (2002) states:

[D]ispersion theory claims that markedness is a property of contrasts as well as of segments. Given the goal of maximizing the distinctiveness of contrasts, a contrast is marked to the extent that it is not auditorily [perceptually] distinct. Individual segments are marked to the extent that they are articulatorily diﬃcult to produce. (36)

In other words, in Dispersion Theory, segment Markedness is based on articulation, and contrast Markedness is based on the perceptual distance between the contrasting segments. This distinction will become more apparent in the constraint descriptions that follow.5 5Flemming uses the term ‘segment’ in a generic way, but his analysis tends to focus on particular properties of these units (e.g. formant frequencies). I do the same here by focusing on particular properties of handshape contrasts (e.g. SF combinations). 94

Articulatory Markedness constraints

Articulatory Markedness constraints in these analyses (i.e. Markedness at the level of the segment), are typically context-free constraints and are found in one of two basic forms, shown in (1).6

(1) Articulatory Markedness constraints:

X : Segments in the output must have a given articulation, X.

*X : Segments in the output can not have articulation X.

Most of the articulation-based constraints proposed here are based on the anatomy of the hand and how it affects the range of independence for the various digits.7 The index finger, pinky and thumb each have independent extensor muscles, which (in addition to the common extensor) allow these fingers to fully extend even if no other fingers are extended. The middle and ring fingers, however, do not have independent extensors. Instead, these two fingers are tethered to the fingers on either side of them by tendons. Furthermore, the intersection of the tendons for the ring finger is more distal than that of the tendons of the middle finger, making the ring finger the most dependent of all of the fingers. These anatomical observations (e.g. Boyes Braem, 1981, 1990; Ann, 1993, 2006; Greftegreff, 1993) as well as experimental evidence (Häger-Ross and Schieber, 2000) suggest that the order of independence is Thumb >> Index >> Pinky >> Middle >> Ring. The specific implications of this ordering are explained in conjunction with the each constraint below. The first constraint—Radial (provided in (2))—addresses the high degree of independence of the index finger. This constraint is grounded in the fact that the

6There is only one context-sensitive constraint used here, which I will explain in more detail in Section 5.3.2. 7I use ‘ﬁngers’ when only talking about the index, middle, ring, and pinky, and ‘digit’ to mean ﬁngers or thumb. 95

index finger is the most independent finger by virtue of anatomy—specifically, by virtue of its having its own independent extensor muscle and its own branch of the common flexor muscle. In addition to increasing its independence, the positioning of these muscles also increases the control of the index finger, which is especially important in components where the maximization of joint contrasts is a priority.

(2) Radial Reference Constraint (Radial): SF must include the index ﬁnger.

The basis of *ExtMid in (3) is the observation that the middle finger, because it has no extensor muscle of its own, is dependent on either on the extension of the index finger to which it is tethered, or on the extension of the whole hand via the common extensor (which would also result in an extended index finger).

(3) Middle Finger Extension Constraint (*ExtMid): Do not allow an extended middle ﬁnger as an SF unless the index ﬁnger is also extended.

The ring finger is highly dependent upon the middle and pinky finger both because they are tethered in extension via ligaments (part of the juncturae tendinum), and because they share a muscle head of a flexor muscle (the flexor digitorum profundus). The *Ring constraint provided in (4) is based in these dependencies.

(4) Selected Ring Finger Constraint (*Ring): Do not allow a selected ring finger unless it has the same configuration as both the middle and pinky fingers. 96

All of the above dependencies have led others to propose principles or constraints concerning the adjacency of selected ﬁngers (but not the thumb) in handshapes (Man- del, 1981; Ann and Peng, 2000). I follow their lead here in the inclusion of the constraint Adjac in (5).

(5) Adjacency Constraint (Adjac): SF must be adjacent to each other.

Finally, the *SpecJnts constraint in (6) is based on the assumption that the joints would prefer not to be specified for a particular joint configuration since (by definition) specification requires deliberate effort. The end result of adherence to this constraint is typically joints in an ‘at rest’ state in handshapes where contrast is dependent on something other than joint flexion (e.g. contact with the thumb).

(6) Joint Speciﬁcation Constraint (*SpecJnts) Do not specify SF for a particular joint conﬁguration.

It is important to emphasize that, like all constraints in OT, these constraints are violable, despite their being based on anatomical facts. We perceive a fairly wide range of joint positions to be extended and closed, so anatomical compensations making ‘impossible’ handshapes possible do occur. (For example, an ‘extended’ middle finger is used in a popular (though impolite) gesture in many Western cultures, but rarely do people notice that the middle finger is not quite fully extended and/or that the other fingers are less than fully closed to compensate for it being difficult to articulate.)

Perceptual Markedness constraints

Perceptual Markedness constraints (i.e. Markedness at the level of segment contrasts) are based on Flemming’s Minimum Distance (Mindist) constraints and 97 take the form shown in (7). These are constraints that concern the perceptual ‘distance’ (n) between the phonemes in an inventory relative to some (in Flemming’s case) auditory or acoustic dimension (D) (e.g. formant frequency, VOT).8

(7) Perceptual Markedness constraints:

Mindist = D: n: The perceptual distance between contrasts across dimension D must be n.

Basically, the theory assumes that languages will reject inventories in which phonemes are too close together because they are too difficult to differentiate perceptually. These constraints are ranked with respect to each other from least distinctive to most (e.g. Mindist = D:1 >> Mindist = D:2 >> Mindist = D:3 etc.).9 To illustrate how this works in traditional Dispersion Theory, I use an example adapted from Flemming (2002). Let us say that D is the formant frequency F1 (i.e. vowel height), and that F1 can be divided into five levels as shown in (8). The Tableau in (9) shows how various inventory candidates would be evaluated by the set of Mindist = F1 constraints. One violation mark (‘*’) is assigned for each instance where a vowel contrast in the inventory is not separated by the minimum distance specified in the constraint. For example, in Candidate A, the [i - a] contrast has a distance of 4 along the F1 continuum, so it does not violate any of the constraints in the set. In Candidate B, while the [i - a] contrast continues to incur no violations, both [i - e] and [e - a] have a distance of 2, therefore violating the minimum distance requirements of Mindist = F1:3 and Mindist = F1:4; consequently, each constraint is violated twice (once per contrast). And so on with Candidate C.

8Padgett (e.g. 1997, 2003) formulated perceptual constraints in a slightly diﬀerent way than Flem- ming, basing them on the fraction of available perceptual space used by each contrasting segment. These constraints may prove useful for sign language analyses in the future, but as of yet we know too little about the relevant perceptual space in these languages to implement them. 9These constraints can then be combined using ‘&’ and/or ‘or’ (e.g. Mindist = (D:1 & D:2) or D:3), although combinations were not necessary in the example analyses presented here. 98

1 2 3 4 5 (8) i I e 2 a

(9) Dispersion Theory Mindist example.

Based on the lack of detailed perceptual research already mentioned, judging the perceptual distance between handshapes is not as straightforward as taking acoustical readings (and even those, Flemming admits, can be somewhat arbitrary measures). Two very basic perceptual constraints (provided in (10) and (11)) are used in these analyses, namely Mindist = SF:n and Mindist = Base:n. Given the complexity of handshape perception, I consider these constraints to be temporary until more perceptual research can be done.

(10) Minimum Distance Constraint: SF (Mindist = SF: n): The perceptual distance between contrasts must be at least n with respect to number of selected ﬁngers not shared in common.

(11) Minimum Distance Constraint: Base Joints (Mindist = Base: n): The perceptual distance between contrasts must be at least n with respect to base joint ﬂexion. 99

The first of these constraints is based on a basic quantification of selected finger difference (e.g. and have one finger different, and have two, etc.), assuming (perhaps somewhat naïvely) that the more selected fingers handshapes have in common, the closer they will be perceptually. The second constraint is based on the observed continuum of degree of base joint flexion in (12).10

1 2 3 4 5

(12)

Using Mindist = Base:n as an example of how these constraints work, according to the continuum in (12), the contrast between an extended conﬁguration and a curved-closed conﬁguration , would have an n of 2. As in the vowel height example, this particular contrast would violate any Mindist = Base:n constraint where n > 2 (i.e. Mindist = Base:3, Mindist = Base:4, etc.) because according to the constraint, the contrasting segments are too close together perceptually.

5.2.2 Faithfulness constraints

Faithfulness in Dispersion Theory is also aﬀected by its focus on contrast. In traditional OT, Faithfulness constraints require aspects of the winning output to remain unchanged from an underlying input. Because Dispersion Theory deals with contrasts within a system rather than words in isolation, Flemming claims that inputs are incompatible with the theory. As a consequence, the only Faithfulness constraints

10From here onward (unless stated otherwise) handshape pictures in tables and tableaux are only exemplifying the handshape feature being discussed (e.g. joint conﬁguration) and do not represent the handshapes themselves. 100 he includes are Maximize Contrast constraints, which only reference the state of contrasts within the grammar as a whole.11 On the other hand, Ito and Mester’s analysis of behavior across lexical components depends upon a reranking of Faithfulness to borrowed inputs. How can the two approaches be reconciled? An answer can be found by looking at Padgett’s (2003) use of Dispersion Theory for historical changes in contrast. Although he acknowledges that it is a departure from Flemming’s original concept, he assumes inputs from earlier stages in a language’s history when modeling sound changes over time. As he puts it:

[I]t is impossible to even describe historical change otherwise. . . it could well be that the analytical assumptions there [in historical models] ought to diﬀer from those relevant to any of synchronic sound patterns such as previous analyses within Dispersion Theory. (80-1)12

I take a similar stance here. My Faithfulness constraints take two forms depending on the speciﬁcity of the analysis being presented. Because lexical components (although they are synchronic divisions) are largely based on diachronic relationships to historically borrowed forms, like Padgett, I need to have some sort of input to the system if I want to explain their behavior. Since I am proposing here that visual aspects can be borrowed from linguistic forms (e.g. ﬁngerspelling or signs from other sign languages) as well as real world referents, and since the latter are not ‘inputs’ in the traditional sense for OT, I refer to all of these borrowed forms interchangeably as either ‘referring entities’ or ‘external referents’. Furthermore, in the Tableaux used to present my analyses, I place these ‘referring entities’ in a side-bar beside the candidate set rather than above them in the traditional ‘input’ slot. That said, referring entities will not always appear in these analyses. This is because the analysis of contrasts within a language is a bit like looking through a camera

11Flemming does not explicitly refer to Maximize Contrast constraints as Faithfulness constraints, but because they deal with the maintenance of contrasts in the system, I label them as such here. 12See Padgett (2003) for further justification. 101 with a zoom lens. When zoomed in close to an analysis (e.g. the analysis of representations of leggedness, Section 5.3.1, or of the diachronic analyses of handshape change in doctor vs. nurse, Section 5.4.1) you can see details like the referring entity. In these cases, Faithfulness constraints take the form IdentX, where X is either a feature category if the referring entity is linguistic (foreign component) or a relation to a visual attribute if it is a real world object (classifiers). It should be noted that when these referring entities are used, they are still in the form of contrasts; they are simply contrasts from sources external to the phonological structure. However, when you zoom out to the level of the inventory, this Faithfulness to individual borrowings becomes necessarily more abstract. Consequently, in my larger scale (synchronic) analyses of inventories (e.g. SF combinations, Section 5.3.1) I do not include referring entities. This is because at the inventory level, in theory, peripheral lexical components should prefer to have as many contrasts as possible in order to faithfully represent the infinite set of possibilities borrowed from real world attributes or foreign signs/systems. Consequently, I assume that in these larger analyses, Faithfulness constraints take the form MaximizeX. In other words, in this approach, MaximizeX constraints (at least those referencing borrowed inputs/referents) are only different from IdentX constraints in terms of scale—MaximizeX for larger scale analyses and IdentX for those on a smaller scale.

Featural Faithfulness constraints

The next group of constraints are based on Flemming’s Maximize Contrasts constraints for the perceptual properties of spoken language phonemes; their basic form is shown in (13). 102

(13) Featural Faithfulness constraints:

Max f : Maximize the number of contrasts with respect to feature f.

Ident f : Corresponding segments in external referents and the output forms should have identical values with respect to feature f.

As a group, these constraints represent the third goal of Dispersion Theory, i.e. languages strive to maximize the number of contrasts in their inventories to allow them more ways to diﬀerentiate between words. All Max f constraints (regardless of their underlying construction) are positive scalar constraints, which means that the optimal candidate has the most observances rather than the least violations. √ Candidates receive check marks (‘ ’) for each instance where the conditions of the constraint are met (i.e. for every contrasting segment present) instead of asterisks (‘*’) for violations. Consequently, for these constraints candidates are eliminated when they have too few marks rather than when they have too many. Ident f constraints, on the other hand, can be violated, and those violations are marked the usual way (i.e. ‘*’). I return to the vowel height example to illustrate how this works in Tableau (14). √ Each candidate receives one ‘ ’ per contrasting segment in their vowel inventory, and the winner (Candidate C in this case) is the candidate with the most marks.

(14) Dispersion Theory Maximize Contrast example. 103

The constraints of this type used here fall into two categories: those that do not make reference to any kind of external referent, and those that do (at least in small scale analyses). Given what was said in Section 5.2.2 about the diﬀerent scale of Faithfulness constraints for peripheral components, Max f constraints have the potential to be a little bit confusing. Smaller scale analyses use both Max f constraints (faithful to system internal contrast) and Ident f constraints (faithful to external input). However, in larger scale analyses, since both core and foreign components are subject to contrasts involving the same distinctive features, the ‘zoomed out’ versions of both the Ident f and Max f constraints take the same form: Max f (Figure 5.2).

Figure 5.2.: Faithfulness constraints in large and small scale analyses.

Only two sets of featural Faithfulness constraints are used in this analyses: those referring to SF features and those referring to joint features (shown in (15) – (18)).13

13These constraints address SF and joint features in general, but I expect similar constraints addressing more speciﬁc feature sets (e.g. quantity, point of reference) to be formulated as needed in future analyses. 104

(15) Maximize SF Contrasts (MaxSF): Maximize the number of contrasts with respect to SF combinations.

(16) SF Identity (IdentSF): Corresponding segments between the external referent and the output should have identical SF combinations.

(17) Maximize Joint Contrasts (MaxJnts): Maximize the number of contrasts with respect to joint conﬁgurations.

(18) Joint Identity (IdentJnts): Corresponding segments between the external referent and the output should have identical joint conﬁgurations.

Iconic Faithfulness constraints

The iconic Faithfulness constraints in (19) behave like the feature Faithfulness constraints just discussed, with the exception that instead of being faithful to a linguistic referent, they are faithful to visual aspects of the real world objects they represent. Like Max f and Ident f, Max R constraints are positive scalar constraints (marked √ with ‘ ’s), and Ident R constraints are not (i.e. they are marked with ‘*’s).

(19) Iconic Faithfulness constraints:

Max R: Maximize the number of possible contrasts with respect to relationship R.

Ident R: Contrasts between external referents and contrasts between output segments should have comparable values with respect to relationship R.

As mentioned in Section 5.1.4, determining what Faithfulness means between an object and a handshape can be difficult. Keeping in mind the Loan Hypothesis 105 and the idea that iconic forms borrow from visual aspects of their referents, I have formulated the iconic Faithfulness constraints based on shared physical characteristics in conjunction with (wherever possible) perceptual principles. These characteristics make use of the various kinds of prominent feature contrasts discussed in Chapter 3 and their iconic relationships to objects (summarized in Figure 5.3). Because the nature of HCL external referents is different than those of other classifier types (i.e. objects vs. hands), I first discuss constraints relevant to WCLs and DCLs, and then those relevant to HCLs. Not all of the constraints proposed in this section will be used in the analyses in Sections 5.3 and 5.4, but I expect them all to be relevant for future analyses of this type.

Figure 5.3.: Available iconic relationships by classiﬁer type. 106

The first group of constraints I will discuss are those dealing with size. As you may recall from Section 3.5.2, size is represented by different handshape features depending on the classifier type. One potential difficulty in the formulation of a Faithfulness constraint referencing a quality like size is that it is typically thought of in relative terms (‘small’, ‘big’, ‘bigger’, etc.). This turns out not to be a problem in an analysis where both output candidates and external referents are expressed as contrasts. For example, given a contrast between two real world objects along a given dimension, if Object A’s size is greater than Object B’s size, then the same should be true of their representations, i.e. Classifier A’s size should be greater than Classifier B’s for that dimension. The dimensions for size used in these constraints are those discussed in Section 3.5.2, and are repeated here:

1. whole object size: overall object size vs. selected ﬁnger size (WCLs)

2. surface size (i.e. thickness): the distance between outside edges of a continuous surface of an object vs. that of the (unspread) hand (WCLs and DCLs)

3. reference point size: the distance between two given reference points on the perimeter of an object vs. the distance between the thumb and ﬁngertip(s) of the hand (DCLs)

The size constraints proposed here are provided in (20) – (25).

(20) Maximize Whole Size Contrasts (MaxSize:W): Maximize the number of contrasts with respect to the the size of the whole object.

(21) Whole Size Identity (IdentSize:W): External referent contrasts and output contrasts should have comparable overall sizes. 107

(22) Maximize Surface Size Contrasts (MaxSize:S): Maximize the number of contrasts with respect to the distance between outside edges of a continuous surface.

(23) Surface Size Identity (IdentSize:S): External referent contrasts and output contrasts should have comparable amounts of space between outside edges of a continuous surface.

(24) Maximize RP Size Contrasts (MaxSize:RP): Maximize the number of contrasts with respect to the distance between reference points.

(25) RP Size Identity (IdentSize:RP): External referent contrasts and output contrasts should have comparable amounts of space between points of reference.

A similar group of constraints, those dealing with shape, are given in (26) – (31). These shape constraints (relevant for both WCLs and DCLs to varying degrees) reference the same physical dimensions as the size constraints (whole, surface and RP) discussed in Chapter 3. However, instead of the comparisons being relative, here they state a preference for round and ﬂat shapes.

(26) Maximize Whole Shape Contrasts (MaxShape:W): Maximize the number of contrasts where the whole hand is round or ﬂat.

(27) Whole Shape Identity (IdentShape:W): External referent contrasts and output contrasts should have comparable overall shapes (curved, circle or ﬂat). 108

(28) Maximize Surface Shape Contrasts (MaxShape:S): Maximize the number of contrasts with round or ﬂat surfaces.

(29) Surface Shape Identity (IdentShape:S): External referent contrasts and output contrasts should have comparable surface shapes (curved or ﬂat).

(30) Maximize RP Shape Contrasts (MaxShape:RP): Maximize the number of contrasts where the points of reference form a circle or a line.

(31) RP Shape Identity (IdentShape:RP): External referent contrasts and output contrasts should have comparable shapes with respect to points of reference (circle or a line).

Basing a preference for round and flat shapes on observational tendencies would be circular reasoning (no pun intended); the hands have the potential to represent many other shapes but do not, and the explanation should be apparent in the analysis. Therefore, we must ground the preference in perceptual principles as much as possible. According to Gestalt principles (see, for example, Pomerantz and Kubovy, 1986), convex surfaces and symmetrical forms are both favored, making circles dou- bly blessed, and we have seen (Section 3.5.2) that the representations of round and circular shapes in perimeter classifiers create the roundest negative space possible (either part of a circle or a whole circle) by means of the SF and thumb. Handshape representations of flat objects also correspond with perceptual favoritism; these configurations (assuming an opposed thumb) either include perpendicular angles internal to the hand or create parallel edges, both of which are featured prominently in theories of object recognition because of their perceptual stability (Rosielle and Cooper, 2001). The physical properties of these shapes are illustrated in Figure 5.4. 109

Figure 5.4.: Shape distinctions in terms of symmetry and angles

Finally, there are two additional constraints that are only relevant for WCLs, concerning subparts of the whole. These two sets of constraints deal with a correspondence between the number of subparts in the object vs. the hand(s), and the arrangement of those parts with respect to each other. The ﬁrst set, (32) and (33), is fairly self-explanatory, and examples of the latter set, (34) and (35), can be seen in spread vs. stacked representations for leg positions (see Section 5.3.2) or in using crossed ﬁngers (e.g. ) to represent the segments internal to rope or crossed legs.

(32) Maximize Part Number Contrasts (MaxPrt:Nbr): Maximize the number of contrasts with respect to perceived parts of the whole.

(33) Part Number Identity (IdentPrt:Nbr): External referent contrasts and output contrasts should have the same number of perceived parts. 110

(34) Maximize Part Arrangement Contrasts (MaxPrt:Arr): Maximize the number of contrasts with respect to the arrangement of parts.

(35) Part Arrangement Identity (IdentPrt:Arr): External referent contrasts and output contrasts should have the same arrangement of perceived parts.

In terms of iconic relationships, HCLs present an interesting situation since they are, in actuality, handshapes representing hands. At first glance, it would seem logical to simply make these real-world handshapes the inputs to the system. Ultimately, that could be a viable solution, but because grips vary so much in everyday use, it would be better (at least for now) to treat them as outside referents much like the objects discussed above. This would also be useful from the perspective of economy; since (as we saw in Chapter 3) HCLs and DCLs use the same mechanisms for representing these characteristics (albeit in a different order), they should be able to share the same constraints. Before this can happen, however, it must be shown that the iconic relationships in grip handshapes still apply. The representation of shape in HCLs based on grip type was presented in Section 3.5.2 and can be used to justify use of the shape constraints. Similarly, utilization of the size constraints can also be justified by examining grip literature, more specifically, by examining constraints on finger position (i.e. RP size) and finger width (surface size) set forth by MacKenzie and Iberall (1994):

Finger Position constraint. The distance between the grasping surface patches of the two VFs [virtual ﬁngers] must be w (the object width).” Finger Width constraint. The object length h can represent an upper limit on the width of a VF (e.g., coﬀee mug handle vs. coin held in palm). (41)

Basically, these constraints tell us that the distance between the ﬁngers does correspond to the size of the object, and that the number of (adjacent) selected 111

fingers can only be as wide as the object is long. These observations allow us to use the size constraints listed in (22) – (25) for HCLs.14 I propose two final sets of Faithfulness constraints for HCLs (shown in (36) – (39)) based on the differences between precision and power for grip types discussed in Section 3.5.2. Max and IdentContr:SF are based on finger use in grip types. The most control would be offered by a grip that uses all of the fingers. Of handshapes without all fingers involved, those making use of the index finger would have the most control, and those without the index would have the least. Sign languages make use of this type of iconicity as well, as is evidenced by the tendency to use middle finger grips ( ) to represent the handling of delicate objects that do not require large amounts of control or grip-strength (e.g. tissue paper). Max and IdentContr:Jnts refer to differentiation between precision and power grasps via the presence or absence of a stacked configuration and an unopposed (or only slightly opposed) thumb.

(36) Maximize SF Control Contrasts (MaxContr:SF): Maximize the number of SF contrasts with respect to degrees of control.

(37) SF Control Identity (IdentContr:SF): External referent contrasts and output contrasts should have comparable degrees of control with respect to SF.

14In addition to their constraints on finger position and width in grip types, MacKenzie and Iberall (1994) also include two constraints regarding the magnitude and orientations of force in grasping hand configurations. Since no actual force is applied by the fingers in an HCL, I do not address those constraints here. However, an examination of force distribution is useful in explaining why the finger groups of a grasping hand (each group referred to as a ‘virtual finger’ in the literature) include only adjacent fingers. Taking, for example, a precision grip on a wooden dowel, the further away the opposing fingers are from the axis of rotation (in this case, the point of contact of the opposing thumb), the smaller the imbalance necessary to move the object (i.e. the less stable the grip is). This suggests that there could also be a constraint requiring adjacency in HCLs from the perspective of their iconic relationship to actual grasping hands, as opposed to (or more likely, in conjunction with) a physiological constraint based on the tethering of the fingers via the ligaments. I leave this to future research. 112

(38) Maximize Joint Control Contrasts (MaxContr:Jnts): Differentiate between the finger and thumb configurations of precision and power grasps.

(39) Joint Control Identity (IdentContr:Jnts): External referent contrasts and output contrasts should have comparable degrees of control with respect to ﬁnger and thumb positions.

5.3 Synchronic analyses

Now that the mechanisms of the theory have been explained, I will use them to explain some of the cross-componential and cross-linguistic observations presented in Chapters 3 and 4.

5.3.1 Selected Fingers

Three-legged boy

The first analysis I will present is of the three-legged boy data. You may recall from Section 3.3 that HKSL signers all used to represent the boy, suggesting that the grammar is unified in its constraint ranking with regard to this contrast. ASL and DSGS signers’ responses, however, varied between using , a variant with the selected digits of , and the same two-part WCL (i.e. ) usually used to represent people, suggesting that other factors are at play beyond the scope of this analysis. Here I present a Tableau representing the selection of in all three languages (most specifically HKSL), in addition to alternative rankings for ASL and DSGS which may prove useful as this variation is explored in future work.15

15Because of the inherent complexities of language, this kind of variation is problematic in all linguistic analysis—spoken and signed—and is not particular to this work. Continued research into this area is always warranted. 113

The preferences for these SF combinations can be expressed using various rankings of four constraints—two articulatory Markedness constraints ( *ExtMid and *Ring), one perceptual Markedness constraint (Mindist = SF:1), and an iconic Faithfulness constraint (IdentPrt:Nbr)—as shown in Tableaux (40) – (42). The candidate set for these tableaux consists of the three observed handshape choices, in addition to one other potential choice, , and all representations are shown in contrast to the two-part version commonly used in all three languages. In this analysis (and all those to follow), an arrow replaces the traditional pointing hand indicating the winner to differentiate it from the handshape font already in use. First, to indicate that no language seems to allow the – handshape contrast for legged-beings, *ExtMid (which eliminates because there is an extended middle finger but no extended index) is ranked highest in all three tableaux, eliminating Candidate D. From that point, it becomes a matter of 1) whether or not iconic Faithfulness is highly ranked, or 2) whether or not perceptual distance is ranked over articulatory ease. In Tableau (40) we see the optimal ranking for HKSL—one which is also possible in ASL and DSGS. This ranking places an importance on Faithfulness to number of parts as compared to the referring entity contrast, but values perceptual salience over articulatory ease. Candidate A (where both kinds of people referents are represented by only one SF combination) is eliminated for violating IdentPrt:Nbr because it has no handshape with three selected digits to match the number of legs of the ‘three-legged person’ member of the referent contrast. Candidate C is then eliminated because it has less than the required distance along the ‘different-SF’ dimension (i.e. because 0 < 1).16 The winning contrast in this case, therefore, is vs. .

16The Mindist = SF:1 only applies to fingers, so the difference in thumb does not affect it. 114

(40) Ranking based on two- vs. three-legged-being classiﬁers in all three languages (especially HKSL).

In Tableau (41) we see an alternative ranking for ASL and DSGS which also values Faithfulness to the numbers of parts in the referent contrast, this time prioritizing ease of articulation.17 Candidate A is again eliminated for violating IdentPrt:Nbr, but this time Candidate B is eliminated next because it has a selected ring ﬁnger without the support of a selected pinky, leaving the contrast vs. the winner.

(41) First alternative ranking in ASL and DSGS based on two- vs. three-legged-being classiﬁers.

Finally, in Tableau (42), we see the alternative ranking for ASL and DSGS that selects as optimal a single representation ( ) for both members of the referring

17This variant may also indicate a conﬂict between iconic Faithfulness and Faithfulness to a number handshape as an instance of numeral incorporation. 115 contrast. In this case, the important ranking distinction is that IdentPrt:Nbr is ranked lowest, i.e. that articulatory ease and perceptual salience (which is not violated since there are no contrasting representations) are given a higher priority than iconic Faithfulness.

(42) Second alternative ranking in ASL and DSGS based on two- vs. three-legged-being classiﬁers.

Cross-componential distribution

I now approach the question of SF contrasts on a grander scale—i.e. the inventory level—for these languages cross-componentially. In each of these analyses (specifically, Tableaux (43) – (47) and (52) – (57)), I examine the constraints and rankings necessary to predict the correct inventory for a given language and lexical component. As candidates, I use the set of possibilities observed (i.e. the winners from all languages across all components) in addition to an inventory with other cross-linguistically attested SF combinations (A), and an inventory with only three contrasts whose members are optimized for maximum perceptual salience (Candi- date J). (These additional candidates are included to help illustrate the predictive power of the analysis.)18 I do not include thumbs in these combinations because their status as selected fingers is often uncertain. 18It is almost certain that not all of the SF combinations in these inventories are directly contrastive when compared in the context of the whole handshape (e.g. joint configuration, thumb position, 116

I begin by looking at the core rankings for each language (Tableaux (43) – (45)) beginning with ASL. Based on the non-initialized handshapes in DASL, core lexical signs in ASL seem to include the six SF combinations listed in Candidate D.19 The choice of Candidate D can be expressed using two constraints, the articulatory Markedness constraint *Ring and the system internal Faithfulness constraint MaxSF. Of these two, the Markedness constraint is higher ranked, as evidenced by the elimination of all inventories that include SF combinations with ring fingers unsupported by both the middle and pinky fingers (i.e. , , and ). Hence, Candidates A, B, C, and E are eliminated, and the optimal candidate is determined to be the one remaining with the largest number of SF contrasts. etc.) By treating them all as equal contrasts, I am actually making the problem more difficult, but until we have more information about what is or is not allowed in certain contexts, I will take the more conservative track.

19The current analysis does not include handshapes only used in numbers (e.g. )). Researchers often note that number signs behave diﬀerently than other core signs, with regard to handshape features, phonological context, and handshape change (e.g. Sandler, 1996). Consequently, numbers may constitute their own native subcomponent of the lexicon, perhaps comparable to the mimetic component of spoken Japanese, (i.e. a native subcomponent whose constraints are ranked diﬀerently based on outside factors.) The phonology of number signs is certainly an area that requires additional research, and as such I do not address it here further. 117

(43) Ranking in the ASL core component based on SF contrast inventory.20

According to the entries in Boyes Braem (in progress), DSGS’s core lexicon (not including foreign borrowings) includes only the ﬁve SF combinations in Candidate F in Tableau (44). To account for this choice, an additional articulatory Markedness constraint is required, namely Adjac. Here I follow the idea that Markedness constraints (at least those involved in universal implicational hierarchies) have a set ranking cross-linguistically (Ito and Mester, 2001). Assuming this is the case (which I will do until evidence appears to the contrary), Adjac must be ranked below *Ring since otherwise it would block the correct result for ASL. It must also be ranked above √ 20To save space, for this and all subsequent Tableaux, a number following a ‘*’ or ‘ ’ indicates the number of marks in that cell (e.g. *9 = nine violations). 118

MaxSF to avoid selecting the same candidate as ASL did in Tableau (43). The resulting ranking eliminates Candidates A-E via Markedness violations, and the optimal inventory is the one remaining with the largest number of contrasts.

(44) Ranking in the DSGS core component based SF contrast inventory.

Based on the entries in the HKSL dictionary (Tang, in press), HKSL’s core utilizes a much larger inventory of SF than the other two languages, as seen in Candidate B of Tableau (45). It is obvious from the large number of SF contrasts in HKSL’s inventory that MaxSF is very highly ranked in this language. The inventory does not, however, include all cross-linguistically attested combinations, and therefore at 119 least one Markedness constraint (and probably more than one) must be ranked higher. Because the physiological and perceptual information on handshape contrast is too limited to fully understand all of the potential factors involved, for this analysis, I will use the constraint Markedness Constraint:X (Marked:X) as a place-holder until more research becomes available. I propose ranking Marked:X above MaxSF, assuming it would eliminate the theoretically possible (but not attested) inventory including the SF combinations , and (i.e. Candidate A).21 The other Markedness constraints—*Ring and Adjac—are then ranked below MaxSF so that Candidate B, the remaining candidate with the most contrasts, is judged to be optimal.

21The SF combination seems to be in a state of transition (e.g. the handshape is used in the HKSL sign/SCL plane by older signers but not younger; see Section 5.4.1) so it may or may not be included in HKSL’s core inventory. Details of this transition need to be explored further to better incorporate this particular combination into the analysis. 120

(45) Ranking in the HKSL core component based on SF contrast inventory.

Here it becomes necessary to revisit the constraint ranking for the three-legged boy analysis in Tableaux (40) – (42). In those rankings, *ExtMid was consistently ranked above *Ring in order to correctly eliminate in all cases. Although it does not contribute directly to the analysis at hand, I include *ExtMid in the constraint ranking here to maintain continuity in the ranking of Markedness constraints overall. To summarize the core rankings of these languages, DSGS ranked articulatory Markedness above system internal Faithfulness, HKSL ranked Faithfulness above (most) articulatory Markedness, and ASL’s ranking fell between the two. Also, the rankings of all three languages so far have yielded an overall Markedness ranking of Marked:X >> *ExtMid >> *Ring >> Adjac. 121

Next, I address the distribution of SF combinations in foreign components. By way of introduction, however, I must point out that the inventories assumed for these components are based on only a small set of forms. The fingerspelling systems of ASL and DSGS are comprised of a relatively finite set of handshapes by virtue of the small number of letters in the written alphabets of the surrounding languages; ASL represents each of the 26 letters of of the Roman alphabet (not always with a unique handshape), and DSGS supplements its representation of those 26 with five more (‘ä’, ‘ö’, ‘ü’, ‘ß’ and ‘ch’). Consequently, while it is possible to observe which handshape features are used within this set, there is no way to tell how many additional features might be available to this type of foreign borrowing if there were more letters to represent.22 For this reason, I use ‘foreign’ here in a broader sense than I have in previous sections; here I include both fingerspelled borrowings and borrowings from other sign languages to get a better sense of what each language is willing to incorporate into its grammar. Furthermore, character signs in HKSL constitute an even smaller set of examples from which to draw conclusions. The study of the handshapes in these forms is further complicated by the fact that most of these signs require a combination of both hands (see Figure 5.5 for examples23). Because such a small amount is known about character signs in HKSL, and because there is such a small number of examples in my data, I do not include these handshapes in my formal analysis. However, I will offer the observation that, with regard to iconicity, these forms seem to be similar to WCLs—since they more directly represent stroke combinations of written characters

22Interestingly, there are a few instances, at least in ASL, where letter combinations are fused together at the handshape level. For example, , the ‘I love you’ handshape, is commonly considered to be a fusion of the ﬁngerspelled letters (‘I’), (‘L’) and (‘Y’). Other fusions of this type include a further alteration of by adding (‘R’) to make (used to symbolize the English phrase ‘I really love you’) and the representation of a double ‘Z’ using a . While these compound-like handshapes are in varying degrees of standardization and nativization, they may give us a glimpse into what is possible in handshapes originating from this kind of foreign borrowing. 23My thanks to Wong Yuet On for providing this pair of examples. 122

(and are not a historical set of handshapes which may or may not look like the written forms they symbolize like ﬁngerspelling alphabets), they may be subject to similar shape and part iconicity constraints.

Figure 5.5.: Examples of character signs.

The constraint rankings for the foreign components of ASL and DSGS are oﬀered in Tableau (46).24 Here we see that the selection of the correct candidate inventory for foreign forms (Candidate C) implies a ranking of MaxSF above *Ring and Adjac, since these constraints are violated by handshapes like and , and would otherwise eliminate the correct candidate. However, I rank MaxSF below *ExtMid based on my current understanding of foreign handshapes in these languages, as SF combinations such as , , , and are not found. Should any of these combinations become attested through foreign borrowings, MaxSF would be moved higher in the rankings. (Note, because Marked:X has not yet proven necessary in

24SF combination does not violate *ExtMid in ASL and DSGS foreign forms because the forms using this combination either do not have an extended middle ﬁnger ( ) or their index ﬁnger is

mostly extended at the base joint oﬀering adequate support ( ). 123

ASL or DSGS analyses, I do not include it here, but its ranking above *ExtMid is implied.)

(46) Ranking in the ASL and DSGS foreign component based on SF contrast inventory.

Next, we move on to the distribution of SF in classifiers. As with foreign forms, the set of SCLs is quite small and difficult to draw conclusions from. Since these handshapes utilize distinctive contrasts instead of prominent ones based on iconic relationships (as discussed in Section 3.5.1), and because I have not found SCLs in any of these languages that include SF combinations not found in their cores, I am 124 working under the assumption that the rankings of SCLs are the same as the core rankings of their respective languages. WCLs, however, are influenced by iconic Faithfulness constraints, as seen in Tableaux (40) – (42). Based on the classifiers discussed previously, all three languages appear to have the same SF contrast inventory for WCLs—namely, Candidate E—as seen in Tableau (47).25 Here we introduce the iconic Faithfulness constraint Max- Prt:Nbr, which evaluates the possible quantities of parts a given inventory can represent via selected fingers. Because I do not include thumbs or two handed signs here, the maximum number of satisfying marks for this constraint here is four (one for each finger-to-part representation). The exact ranking of MaxPrt:Nbr with respect to Marked:X and *ExtMid is uncertain, since all combinations yield the correct elimination of Candidates A, B and C, leaving Candidate E as the inventory with the most check marks. MaxPrt:Nbr must be ranked above *Ring and Adjac, however, since those constraints would incorrectly eliminate this candidate.

25Even though the SF combination has not been found in HKSL WCLs so far, there is reason to believe that, at least historically, this combination may have been used in the iconic representation of parts (e.g. bed which may have been derived from a WCL using the extended ﬁngers to represent bedposts). 125

(47) Ranking in all three languages for WCLs based on SF contrast inventory.

Recall from Section 3.5.2 that SF combinations in WCLs are used to represent size in addition to number of parts. Therefore, even though it is not required for the selection of the optimum candidate, we should determine the ranking of those constraints as well. I leave the ranking of MaxSize:W (whole object size) to future work, since examples are quite rare. The ranking of MaxSize:S (surface size, or thickness), however, is justiﬁed in Tableaux (48) – (51). Given that 3FHSs like were used to represent the three-legged boy and other three-part objects by at least some signers from all three languages (see Section 3.3), and given that I have found no surface size related uses of 3FHSs for WCLs in any of these languages, we can assume that Candidate B (three possible surfaces size rep- 126 resentations and four possible part representations)26 is optimal for these languages (especially HKSL), as is indicated via the ranking in Tableau (48). Here, Faithful- ness to parts (MaxPrt:Nbr) is ranked above the Markedness of an unsupported ring ﬁnger (*Ring), but the Markedness constraint is then ranked above a desire to faithfully represent the largest number of possible thicknesses (MaxSize:S). In some situations, however, signers from ASL and DSGS rejected 3FHSs, even for representing number of parts, indicating that the ranking in Tableau (49) is also possible in these two languages.

(48) WCL ranking based on part number and surface size contrasts.

26The SF combination is counted twice in these inventories—once as a surface size, and once as a part number. 127

(49) Alternative WCL ranking based on part number and surface size contrasts.

Meanwhile, Tableaux (50) and (51) show that ranking either *Ring below both iconicity constraints, or ranking Faithfulness to surface size higher than the other two, yield incorrect candidate predictions. We can therefore rank MaxSize:S below *Ring in the overall ranking for WCLs, as seen in Tableau (52). (Its ranking with respect to Adjac is unknown.)

(50) Incorrect WCL ranking based on part number and surface size contrasts. 128

(51) Incorrect WCL ranking based on part number and surface size contrasts. 129

(52) Ranking in all three languages for WCLs based on SF contrast inventory (revised).

We continue with the ranking for DCLs. In Tableau (53) we see the constraint ranking for ASL, which allows only three SF combinations for these representations (Candidate H). For this classiﬁer type, only MaxSize:S is relevant for discussions of SF combination (since the other size and shape constraints concern joint features), so it is the only iconic constraint in the analysis. In order to ensure the correct candidate selection, a new Markedness constraint, Radial, must be added (indicating that the index ﬁnger must be selected), and it must be ranked below Adjac, as a higher ranking would interfere with previous analyses. MaxSize:S must be ranked below 130

*Ring to prevent the selection of the incorrect Candidate E, but its ranking with respect to the other two Markedness constraints is unclear.

(53) Ranking in ASL for DCLs based on SF contrast inventory.

DSGS and HKSL only seem to allow two SF combinations in DCLs, as seen in Candidate I of Tableau (54). Here, the MaxSize:S must still be ranked somewhere below *Ring, but the Radial constraint is no longer suﬃcient to predict the correct candidate given the rest of the Markedness constraints. To remedy this, I introduce a perceptual Markedness constraint Mindist = SF:2, since the winning candidate consists of only one contrast that is perceptually quite salient (cf. Lane et al., 1976). 131

Recall that this constraint is violated each time a contrast within the inventory has less than two ﬁngers diﬀerent (e.g. SF = vs. ). It must be ranked lower than Adjac because it would otherwise disrupt the analysis of DSGS’s core, but above MaxSize:S to prevent the optimization of Candidate H.

(54) Ranking in DSGS and HKSL for DCLs based on SF contrast inventory.

The ﬁnal rankings presented in this section pertain to HCLs. Tableau (55) illustrates the ranking for HCLs in ASL, for which the four contrasts in Candidate G are 132

optimal.27 The relevant iconicity constraints for SF combinations in HCLs are Max- Contr:SF (relating to the relative amounts of control available to ﬁnger combinations) and MaxSize:S. The exact ranking of these constraints compared to Marked- ness is unclear—MaxContr:SF must be ranked somewhere above Radial and MaxSize:S must be ranked somewhere below *Ring— but since MaxContr:SF can be ranked above *Ring and MaxSize:S cannot, I assume the two are ranked MaxContr:SF >> MaxSize:S with respect to each other. In this ranking, the winner of the ﬁnal two candidates is determined by Radial and the candidate with the fewest violations (Candidate G) wins.

27Since extended ﬁngers rarely occur in HCLs, *ExtMid would probably not apply here, so I do not include it. Even if it were included, the violations would not interfere with the selection of the correct candidate. 133

(55) Ranking in ASL for HCLs based on SF contrast inventory.

In HKSL HCLs, Tableau (56), the rankings and optimal candidates are the same as those of ASL with one diﬀerence. Because Mindist = SF:2 is a Markedness constraint, I believe that its ranking with respect to other Markedness constraints should remain consistent within a language. However, since it is not articulatory, I do not consider it to be part of the universal ranking that applies to all languages. I therefore include it in the DSGS and HKSL analyses, but not for ASL. With Mindist = SF:2 ranked above Radial as it was in Tableau (54), it becomes the determining constraint—again, the candidate with the fewest violations wins. 134

(56) Ranking in HKSL for HCLs based on SF contrast inventory.

As it did for DCLs, DSGS seems to have only two contrastive SF combinations for HCLs (Candidate I), as seen in Tableau (57). This ranking is the same as the one for DCLs (Tableau (54)) except for the addition of MaxContr:SF, which is inconsequential in candidate evaluation since it is ranked lower than Mindist = SF:2 (i.e. the constraint that determines the optimal candidate). 135

(57) Ranking in DSGS for HCLs based on SF contrast inventory.

Discussion of SF analyses

In summary, these analyses of SF distribution across lexical components and languages represent the diﬀerent ways each one resolves the conﬂict between competing forces in the grammar (i.e. articulatory and perceptual Markedness and featural and iconic Faithfulness). Some generalizations and observations can be made about the analyses presented here. First, they all support a universal ranking of the articulatory Markedness constraints introduced, namely: Marked:X >> *ExtMid >> 136

*Ring >> Adjac >> Radial. Second, generally speaking, HKSL tended to prefer Faithfulness over Markedness, DSGS Markedness over Faithfulness, and ASL was somewhere in the middle. It is also interesting to note that in ASL and DSGS, the relationships between SF contrast Faithfulness in the core rankings (i.e. MaxSF) and external Faithfulness from the more peripheral components (e.g. IdentSize:S) parallel the continuum of SF vs. joint inventories of the various classiﬁer types from Section 3.5.3. For example, for DCLs and HCLs in these two languages (i.e. the components with relatively large numbers of joint contrasts; see Chapter 3), iconic Faithfulness constraints for SF were ranked lower with respect to Markedness constraints than MaxSF was in the core components, indicating that that Markedness has priority over iconic Faithfulness to SF in these classiﬁer types. Conversely, SCLs (which have a larger inventory of SF contrasts) have equal Faithfulness rankings with the core (if my assumption that they share core rankings is correct), and Faithfulness in foreign and WCL contrasts is of higher or equal ranking as compared to MaxSF in core forms (i.e. external Faithfulness to SF contrasts has greater or equal priority in these forms as compared to the core). These observations further support the idea that there is a balancing of complexity across the lexicon with regard to the number of joint and SF contrasts. In contrast, HKSL rankings always had core Faithfulness higher than external Faithfulness constraints with respect to Markedness. I hypothesize that this higher ranking of MaxSF in the HKSL core with regard to SF contrasts is the end result of either a historical willingness to borrow and maintain contrasts from sources outside the grammar (linguistic or iconic), or a consequence of being a relatively young language (cf. Frishberg, 1975); either way, to put it in the terms of the current theory, the HKSL grammar currently values Faithfulness to borrowed forms (and the consequent wealth of contrasts it creates) over articulatory ease and perceptual distinctiveness.28 Based on this hypothesis, I would predict that HKSL handshape contrasts would be

28This could be balanced out with higher ranked Markedness constraints in other parts of the grammar, e.g. movement type. 137 less prone to visible nativization and lexicalization processes as compared to ASL and DSGS, where Faithfulness to contrast is ranked lower.

5.3.2 Joints

Joint ﬂexion

Next we turn to an analysis of the O and C group joint configurations examined in Chapter 4 (Tableaux (58) and (59)). In these Tableaux, I examine four potential inventory candidates across the six handshapes: Candidate A is the set of all possible contrasts, Candidate B contains all possible C group handshapes but only the endpoints of the O group, Candidate C contains only the endpoint configurations, and Candidate D consists of a three-way contrast between fully extended joints (the flat endpoint of the C group), curved joints (including the round and mid C group handshapes), and handshapes with contact (including all O group handshapes).29 There are only two relevant constraints for this discussion of joint contrasts—MaxShape:S, which maximizes the number of round and flat surface shapes (i.e. those primarily represented by nonbase joint flexion), and Mindist = Base:2 which receives a violation mark for every contrast closer than 2 on the continuum of base joint flexion provided in Section 5.2.1. Tableau (58) shows the ranking needed to correctly predict the contrasts found in the classifier forms (DCLs and HCLs) from the experiment. We see here that the iconic Faithfulness constraint for shape is ranked higher than the Markedness constraint judging perceptual distance. Candidates C and D are eliminated first because they have too few favored shape contrasts (i.e. symmetrical curves and straight lines at preferred angles). Of the two that remain, Candidate A is then eliminated because it has the most violations of Mindist = Base:2. 29I assume an opposed thumb is present in all inventory members. 138

(58) Ranking in ASL for perimeter classiﬁers based on Experiment joint contrasts.

In Tableau (59) we see the reverse ranking applies for core and initialized contrasts as seen in the experiment. Here, the grammar prioritizes perceptual Markedness over iconic Faithfulness. All candidates violate Mindist = Base:2, but Candidate D in- curs fewer violations than the others (i.e. just one between extended and curved).30,31

30In determining whether or not Mindist = Base:2 is satisﬁed, I assume the midpoint of each set of non-contrastive joint conﬁgurations, i.e. somewhere between C group HS 1 and HS 2 for {curved} and for {contact}.

31The diﬀerences in contrast between , and are similar to an example cited by Flemming concerning central vowels. In systems with front-back vowel contrasts, central vowels are relatively uncommon, but in situations where front-back contrasts are lacking, central vowels are typically the unmarked case. Similarly, in the DCLs, the round-ﬂat contrast exists without , but in the core, which lacks this prominent contrast, is the target. 139

(59) Ranking in ASL for core component based on Experiment joint contrasts.

Stacked

The final synchronic analysis presented here concerns the stacked configuration across the lexicon. Recall from Section 3.2 that the contrast between spread and stacked fingers in ASL was shown to be distinctive in initialized signs, prominent in WCLs, and active in both plus the core.32 Furthermore, recall that the active contrast was blocked by the prominent contrast in the WCL classifiers (e.g. ‘people on their backs’), but not by the distinctive contrast in the foreign forms (e.g. verb). Tableaux (60) – (62) offer rankings to explain these phenomena using three types of constraints: a context sensitive constraint ( → [+ori] ) that says spread will become stacked in specified orientations (see below), external Faithfulness constraints to either features (IdentJnts) or iconic relationships (IdentPrt:Arr), and an internal faithfulness constraint to featural contrasts (MaxJnts). The context sensitive Markedness constraint is based on anatomy—in handshapes with the index and middle fingers selected, the middle is tethered to the ring and pinky fingers with which it shares a flexor muscle, therefore it is more difficult to extend (Mandel, 1981); this

32Because the active status of stacked in DSGS and HKSL was not clear to me from the available data, I concentrate on ASL here. 140

may be especially true in orientations where the rotation of the wrist tightens that ﬂexor.33 The candidates in these Tableaux consist of three possible contrasts in the appropriate contexts for ( → [+ori]): vs. , vs. , and vs. . External referring entities are given to the side of the candidate set. In Tableau (60) we see the ranking for the core where [stacked] is active. In the orientation context, becomes in the core, which is indicated by a ranking of the context sensitive constraint above MaxJnts. Candidates A and B are eliminated because elements in their contrast sets (i.e. the s) did not become stacked, leaving Candidate C optimal.

(60) Ranking in ASL for core component based on [stacked] observations.

In Tableau (61) we see the ranking for initialized forms in the foreign component, which have as their referring entities the ﬁngerspelled handshapes and (a distinctive contrast). As in the core, becomes in the speciﬁed orientations in initialized forms, despite their distinctive contrast in other contexts. This is indicated by a ranking of the orientation constraint above IdentJnts. As before, Candidates A and B are eliminated because the s did not become stacked, leaving Candidate C optimal.

33The change from spread to stacked in hands facing the midline may also have perceptual motivation, i.e. it helps maintain a visual distinction between the ﬁngers at an angle where it would be harder to perceive by the interlocutor. 141

(61) Ranking in ASL for foreign component based on [stacked] observations.

Finally, in Tableau (62), we see the ranking that determines the winner for WCLs, in which the [spread] vs. [stacked] contrast is prominent. Here, the active contrast is blocked, as shown by the ranking of IdentPrt:Arr over the context sensitive constraint. This time, Candidates B and C are eliminated because they do not faithfully represent the diﬀerences in part arrangement found in the referring entities (i.e. the leg positions), leaving the Candidate A the winner.

(62) Ranking in ASL for WCLs based on [stacked] observations. 142

5.4 Diachronic Analyses

As previously mentioned, over time, signs have a tendency to become less iconic and more arbitrary (Frishberg, 1975). This can be explained by a re-ranking of constraints—Faithfulness to the original referring entity (i.e. borrowed input) becomes increasingly lower ranked over time, giving language particular Markedness constraints more priority in the determination of the ﬁnal form.

5.4.1 Selected ﬁngers

The ﬁrst historical example illustrates the ranking changes from an example mentioned in Section 3.3, namely, the gradual change of handshape from to in the sign doctor. This sign was originally an initialized sign in Old French Sign Language from which it was borrowed (médecin), but over time it has lost its connection to the original form. This lowering of rank for external Faithfulness to the ﬁngerspelled letter as contrasted with (shown here in the sign pair, doctor vs. nurse) is illustrated in Tableaux (63) (the original initialized ranking) and (64) (the newer core ranking). We see in Tableau (63) that the Faithfulness constraint IdentSF is ranked higher than the Markedness constraints *Ring and Mindist = SF:2, thus eliminating the vs. contrast in Candidate B.34 In Tableau (64), however, external Faithfulness has been lowered below the Markedness constraints, now making the vs. optimal.35

34I include both articulatory and perceptual Markedness constraints here because it is unclear which has a greater effect on this change. 35Interestingly, the newer four-finger handshape has further differentiated itself from M perceptually via an extended thumb, but because so little is known about thumb positions in general, I do not include it in my analysis here. 143

(63) Ranking for ASL doctor (vs. nurse) as an initialized sign.

(64) Re-ranking for ASL doctor (vs. nurse) as a core sign.

Similar alterations are found in the ASL sign thousand, borrowed from the initialized Old French Sign Language sign mille (cf. Long, 1918) and in the sign scout in HKSL (Tang, in press) which was almost certainly borrowed from the three-fingered cultural gesture used by the scouts themselves.36 Historical information suggests that there may also be other iconic relationships grounded in perceptual principles that influence SF contrasts. For instance, HKSL offers an example that might indicate the existence of a symmetry constraint based on the Gestalt principle that we tend to organize visual information using symmetry (Pomerantz and Kubovy, 1986). The older form of the airplane classifier is —an extremely marked handshape given the previously established anatomical constraints,

36In scout, the handshape seems to be changing to instead of , possibly because there is

already a sign in the same location using . 144

but one which is symmetrical around the middle ﬁnger much like the visual form of the airplane it represents is symmetrical around the fuselage. In the younger generation, however, this classiﬁer is being replaced by which does not violate *ExtMid, indicating that, if an iconic Faithfulness constraint to symmetry exists, it is becoming lower ranked in HKSL over time.

5.4.2 Joints

Joint configurations can also change over time as a result of lowered Faithfulness rankings. For example, the sign for berry was likely once a combination of shape- representing classifier handshapes—a DCL representing the round berry and a WCL representing the thin, straight stem.37 Based on personal observation, it seems that the production of this sign now varies across the to handshape range tested in the experiment in Chapter 4, indicating that IdentShape:S is no longer ranked high enough to maintain the vs. contrast from the DCL stratum of the classifier component. This change in contrast is illustrated by Tableaux (65) and (66), the first representing the joint contrast in productive DCLs, and the second representing the contrast (or lack thereof) in lexicalized DCLs.

(65) Ranking for shape contrasts in ASL perimeter classiﬁers.

37Until production experiments are undertaken to see how lexicalized DCLs and WCLs representing shape are actually articulated with respect to joint ﬂexion, such analyses are mostly speculation based on observation. 145

(66) Re-ranking for lexicalized shape contrasts in ASL (e.g. berry).

Another example lies in the lexicalization of HCLs into signs like control and key. In Long’s (1918) dictionary, the sign control uses what he refers to as hands “in grasping attitude”, pictured as lateral grasp handshapes ( similar to ). The movement in this entry is more complex than in the modern version of the sign, but based on other similarly likely handshape changes (e.g. give, key, (ex)change) and the fact that variants with both handshapes can be found in more modern dictionaries (e.g. in Sternberg, 1998, and in Valli, 2005), the likelihood is high that the two forms are related. If this is, indeed, the predecessor for the modern sign, the grasping hands have become handshapes over time. We currently lack enough information to do a detailed analysis of this kind of change (i.e. to predict all of the relevant perceptual dimensions and constraints involved), but generally speaking, it can be seen as a case of re-ranking perceptual Markedness above iconic Faithfulness. I would predict the grasping hand being referenced in the HCL (often attributed to ‘holding reins’) would be a power grip, with all fingers selected—fully flexed and stacked—with an unopposed thumb (for both power and control). I would also predict such a handshape to contrast with which has the same specification for joint flexion, but is unstacked with a closed thumb (see Section 3.5.2 for justification). Under the current approach, as this handshape is lexicalized, its Faithfulness to IdentContr:SF (which chooses all fingers selected) and IdentContr:Jnts (which chooses the stacked configuration and unopposed thumb) becomes lower ranked, and 146

the grammar instead selects a contrast that is more perceptually salient—namely vs. .38

5.4.3 Iconicity in core forms

Experimental research suggests that iconicity in (frozen) signs does not have a significant effect on memory or neural processing (Poizner et al., 1981; Emmorey et al., 2004), supporting the idea that iconicity is no longer relevant in lexicalized forms. However, Boyes Braem (1981) and van der Kooij (2002) both conclude that the phonetic behavior of handshapes in frozen forms (at least in ASL and NGT, respectively) can not be explained without including some mention of (and even dependence on) semantic motivation in the analysis. From the perspective of the core-periphery model, many of the forms they discussed could be considered boundary cases between the core and classifier components of their languages’ respective lexicons, either because they are in intermediate stages of the lexicalization process, or because they are in a perpetual state of morphological ‘limbo’. In the former case, at least in terms of perceptual contrast, I would expect these forms to behave like the initialized O and C group signs did in the experiment, namely that the range of variants acceptable would be more like that of core forms than classifiers, but with a remaining residual preference for the variant closest to the original classifier handshape.39 Furthermore, in a footnote of her dissertation, van der Kooij (2002) mentions that in a preliminary and unsystematic comparison of video data to isolated signs, she noticed that “semantically unmotivated form elements seem to yield more phonetic variation. When form elements are motivated, the phonetic implementation space seems to vary less.” (141) To my knowledge, more rigorous investigations of this observation have not been undertaken, but if true, it might support a greater strength in the residual phonetic preference when the feature

38The change from the stacked, flexed grasp hand to makes sense because the position of the index finger is similar in both cases. 39These connections to iconic aspects of classifiers are often called ‘motivated’ elements of a sign. 147 originated as a prominent contrast with morphological significance. Diachronically speaking, I would expect this preference (however strong) to fade in most forms over time as the signs are further lexicalized (Frishberg, 1975). Synchronically, however, a case could also be made for assigning some of these forms (particularly the ones that seem to jump in and out of the core as needed) to an additional stratum of the lexicon, as suggested by Perlmutter (2006). This is supported by Brentari and Padden’s (2001) observation that core forms whose handshapes retain their classifier status can not be replaced by fingerspelling handshapes to create initialized signs. I would predict that handshapes of boundary forms such as these would have different behavior when they are in their frozen/core state. For example, the handshapes of the core versions should no longer be alterable in terms of size, shape, etc.40 Before we can make more definite arguments for or against an additional substrata, more research is needed on the perceptual contrasts in these boundary forms.

5.5 Summary

In this chapter I have presented an analytic approach toward explaining the cross- linguistic and cross-componential variation in handshape contrasts using various in- stantiations of OT. First, I argue that forms in the core component strike a language particular balance between universally ranked articulatory Markedness constraints, Markedness constraints maximizing perceptual salience, and Faithfulness constraints maximizing the number of contrasts in the system. In other words, I believe that the behavior of the core component can be described using the basic tenets of Dispersion Theory, following Flemming (2002). In non-core components, however, I believe that there is an additional group of Faithfulness constraints at play—ones that depend on Faithfulness to forms outside of

40Many of these boundary handshapes already belong to the SCL category where iconic alterations are no longer possible, (e.g. the airplane classifier handshape in the sign fly-by-plane). 148 the language in question, either in the form of foreign linguistic borrowings or visual aspects of real world objects. In either case, following Ito and Mester’s work, I believe that these Faithfulness constraints come into conflict with native constraint rankings to varying degrees depending on the lexical component. In certain classifiers substrata (WCL, DCL, and HCL), these constraints are ranked higher than they are in SCLs or the core, allowing the more peripheral lexical components to have a greater number of phonological contrasts. Diachronically and synchronically, the closer forms are to the core, the lower these Faithfulness constraints are ranked, leaving the phonological contrasts to be shaped by both articulatory and perceptual Markedness constraints and system-internal Faithfulness to contrasts. Additionally, in forms that do eventually become nativized (from the foreign components) or lexicalized (from the classifier system), Faithfulness to these outside forms becomes increasingly lower ranked over time. The Faithfulness I propose with respect to iconic relationships in the classifier component is not arbitrary, nor is it conceptually or semantically based. As much as possible (given the current amount of information available) I base my iconic Faithfulness constraints on visually perceptible aspects of both the referent and the articulators of the hand. That said, the decision to include iconicity in phonological representations has always been a precarious one. In discussing the inclusion of iconicity in linguistic description, Wilbur (1979) states:

Any linguistic model claiming to provide a consistent description of language should be an appropriate framework in which to attempt an analysis. This framework should allow one to develop tests, arguments, and predictions. In other words, it provides the metalinguistic mechanisms with which to work. . . All of this adds up to a single point: describing something as iconic may be true, but it is also linguistically insuﬃcient. (156) 149

It is my hope that the approach presented here has created a framework that includes iconicity, but still remains suﬃcient for the analysis of handshape at the phonological level. 150

6. CONCLUSION

The main goal of this dissertation was to explore the nature of contrast in sign language handshapes. My first hypothesis in this work was that the distribution of handshape contrasts is not homogeneous, either within or across sign languages. The examples presented in Chapter 3 showed that this was indeed the case—contrasts in selected finger combination (e.g. three-fingered forms vs. two or four) and in joint configuration (e.g. [stacked] vs. [spread]) were shown to differ based on contrast type (distinctive, active, or prominent) when compared both language-internally (i.e. across lexical components) and cross-linguistically. Chapter 3 also showed that the distributions of selected finger and joint contrasts were asymmetrical even across a given lexical component—in this case, across classifier types in the spatial component of ASL—and that these asymmetries were largely due to the extent to which each type was influenced by iconic relationships. Chapter 4 also showed differences in contrasts across components, this time by testing signers’ perceived meanings and preferences for various handshapes in forms from across the ASL lexicon (core, foreign and spatial components). These results also indicated that differences in contrast did exist cross-componentially. The second hypothesis was that the differences in contrast distributions could be explained as differences in the ranking of constraints using Optimality Theory (OT). In Chapter 5, I proposed Markedness constraints for handshape in sign languages based on articulatory ease and perceptual distinctiveness, as well as Faithfulness constraints based on a desire to maximize language-internal contrasts and a desire to maintain contrasts borrowed into the language from external sources (both linguistic and iconic). I then illustrated how contrast differences across lexical components, 151 across languages, and even across time (i.e historical change) could be accounted for by re-ranking these constraints. OT analyses like the one presented here have potential explanatory power well beyond the parameter of handshape. For example, the use of movement in classifier constructions has often been considered a problem for phonological theories, with certain movements being used in core forms and other, more ‘mimetic’ or ‘imitative’ movements being used in certain classifier constructions (e.g. Schick, 1987). Others, however, have claimed that movement follows linguistically determined patterns (e.g. Supalla, 1982). An OT analysis pitting some sort of iconic faithfulness constraint for movement shape against the markedness restrictions of the system may help explain these apparent inconsistencies. These results also have implications for the phonological representation of handshape using feature trees. They suggest that there may be different feature geometries for handshape in the three lexical components, each feature geometry reflecting the differences revealed here. Constructing these feature trees in detail is a fruitful topic for future work. Additionally, this kind of analysis could be used to explain a new finding by Aronoff et al. (2008) that languages tend to favor either HCLs or Instrument CLs (here, part of the WCL group) for lexicalization. I would hypothesize that this type of preference is the result of different constraint rankings for the languages in question, giving priority to certain types of iconicity in relation to articulatory and perceptual factors. Finally, this analysis might have implications for those rare occasions where iconicity does play a part in spoken language structure. For example, Ito and Mester, in their presentation in support of stratal faithfulness (i.e. a re-ranking of Faithfulness constraints with regard to the Markedness rankings of the phonology as a whole) they mention that accounting for the behavior of the spoken Japanese lexical substrata based in sound symbolism and onomatopoeia (called ‘Mimetics’) may require 152 additional and more specific intervening constraints than other strata (1995a:190).1 Perhaps faithfulness to aspects of the original sound structures could be used to help explain these differences. As a field, we have only scratched the surface when it comes to understanding production and perception in sign languages, not to mention reaching agreement on the inventories of contrasts available to these languages. This dissertation work has tried to contribute in this area, but there is still much work to be done.

1See Mester and Ito (1989) for more discussion of the phonological diﬀerences of mimetics outside of an OT framework. LIST OF REFERENCES 153

LIST OF REFERENCES

Allen, J. S., and Miller, J. L. (2001) Contextual influences on the internal structure of phonetic categories: A distinction between lexical status and speaking rate. Perception & Psychophysics 63 (5): 798–810. Ann, J., (1993) A linguistic investigation of the relationship between physiology and handshape. University of Arizona dissertation. Ann, J. (2006) Frequency of occurrence and ease of articulation of sign language handshapes: The Taiwanese example. Washington, D.C.: Gallaudet University Press. Ann, J., and Peng, L. (2000) Optimality and opposed handshapes in Taiwan Sign Language. University of Rochester Working Papers in the Language Sciences 1 (2): 173–194. Aronoff, M., Meir, I., Padden, C., and Sandler, W. (2003) Classifier constructions and morphology in two sign languages. In Perspectives on classifier constructions in sign languages, pp. 53–84. Mahwah, NJ: Lawrence Erlbaum Associates. Aronoff, M., Meir, I., Padden, C., and Sandler, W., (2008) Instrument vs. handling: Sign language lexicalization patterns. Poster presented at the Linguistic Society of America Annual Meeting, Chicago, IL. Aronoff, M., Meir, I., and Sandler, W. (2005) The paradox of sign language morphology. Language 81 (2): 301–344. Avery, P., and Idsardi, W. J. (2001) Laryngeal dimensions, completion and enhancement. In Distinctive Feature Theory, ed. by T. A. Hall, pp. 41–70. Berlin: Mouton de Gruyter. Battison, R. (1978) Lexical Borrowing in American Sign Language. Silver Spring, MD: Linstok Press. Benedicto, E., and Brentari, D. (2004) Where did all the arguments go?: Argument-changing properties of classifiers in ASL. Natural Language and Linguistic Theory 22: 743–810. Boyes Braem, P., (1981) Features of the handshape in American Sign Language. Berkeley: University of California dissertation. Boyes Braem, P. (1990) Acquisition of the handshape in American Sign Language: A preliminary analysis. In From gesture to language in hearing and deaf children, ed. by V. Volterra and C. Erting, pp. 107–127. New York: Springer. 154

Boyes Braem, P., (in progress) A multimedia bilingual databank basis for the lexicon of swiss german sign language (deutschschweizerische gebärdensprache, dsgs). . Brentari, D. (1998) A Prosodic Model of Sign Language Phonology. Cambridge, MA: MIT Press. Brentari, D. (to appear) Sign language phonology. In Handbook of phonological theory, ed. by J. A. Goldsmith, J. Riggle, and A. Yu. New York/London: Blackwell. Brentari, D., and Eccarius, P. (in press) Phonological contrast in sign language handshapes. In Sign Languages, ed. by D. Brentari. Cambridge/New York: Cambridge University Press. Brentari, D., and Padden, C. (2001) Native and foreign vocabulary in Ameri- can Sign Language: A lexicon with multiple origins. In Foreign vocabulary in sign languages, ed. by D. Brentari, pp. 87–120. Lawrence Erlbaum Associates. Channon, R. E., (2002) Signs are single segments: Phonological representations and temporal sequencing in ASL and other sign languages. College Park: University of Maryland dissertation. Chomsky, N., and Halle, M. (1968) The sound pattern of English. New York: Harper & Row. Clements, G. N. (2001) Representational economy in constraint-based phonology. In Distinctive Feature Theory, ed. by T. A. Hall, pp. 71–146. Berlin: Mouton de Gruyter. Cogill-Koez, D. (2000) Signed language classifier predicates: Linguistic structures or schematic visual representation? Sign Language & Linguistics 3 (2): 153–207. Corazza, S. (1990) The morphology of classifier handshapes in Itailian Sign Language (lis). In Sign language research: Theoretical issues, ed. by C. Lucas, pp. 71–82. Washington, D.C.: Gallaudet University Press. Corina, D. P. (1993) To branch or not to branch: Underspecification in ASL handshape contours. In Phonetics and Phonology: Current issues in ASL phonology, ed. by G. R. Coulter, volume 3, pp. 63–95. San Diego: Academic Press. Cutkosky, M. R. (1989) On grasp choice, grasp models, and the design of hands for manufacturing tasks. IEEE Transactions on Robotics and Automation 5 (3): 269–279. DeMatteo, A. (1977) Visual imagery and visual analogues in American Sign Language. In On the other hand: New perspectives on American Sign Language, ed. by L. A. Friedman, pp. 109–136. New York: Academic Press. Dresher, B. E. (2003) Contrast and asymmetries in inventories. In Asymmetry in Grammar, Volume 2: Morphology, Phonology, Acquisition, ed. by A.-M. Di Sciullo, pp. 239–257. Amsterdam: John Benjamins. Dresher, B. E., Piggott, G., and Rice, K. (1994) Contrast in phonology: Overview. In Toronto Working Papers in Linguistics, ed. by C. Dyck, volume 13.1, pp. iii–xvii. Department of Linguistics, University of Toronto. 155

Eccarius, P., (2002) Finding common ground: A comparison of handshape across multiple sign languages. Master’s thesis, Purdue University, West Lafayette, IN. Emmorey, K., Grabowski, T., McCullough, S., Damasio, H., Ponto, L., Hichwa, R., and Bellugi, U. (2004) Motor-iconicity of sign language does not alter the neural systems underlying tool and action naming. Brain and Language 89: 27–37. Emmorey, K., and Herzig, M. (2003) Categorical versus gradient properties of classifier constructions in ASL. In Perspectives on classifier constructions in sign languages, ed. by K. Emmorey, pp. 221–246. Mahwah, NJ: Lawrence Erlbaum Associates. Engberg-Pedersen, E. (1993) Space in Danish Sign Language. Hamburg: Signum-Verlag. Flemming, E., (1995) Auditory representations in phonology. Doctoral Disserta- tion, University of California, Los Angeles[published 2002 by Routlege, New York]. Flemming, E. (2002) Auditory representations in phonology. New York: Routledge. Friedman, L. A. (1977) Formational properties of American Sign Language. In On the other hand: New perspectives on American Sign Language, ed. by L. A. Friedman, pp. 13–56. New York: Academic Press. Frishberg, N. (1975) Arbitrariness and iconicity: Historical change in American Sign Language. Language 51 (3): 696–719. Goldsmith, J. A., (1976) Autosegmental phonology. Doctoral dissertation, MIT, Cambridge, MA. [Published 1979, Garland Press, New York]. Goldsmith, J. A. ed. (1995) The Handbook of phonological theory. Cambridge, MA: Blackwell. Greftegreff, I. (1993) A few notes on anatomy and distinctive features in NTS handshapes. In Working Papers in Linguistics, volume 17, pp. 48–68. Department of Linguistics, University of Trondheim. Guion, S. G., Flege, J. E., Akahane-Yamada, R., and Pruitt, J. C. (2000) An investigation of current models of second language speech perception: The case of Japanese adults’ perception of English consonants. Journal of the Acoustical Society of America 107 (5): 2711–2724. Häger-Ross, C., and Schieber, M. H. (2000) Quantifying the independence of human finger movements: Comparisons of digits, hands, and movement frequencies. Journal of Neuroscience 20 (22): 8542–8550. Hall, T. A. ed. (2001) Distinctive Feature Theory. Berlin: Mouton de Gruyter. Hara, D., (2003) A complexity-based approach to the syllable formation in sign language. Chicago, IL: University of Chicago dissertation. Ito, J., and Mester, A. (1995a) The Core-Periphery structure of the lexicon and constraints on reranking. In Papers in Optimality Theory, ed. by J. N. Beck- man, L. Walsh Dickey, and S. Urbanczyk, number 18 in University of Massachusetts Occasional Papers in Linguistics [UMOP], pp. 181–209. Amherst, MA: GLSA. 156

Ito, J., and Mester, A. (1995b) Japanese phonology: Constraint domains and structure preservation. In The Handbook of Phonological Theory, ed. by J. A. Gold- smith, pp. 817–838. Oxford: Blackwell. Ito, J., and Mester, A. (2001) Covert generalization in Optimality Theory: the role of stratal faithfulness constraints. Studies in Phonetics, Phonology and Morphology 7: 273–299. Ito, J., and Mester, A. (2003a) Japanese Morphophonemics: Markedness and word structure. Cambridge, MA: MIT Press. Ito, J., and Mester, A. (2003b) Systemic markedness and faithfulness. In CLS 39: Main session. (Papers from the regional meeting of the Chicago Linguistic Society; 39), pp. 665–689. Chicago: Chicago Linguistics Society. Johnson, R. E., (1990) Distinctive features for handshapes in American Sign Language. Presented at the International Conference on Theoretical Issues in Sign Language Research, Boston, MA. Johnston, T., and Schembri, A. (2007) Australian Sign Language: An introduction to sign language linguistics. Cambridge, UK: Cambridge University Press. Kager, R. (1999) Optimality Theory. Cambridge, UK: Cambridge University Press. Kamakura, N., Matsuo, M., Ishii, H., Mitsuboshi, F., and Miura, Y. (1980) Patterns of static prehension in normal hands. The American Jornal of Occupational Therapy 34 (7): 437–445. Klima, E., and Bellugi, U. (1979) The Signs of language. Cambridge, MA: Harvard University Press. Lane, H., Boyes Braem, P., and Bellugi, U. (1976) Preliminaries to distinctive feature analysis of handshapes in american sign language. Cognitive Psychology 8: 263–289. Lindblom, B. (1986) Phonetic universals in vowel systems. In Experimental Phonology, ed. by J. J. Ohala and J. J. Jaeger, pp. 13–44. Orlando: Academic Press. Long, J. S. (1918) The Sign language: A Manual of signs. Iowa City, IA: Athens Press. MacKenzie, C. L., and Iberall, T. (1994) The grasping hand. Amsterdam: North-Holland. Mandel, M. (1977) Iconic devices in American Sign Language. In On the other hand: New perspectives on American Sign Language, ed. by L. A. Friedman, pp. 57– 107. New York: Academic Press. Mandel, M. (1979) Natural constraints in sign language phonology: Data from anatomy. Sign Language Studies 24: 215–229. Mandel, M., (1981) Phonotactics and morphophonology in American Sign Lan- guage. Berkeley: University of California dissertation. 157

McCarthy, J. J. (2002) A Thematic guide to Optimality Theory. Cambridge, UK: Cambridge University Press. McDonald, B. H., (1982) Aspects of the American Sign Language predicate system. Buffalo, NY: University of Buffalo dissertation. McQueen, J. (1996) Phonetic categorisation. Language and Cognitive Processes 11 (6): 655–664. Mester, A., and Ito, J. (1989) Feature predictability and underspecification: Palatal prosody in Japanese mimetics. Language 65 (2): 258–293. Napier, J. R. (1956) The prehensile movements of the human hand. The Journal of Bone and Joint Surgery 38B (4): 902–913. Okrent, A. (2002) A modality-free notion of gesture and how it can help us with the morpheme vs. gesture question in sign language linguistics (or at least give us some criteria to work with). In Modality and structure in signed and spoken languages, ed. by R. P. Meier, K. Cormier, and D. Quinto-Pozos, pp. 175–198. Cambridge, UK: Cambridge University Press. Padden, C. (1988) Interaction of morphology and syntax in American Sign Lan- guage. New York: Garland Press. Padden, C., and Gunsauls, D. C. (2003) How the alphabet came to be used in a sign language. Sign Language Studies 4 (1): 10–33. Padden, C., and Le Master, B. (1985) An alphabet on hand: The acquisition of fingerspelling in deaf children. Sign Language Studies 47: 161–172. Padgett, J. (1997) Perceptual distance of contrast: Vowel height and nasality. In Phonology at Santa Cruz 5 , ed. by R. Walker, M. Katayama, and D. Karvonen, pp. 63–78. Santa Cruz: Linguistics Research Center, University of California. Padgett, J. (2003) Contrast and post-velar fronting in Russian. Natural Language and Linguistic Theory 21: 39–87. Perlmutter, D. M. (2006) Some current claims about sign language phonetics, phonology, and experimental results. In Laboratory Phonology 8 , ed. by L. Goldstein, D. H. Whalen, and C. T. Best, pp. 315–338. Berlin: Mouton de Gruyter. Pierrehumbert, J. B. (2001) Exemplar dynamics: Word frequency, lenition and contrast. In Frequency and the emergence of linguisitc structure, ed. by J. L. Bybee and H. P. J., pp. 137–158. Amsterdam: John Benjamins. Poizner, H., Bellugi, U., and Tweney, R. D. (1981) Processing of formational, semantic, and iconic information in American Sign Language. Journal of Experimental Psychology: Human Perception and Performance 7 (5): 1146–1159. Pomerantz, J. R., and Kubovy, M. (1986) Theoretical approaches to perceptual organization: Simplicity and likelihood principles. In Handbook of perception and human performance: Volume II. Cognitive processes and performance, ed. by K. R. K. R. SIMPLICITY VERSUS LIKELIHOOD 581 Boff, L. Kaufman, and J. P. Thomas, pp. 36:1–45. New York: Wiley. 158

Prince, A., and Smolensky, P., (1993) Optimality theory. Technical Report #2 of the Rutgers Center for Cognitive Science, Rutgers, NJ. (ROA Version, 8/2002). Rosielle, L. J., and Cooper, E. E. (2001) Categorical perception of relative orientation in visual object recognition. Memory & Cognition 29 (1): 68–82. Sandler, W. (1989) Phonological representation of the sign. Dordrecht: Foris. Sandler, W. (1996) Representing handshapes. In International review of sign linguistics, ed. by W. H. Edmondson and R. B. Wilbur, volume 1, pp. 115–158. Mahwah, NJ: Lawrence Erlbaum Associates. Schembri, A. (2003) Rethinking ‘classifiers’ in signed languages. In Perspectives on classifier constructions in sign languages, ed. by K. Emmorey, pp. 3–34. Mahwah, NJ: Lawrence Erlbaum Associates. Schick, B. S., (1987) The Acquisition of classifier predicates in American Sign Language. West Lafayette, IN: Purdue University dissertation. Slobin, D. I., Hoiting, N., Kuntze, M., Lindert, R., Weinberg, A., Py- ers, J., Anthony, M., Biederman, Y., and Thumann, H. (2003) A cognitive/funtional perspective on the acquisition of "classifiers". In Perspectives on classifier constructions in sign languages, ed. by K. Emmorey, pp. 271–296. Mahwah, NJ: Lawrence Erlbaum Associates. Sternberg, M. L. A. ed. (1998) American Sign Language, (Unabridged Edition). New York: HarperCollins. Stokoe, W. C., Casterline, D. C., and Croneberg, C. G. eds. (1965) A Dictionary of American Sign Language on Linguistic Principles. Silver Spring, MD: Linstok Press. Stungis, J. (1981) Identification and discrimination of handshape in American Sign Language. Perception & Psychophysics 29 (3): 261–276. Supalla, T. (1978) Morphology of verbs of motion and location in American Sign Language. In Proceedings of the Second National Symposium on Sign Language Research and Teaching, ed. by F. Caccamise, pp. 27–45. Silver Spring, MD: National Association of the Deaf. Supalla, T., (1982) Structure and acquisition of verbs of motion and location in American Sign Language. San Diego: University of California dissertation. Supalla, T. (1986) The classifier system in American Sign Language. In Noun classification and categorization, ed. by C. Craig, pp. 181–214. Philadelphia: John Benjamins. Tang, G. ed. (in press) Dictionary of Hong Kong Sign Language. Hong Kong: Chinese University Press. Taub, S. F. (2001) Language from the body: Iconicity and metaphor in American Sign Language. Cambridge, UK: Cambridge University Press. Valli, C. ed. (2005) The Gallaudet dictionary of American Sign Language. Wash- ington, D.C.: Gallaudet University Press. 159 van der Hulst, H. (1995) The composition of handshapes. University of Trond- heim Working Papers in Linguistics 23: 1–17. van der Hulst, H., and van der Kooij, E. (2006) Phonetic implementation and phonetic pre-specification in sign language phonology. In Laboratory Phonology 8 , ed. by L. Goldstein, D. H. Whalen, and C. T. Best, pp. 265–286. Berlin: Mouton de Gruyter. van der Kooij, E., (2002) Phonological categories in Sign Language of the Nether- lands: The role of phonetic implementation and iconicity. Utrecht, The Netherlands: University of Leiden, Published by LOT dissertation. Wallin, L., (1996) Polysynthetic signs in Swedish Sign Language. Stockholm (English translation): Stockholm University dissertation. Wilbur, R. B. (1979) American Sign Language and sign systems. Baltimore: University Park Press. Woll, B. (1987) Signs: Fingerspelling. In Gallaudet encyclopedia of deaf people and deafness, ed. by J. V. Van Cleve, volume 3 (S-Z, Index), pp. 120–124. New York: McGraw-Hill. Woodward, J. (1981) Signs of marking: "stage" three handshapes. In Issues in language: Studies in honor of Robert J. Di Pietro presented to him by his students, ed. by M. Danesi and R. J. Di Pietro, volume 9 of Edward Sapir monograph series in language, culture, and cognition, pp. 47–59. Lake Bluff, IL: Jupiter Press. Zwitserlood, I., (2003) Classifying hand configuration in Nederlandse Gebarentaal. Utrecht, The Netherlands: University of Utrecht, Published by LOT dissertation. APPENDICES 160

A. EXPERIMENTAL STIMULUS ITEMS

Targeted Meaning Non-Targeted Meaning O handshapes

Initialized opinion organization new orleans Core teach fix eat Classiﬁers ‘cardboard tube’ ‘sheet of cardboard’ ‘curved plumbing pipe’ ‘curved sheet of metal’ ‘rope belt’ ‘ﬂat belt’

F handshapes

Initialized foreign french fries get-an-f Core count soon benefit Classiﬁers ‘pole’ ‘thread’ ‘coin on a table’ ‘holding a nail’ ‘hole in a wall’ ‘holding a nail’ 161

Targeted Meaning Non-Targeted Meaning C handshapes

Initialized chicago never church business cafeteria bachelor Core search worry strange force Classiﬁers ‘drink from a bowl’ ‘drink from a box’ ‘look through a ‘look through a round window’ square window’ ‘sewer pipe’ ‘rectangular beam’ 162

B. LIST OF PROPOSED CONSTRAINTS

Articulatory markedness constraints:

Radial Reference Constraint (Radial): SF must include the index ﬁnger.

Middle Finger Extension Constraint (*ExtMid): Do not allow an extended middle ﬁnger as an SF unless the index ﬁnger is also extended.

Selected Ring Finger Constraint (*Ring): Do not allow a selected ring finger unless it has the same configuration as both the middle and pinky fingers.

Adjacency Constraint (Adjac): SF must be adjacent to each other.

Joint Speciﬁcation Constraint (*SpecJnts) Do not specify SF for a particular joint conﬁguration.

Perceptual markedness constraints:

Minimum Distance Constraint: SF (Mindist = SF: n): The perceptual distance between contrasts must be at least n with respect to number of selected ﬁngers they do not share in common.

Minimum Distance Constraint: Base Joints (Mindist = Base: n): The perceptual distance between contrasts must be at least n with respect to base joint ﬂexion. 163

Featural faithfulness constraints:

Maximize SF Contrasts (MaxSF): Maximize the number of contrasts with respect to SF combinations.

SF Identity (IdentSF): Corresponding segments between the external referent and the output should have identical SF combinations.

Maximize Joint Contrasts (MaxJnts): Maximize the number of contrasts with respect to joint conﬁgurations.

Joint Identity (IdentJnts): Corresponding segments between the external referent and the output should have identical joint conﬁgurations.

Iconic faithfulness constraints:

Maximize Whole Size Contrasts (MaxSize:W): Maximize the number of contrasts with respect the the size of the whole object.

Whole Size Identity (IdentSize:W): External referent contrasts and output contrasts should have comparable overall sizes.

Maximize Surface Size Contrasts (MaxSize:S): Maximize the number of contrasts with respect to the distance between outside edges of a continuous surface.

Surface Size Identity (IdentSize:S): External referent contrasts and output contrasts should have comparable amounts of space between outside edges of a continuous surface.

Maximize RP Size Contrasts (MaxSize:RP): Maximize the number of contrasts with respect to the distance between reference points.

RP Size Identity (IdentSize:RP): External referent contrasts and output contrasts should have comparable amounts of space between points of reference.

Maximize Whole Shape Contrasts (MaxShape:W): Maximize the number of contrasts where the whole hand is round or ﬂat. 164

Whole Shape Identity (IdentShape:W): External referent contrasts and output contrasts should have comparable overall shapes (curved, circle or ﬂat).

Maximize Surface Shape Contrasts (MaxShape:S): Maximize the number of contrasts with round or ﬂat surfaces.

Surface Shape Identity (IdentShape:S): External referent contrasts and output contrasts should have comparable surface shapes (curved or ﬂat).

Maximize RP Shape Contrasts (MaxShape:RP): Maximize the number of contrasts where the points of reference form a circle or a line.

RP Shape Identity (IdentShape:RP): External referent contrasts and output contrasts should have comparable shapes with respect to points of reference (circle or a line).

Maximize Part Number Contrasts (MaxPrt:Nbr): Maximize the number of contrasts with respect to perceived parts of the whole.

Part Number Identity (IdentPrt:Nbr): External referent contrasts and output contrasts should have the same number of perceived parts.

Maximize Part Arrangement Contrasts (MaxPrt:Arr): Maximize the number of contrasts with respect to the arrangement of parts.

Part Arrangement Identity (IdentPrt:Arr): External referent contrasts and output contrasts should have the same arrangement of perceived parts.

Maximize SF Control Contrasts (MaxContr:SF): Maximize the number of SF contrasts with respect to degrees of control.

SF Control Identity (IdentContr:SF): External referent contrasts and output contrasts should have comparable degrees of control with respect to SF.

Maximize Joint Control Contrasts (MaxContr:Jnts): Differentiate between the finger and thumb configurations of precision and power grasps. 165

Joint Control Identity (IdentContr:Jnts): External referent contrasts and output contrasts should have comparable degrees of control with respect to ﬁnger and thumb positions.

Other constraints:

→ [+ori]: Spread will become stacked in speciﬁed orientations. VITA 166

VITA

Petra Nichole Eccarius was born in Iowa, grew up primarily in Nebraska, and is now a resident of Milwaukee, Wisconsin. She graduated with a Bachelor of Arts degree from the University of Nebraska in 1999 with a major in Anthropology and an Integrated Studies major in Linguistics; much of her coursework for the latter was in classical languages and American Sign Language, and both majors were supplemented with minors in both History and Religious Studies. She earned a Master of Arts degree from Purdue University’s interdisciplinary Linguistics Program in 2002 with concentrations in Phonology and Sign Language Linguistics. During her time in graduate school she has enjoyed a number of research and teaching opportunities, including teaching undergraduate Introduction to Linguistics at Purdue and graduate level Phonological Theory at Gallaudet University. In her free time, she loves singing in groups and dabbling in a variety of visual arts. Petra strives to approach the study of language from as many perspectives as possible in order to better understand the means by which it changes and grows.