<<

END-WEIGHT MANIFESTATIONS IN GREEK BINOMIALS

By ATHINA KIKIOPOULOU

A Thesis submitted to the School of English Aristotle University of Thessaloniki in partial fulfillment of the requirements for the degree of

MA in LINGUISTICS

Thesis Supervisor: Dr. Nina Topintzi

Thessaloniki 17 February 2020

TABLE OF CONTENTS

LIST OF FIGURES ...... i

LIST OF TABLES ...... ii

LIST OF ABBREVIATIONS ...... iii

ACKNOWLEDGMENTS ...... iv

ABSTRACT ...... v

INTRODUCTION ...... 1

MAIN PARAMETERS OF END-WEIGHT MANIFESTATIONS ...... 6

2.1 Non-Phonological ...... 6

2.2 Phonological Parameters and Partial Evidence from Greek ...... 7

2.2.1 ...... 8

2.2.2 Count ...... 12

2.2.3 Codas ...... 14

2.3 The weight-interface and Prosodic End Weight (Ryan, 2019) ... 18

EMPIRICAL STUDY ...... 21

3.1 Parameters tested and Expectations...... 21

3.1.1 Vowels ...... 22

3.1.2 Syllable Count ...... 22

3.1.3 Codas ...... 22

3.2 General Points in Methodology ...... 23

3.3 First experiment ...... 24

3.3.1 Item Selection ...... 24

3.3.2 Stimuli ...... 25

3.3.3 Results and Discussion ...... 26

3.4 Second experiment ...... 31

3.4.1 Nonce word construction ...... 31

3.4.2 Stimuli ...... 32

3.4.3 Results and Discussion ...... 33 3.5 Summary of Results ...... 36

GENERAL DISCUSSION ...... 39

4.1 Frequency, Markedness, and Accessibility ...... 39

4.2 Phrasal Stress and Weight ...... 43

4.3 Limitations ...... 48

CONCLUSION ...... 52

REFERENCES ...... 54

APPENDICES ...... 63

LIST OF FIGURES (1) Ryan (2016:6) End Weight Manifestations…………………………….….....1 (2) Anttila (2008:53): Satisfying the Stress-to-Stress Principle…………………19 (3) Ryan (2019:328): Satisfying the Stress-to-Stress Principle …………….…..20 (4) Parker‟s (2002:69) sonority scale for English…………………………….....23 (5) First experiment. Item example: sentence #7, targeting [i] versus [u]………26 (6) Legal word-medial clusters in Greek……………………………………..….30 (7) Second experiment. Item example: sentence #13 targeting [e] and [a]……...33 (8) Violations of lapse……………………………………………………………35

i

LIST OF TABLES Table 1: Ablaut patterns in Greek…………………………………………………….…9 Table 2: Relative and F1 and F2 values for the five Greek vowels (Hz).……………...10 Table 3: Relative and Absolute duration for the five Greek vowels (ms)………12 Table 4: Syllable effects: Evidence form Greek………………………………………..13 Table 5: Syllable effects in multinomials: Evidence form Greek…..………………….14 Table 6: Summary of parameters tested………..………………………………………21 Table 7: First experiment: Examples of binomials utilized…………………………….25 Table 8: First experiment: Results per parameter………………………………………27 Table 9: First experiment: Number of times that vowels appear in slot B………..……27 Table 10: First experiment: Wilcoxon‟s results per Final Coda sub-parameter……..…28 Table 11: Friedman‟s test mean values per group………………...……………………29 Table 12: First experiment: Binomial distributions per group…………………...…….29 Table 13: Second experiment: Examples of binomials utilized…………………..……32 Table 14: Second experiment: Results per parameter………………………………….34 Table 15: Second experiment: Number of times that vowels appear in slot B…..…….34 Table 16: Second experiment: Wilcoxon‟s results per Final Coda sub-parameter….....36 Table 17: Summary of Results………...……………………………………………….37

ii

LIST OF ABBREVIATIONS Greek: Standard Modern Greek

GG: binomial consisting of two Greek words

LL: binomial consisting of two loanwords

GL: binomial consisting of a Greek word and a loan word

BNC: Brithish National Corpus

HNC: Hellenic National Corpus

iii

ACKNOWLEDGMENTS I would like to thank my supervisor, dr. Nina Topintzi for being a role model and instilling in me the critical skills required in academic research and linguistic analysis. I am grateful for her constant support and invaluable feedback throughout the whole process. Her guidance and eagerness to assist me with any concern, along with her sheer belief in me has been the greatest motivation in completing this thesis. I am also grateful to dr. Maria Dimitrakopoulou for her valuable comments on the practical aspects of this study, such as stimuli selection and construction, and the collection and statistical analysis of data. I also wish to thank Prof. Katerina Nicolaidis for introducing me to the field of , the practical advice on the construction of the two tasks of this study, and her ever-encouraging words. Finally, I am thankful to my family and friends, who always believe in me.

iv

ABSTRACT Prosodic End-Weight refers to the tendency of prosodically heavier constituents to occupy a position closer to the end of a sentence, so as to coincide with nuclear stress (Ryan, 2019). Coordination in binomial structures is probably its most well-studied aspect. Maybe the most well-known analysis comes from Cooper and Ross (1975), who suggest that, among other factors, phonological properties that can be associated with the second slot in an irreversible binomial of the type A and B, are lowness and backness in vowels, greater word length (in terms of syllable count), as well as complexity of codas (Oakeshott-Taylor, 1984; Benor and Levy, 2006; Mollin 2012). The current study seeks to investigate the phonological aspects that can potentially determine word order in coordination in Standard Modern Greek under the principle of End-Weight. The exemplifies certain tendencies that seem to coincide with manifestations related to Prosodic End-Weight. An observed potential effect of vowel quality is illustrated in patterns of ablaut structures like din-dan and platsa-plutsa, and a syllable-count one in fixed expressions like “ήζε θαη έζηκα” [ˈiζi ce ˈeζima] „customs‟ Moreover, a stress-related tendency has been observed in Greek acronyms by Topintzi and Kainada (2012) and Revithiadou et al. (2015), according to which acronyms, the final of which bear codas (especially illegal ones) are more likely to attract stress than open final syllables. The main aim is to experimentally explore the potential effects of vowel quality, syllable count, and final syllable coda status (null and singleton, in stressed and unstressed environments) in the ordering of constituents in Greek. Two forced choice tasks have been designed to test the potential legitimacy of these presumed tendencies, following a methodology similar to Bolinger (1962), Oakeshott-Taylor (1984), and Parker (2003); in both tasks, the participants read a Greek sentence ending in a binomial and are asked to indicate which order of the two constituents they prefer (A and B or B and A). In the first task, the constituent words are not parts of a frozen form, but random pairings. The second task follows the same methodology, but utilizes nonce words, after Pinker and Birdsong (1979), Parker (2003), and Ryan (2019), aiming at completely eliminating any semantic effects, and grasping better phonological control of word structure, through the creation of minimal pairs. The tendencies emerging from the results indicate that there are indeed effects of vowel backness for the nonce word task, while significant effects derive from both tasks for the word length and the coda parameter. These results in accord with Ryan‟s (2019) predictions for prosodically heavier constituents, suggest that Greek words with back vowels, more syllables, or word-final codas are more preferred in positions of nuclear phrasal stress, a conclusion which is especially interesting for Greek, since it is a weight-insensitive system (Drachman and Malikouti-Drachman, 1999). Finally, the motivation behind this preference is examined considering issues such as frequency (Fenk-Oczlon, 1989) and markedness (Benor and Levy, 2006), but also the relevance of Greek to a weight-sensitive system of gradient nature (Duarte-Garcia 2017). This study concludes that neither theory can successfully describe the findings of this study, and that the status of the stress-weight interface currently remains undetermined for Greek.

Keywords: End-Weight, binomials, word-order, coordination, frequency, markedness, gradient weight

v

CHAPTER 1 INTRODUCTION This study aims at experimentally exploring potential effects of the End-Weight principle in Standard Modern Greek (henceforth Greek) binomial structures. The phenomenon termed as End-Weight refers to a tendency of constituents of certain qualities to occupy a position closer to the end of a prosodic phrase or sentence (Ryan, 2019). One of the first appearances of the term is in Quirk et al. (1972), where it is presented as a principle responsible for the ordering of clauses and their sub-parts within a sentence and the reservation of final positions for “complex” structures, what has also been examined as a constraint of Markedness by Benor and Levy (2006). The phenomenon and its multiple extensions have been observed in various throughout the centuries and received different names, but have mostly been studied in coordination (Ryan 2019). End-Weight was first recorded by Panini for Sanskrit dvandva compounds (Cooper and Ross, 1975; Pinker and Birdsong, 1979; Ryan, 2019) and then again much later by Behaghel (1909). End-Weight has been studied under various scopes and is believed to have a great range of linguistic manifestations; in (1), Ryan (2016:6) provides a variety of environments where End-Weight could possibly manifest itself in English: (1) Ryan (2016:6) End Weight Manifestations Order A Order B a. Particle verbs e.g. picked X up picked up X b. Coordination e.g. X and Y Y and X c. Dative alternation e.g. gave X to Y gave Y X d. Heavy NP shift e.g. revealed X to Y revealed to Y X e. Genitive alternation e.g. X's Y Y of X f. Locative alternation e.g. spray X with Y spray Y on X g. Extraposition e.g. N Rel-X V N V Rel-X h. AP stacking e.g. AP-X, AP-Y N AP-Y, AP-X N i. PP stacking e.g. PP-X PP-Y PP-Y PP-X

When first formulated, the principle of End-Weight was not intended to refer to, as the name may suggest, phonological weight; Quirk et al. (1972) regard it as a matter of constituent complexity, whether these constituents are entire clauses or the internal parts of a clause (p. 776). In order to refer to the phonological aspects of the principle,

1

Ryan (2019) refers to it as Prosodic End-Weight, but for this study, the terms Prosodic End-Weight and End-Weight are used interchangeably. The present study aims to adding to the discussion of the study of two-part structures usually referred to as binomial structures. The term binomial is used by Malkiel (1959) to refer to a sequence of two words connected by some type of coordinator (which can also be null); the two words in question, which he names kernels or members, must belong to the same word class and syntactic level. The two positions available for the connected words within the binomial can occupy position or slot A or B. Benor and Levy (2006:235) illustrate this convention with the example of salt and pepper, for which salt occupies SLOT A and pepper SLOT B. It would be expected that the order that the two members should be irrelevant to the semantic content of the construction, and, indeed, Malkiel (1959) further asserts that the sequence of some “kernels” is flexible and free for the speaker‟s choice; a structure “cold and snow” generally carries the same semantic value as “snow and cold” (Malkiel, 1959:113). There are, however, binomial structures, the order of which is frozen, as in “odds and ends,” for the reverse order “ends and odds” is ungrammatical (p. 113). Malkiel (1959) named such constructions irreversible binomials, but they have also been called fixed- order coordinates by Abraham (1950), binomial expressions by Gustafsson (1975), or freezes by Cooper and Ross (1975). The definitive description of reversibility within these structures has been a matter of discussion by many scholars. While Gustafsson (1976) asserts that reversibility in binomial structures is a rare occurrence, Mollin (2012) widens her scope and data, by classifying the degrees of order flexibility for English into four categories upon analyzing 544 binomials extracted from the British National Corpus (BNC). She reports that only an 18.20% of the items examined are true irreversible binomials in the manner Malkiel (1959) predicts. However, she also notes that a mere 19.49% belongs to the freely reversible category, meaning that the majority of binomials examined belong to a reversible category with a strong (39.52%) or a moderate (27.79%) preference for one of the two orders possible. Considering the percentages of the binomials with a fixed or a more rigid order, it could be argued that the majority of binomials in English (57.72%) have a more or less frozen order, which indicates that there could be an underlying force dictating a specific order of constituents. The circumstances that drive this tendency of binomial phrases to have a more rigid order or to “sound better” when coordinated in a particular order, or in other words the freezing process has been the matter of a discussion initiated by Cooper and Ross 2

(1975). Cooper and Ross (1975) endeavor to specify the parameters which lead to the formation of (what they name) freezes or frozen irreversible binomial structures in English. They devised a comprehensive list of multiple parameters alluding to semantic, pragmatic, and phonological aspects which regulate the ordering of constituents within a binomial phrase. While the semantic and pragmatic motivations for word ordering have proved to be influential indicators, the present study focuses mainly on the phonological and prosodic basis of ordering constraints, as brought forward by Cooper and Ross (1975) and others. The issue of item ordering in coordination has been long discussed, as various parameters appear to have direct effects on word order. Frequency (Fenk-Oczlon, 1989), markedness (Benor and Levy, 2006), as well as cognitive (McDonanld et al. 1993), and even social (Wright et al., 2005) aspects are reported to determine the order of constituents. Cooper and Ross (1975) are among the first to ever consider phonological aspects that may influence word order in binomial structures in English. Their analysis of idiomatic binomial structures yields their MEFIRST constraint, which includes a series of semantic and pragmatic parameters that affect word order. However, it also contains a list of phonological parameters, which suggest that words with certain phonetic and phonological qualities have greater propensity for the occupation of the right edge position. These include, among others, more syllables, back vowels, and fewer or more sonorous coda segments. The phonological aspect of word ordering has also been a challenge. Since phonology is reported to operate on a lower level in such structures (Pordany, 1986; McDonald et al., 1993; Benor and Levy, 2006, Mollin, 2012), it has been difficult to control all other parameters intercepting the operation of phonological factors. While some scholars suggest that phonology does not have any effect on word ordering (Fenk- Oczlon, 1989), and others suggest that it is a cognitive and perceptual matter (Benor and Levy, 2006), an analysis favoring the importance of phonological aspects is provided by Ryan (2019), who suggests that End-Weight effects are observed because phrasal stress must coincide with the heaviest constituent of a prosodic phrase, as a general function of the stress-weight interface. The hypothesis regarding the End-Weight effects that are observed in Greek are based on two distinct foundations. First, according to Cooper and Ross (1975) frozen pairs constitute examples of optimal ordering, meaning that the study of such structures can reveal certain phonological operations within a language. The Greek language includes a wide array of irreversible binomial structures, that seemingly conform to the 3 phonological predictions of MEFIRST mentioned above; vowel quality and word length. Second, recent findings by Topintzi and Kanaida (2012), and Revithiadou et al. (2015), report that word-final codas seem to attract final stress in experiments of acronym stressing. This finding is quite comparable to effects observed pertaining to the stress- weight interface described by Gordon (2006), but also discussed in Ryan (2019). This study delves into the study of phonological parameters potentially affecting word order in novel binomial structures, with the utilization of forced-choice tasks. Similar studies assess possible phonetic, phonological, or prosodic effects on word ordering, via the elicitation of native speakers‟ preferences on word order within such controlled contexts, in languages such as English (Bolinger, 1962; Parker, 2003; Ryan, 2019), but also French (Pinker and Birdsong, 1979) and even German and Africaans (Oakeshott-Taylor, 1984). In order to assess the existence of these parameters in Greek, two tasks were designed, one using real words, and a second using nonce words, and then distributed to two groups of native speakers, consisting of 85 and 86 participants respectively. The tasks targeted the parameters of vowel quality (backness, height and intrinsic length), syllable count, and finally, with regard to the last syllable of the words, singleton versus null coda (also under stress and unstressed conditions). Considering the main parameters listed by Cooper and Ross (1975), linguistic patterns and recent findings on the Greek language, the research questions of the current study revolve around the following parameters having an effect on word ordering in novel Greek binomial structures (A and B): 1. Vowel quality:  Do words with vowels of further back and/or lower quality tend to appear at the right edge?  Is intrinsic a parameter that affects word order? 2. Syllable count:  Do words consisting of more syllables show a greater tendency to occupy the right edge? 3. Final Syllable:  Do words with final codas appear in the second position more often than words with open final syllables?  Does final stress have an effect? Do coda-bearing words appear in the second position more often when they are stressed word finally?  Do illegal codas influence word ordering?

4

 Do more sonorous codas appear at position B more often than less sonorous codas do? The results indicate that there is a statistically significant tendency for words to be localized at the second position of a binomial structure, when they include back vowels, more syllables, as well as a coda word-finally, while stress, legality, and sonority aspects of coda segments do not seem to have a significant effect. The study is organized as follows: chapter 2 includes an overview of the main word ordering parameters which are targeted in this study, along with corresponding evidence from Greek. The methodology of the empirical part of the study, the two experiments, the results and elaboration is laid out in Chapter 3. Chapter 4 includes a discussion of the results and an endeavor to interpret them and determine its relevance to phonological theories of word ordering, as well as the limitations of the study. Finally, the conclusions of the study can be found in Chapter 5, along with some proposals for future research, while an appendix section which includes all materials and results can also be found and the end.

5

CHAPTER 2 MAIN PARAMETERS OF END-WEIGHT MANIFESTATIONS 2.1 Non-Phonological The list formulated by Cooper and Ross (1975) is summarized in a single semantic constraint ME FIRST, which predicts that the first item in a binomial will be more prototypical to the speaker, or closer to the speaker‟s experience. The words that tend to occupy the first slot in their freezes list are more relevant to notions such as: here, now, adult, male, positive, animate, front, agentive, and general. Such effects have indeed been observed in other studies as well. Benor and Levy (2006) report significant effects of prototypicality, suggesting that the first item is always less marked, in a constraint they name MARKEDNESS; the first item is more general, simple, a hypernym, and/or has greater distribution. McDonald et al. (1993) report significant effects for animacy, with the animate member appearing first in sentences, probably due to agentive connotations. Fenk-Oczlon (1989) and Kiparsky (2009) deem iconicity as one of the most decisive criteria. Pragmatic constraints have been extensively discussed as strong indicators of word order; the first member is prioritized chronologically or socially (Malkiel, 1959; Fenk-Oczlon, 1989). Furthermore, prioritization of proximity is reported by Allan (1987) and the male gender by Wright et al. (2005), Hegarty (2014), and Hegarty et al. (2011). Finally, parameters such as specificity, focus, and topicalization have been known to influence word order (Quirk et al., 1972; Selkirk, 2011) and are also reported to do so in Greek (Horrocks, 1983; Alexiadou, 2002; Laskaratou and Georgiafentis, 2006). Syntactic and morphological complexity is also a parameter affecting constituent order. Ryan (2019) presents extraposition and heavy NP shift as potential syntactic parameters, and defines syntactic weight in terms of “word, node, or phrase count” (p. 319). Laskaratou and Georgiafentis (2006) provide an analysis for Greek, according to which constituents that have complex morphology or belong to complex categories, in the sense of consisting of multiple words, are more likely to occupy a position that is closer to the end of the sentence. Their examples comprise nominal phrases which also include adjuncts, and, consequently, they assert that their syntactic complexity is responsible for the positioning of these constituents closer to the end of the sentence, notwithstanding their syntactic function. Their conclusion is that even though Greek is characterized by a so-called “free word order,” sentences that include such complex

6 constituents at the beginning or at least not at the end of the sentence may not be deemed as grammatical. Another parameter that seems to have an immediate effect in the order of constituents is frequency as in Wright et al. (2005). McDonald et al. (1993) suggest that word order is predicted by cognitive parameters; since speech includes a memory component, the more readily available words are the ones that tend to occupy the first position of a binomial structure. This effect could be connected to the prototypicality argument of Cooper and Ross (1975) and Benor and Levy (2006), since what is frequent for a given speaker should be more central to a speaker‟s experience. Fenk-Oczlon (1989) follows a similar train of thought formulating a new rule of frequency which she presents as the most accurate in predicting the order of words in binomial structures. In her theory, the general frequency of a word, but also its frequency within discourse, is responsible for the speaker‟s familiarity with it and therefore influences its position within a prosodic structure. In the experiment of Siyanova-Chanturia et al. (2011), frequency effects are significant for the faster processing and recognition of words for both native and non-native speakers of English, which could be considered to be in accord with the arguments of cognitive strain influencing word ordering (Cooper and Ross, 1975; Fenk-Oczlon, 1989). Testing for frozen and non-frozen binomials, Morgan and Levy (2016) too come to the conclusion that frequent structures are faster and more easily recognized and furthermore processed holistically, in contrast to novel combinations. Due to its generally accepted great effect in word ordering, frequency has been controlled for many studies on binomials with the utilization of non-words, which can eliminate semantic and frequency effects, so that potential phonological effect can emerge (Bolinger, 1962; Pinker and Birdsong 1979; Oden and Lopes 1981; Oakeshott- Taylor 1984; Parker 2003), or through statistical regression (Wright et al., 2005; Benor and Levy, 2006). 2.2 Phonological Parameters and Partial Evidence from Greek Cooper and Ross (1975) provide a list of phonological constraints which seem to regulate word ordering in freezes. These constraints are only to be taken into account if semantic and pragmatic factors are controlled for, since the basis of their hypothesis is a collection of idiomatic binomials, the members of which have no semantic distinction with one another, but also no semantic content in isolation. The parameters that draw a word towards occupying the second slot of a binomial are (in the original order of formulation):

7

a. More syllables (which they term Panini‟s law) b. A longer vowel c. An onset comprising of more consonantal segments d. A more obstruent onset if both members start with one segment e. A vowel of lower F2 (i.e back vowels) f. Fewer final consonants g. A less obstruent final segment if both members end in a consonant. Based on evidence from Greek, this study focuses on experimentally testing and exploring potential effects of word length (in terms of syllable number), vowel backness (a lower F2), and the parameter of fewer final consonants by comparing null and singleton final codas. 2.2.1 Vowels 2.2.1.1 Quality The parameter of vowel quality has been a very central one in word ordering studies. Mentioned extensively in Jespersen (1961), Thun (1963), and Wescott (1970), are word order and alternation patterns that are based on vowel contrast. Cooper and Ross (1975) capture this contrast by proposing that a vowel with a lower F2 frequency, meaning of a further back quality, is more likely to occupy the right-most position. Jespersen (1961) also mentions height as a definitive aspect, which is also mentioned in Cutler and Cooper (1978), favoring a higher vowel for the first item. Ross (1982) has also reformulated the rule to further include vowel height, claiming that backness can only be considered a parameter if height is controlled for, remaining constant. Benor and Levy (2006) suggest that both height and backness could be considered important parameters in binomial ordering and should be analyzed individually. Effects of both are reported in Pinker and Birdsong (1979), who tested with English and French speakers using binomials of English and French nonce words, allowing the participants to take the test of their non-native language as well, in search of a potential universal principle of vowel quality in word ordering. They found that there is an effect of both vowel height and backness, with SLOT A being filled by vowels of lower F1 and higher F2, meaning that this position is expected to hold higher and fronter vowels than SLOT B. They further note that there must be multiple instances where this rule is not applied in its entirety, violating one of the two formant predictions. Strong effects for height are also reported by Oden and Lopes (1981), Müller (1997), but also Lohmann (2013), who presents significant effects for height, but not backness. In his nonce word experiment, Oakeshott-Taylor (1984) too tests for vowel effects for English, German, and Afrikaans, 8 with native speakers completing the task for their own native language respectively. His design involves the use of a binomial structure of A and B, and the placement of the target vowels in a [p_t] environment. His results illustrate a significance of vowel quality; effects of height for Afrikaans and backness for English and German. Cooper and Ross‟s (1975) main basis for their arguments on vowels has been frozen (near) -like expressions, the members of which are of low or no semantic content when in isolation. The Greek language too illustrates such structures in some echomimetic patterns, which could be considered so as to potentially reflect vowel quality alternations similar to the ones seen in End-Weight effects. Indeed, in his discussion of reduplicative compounds, Jespersen (1961) mentions that in Greek there are patterns where the kernels are identical, with the exception of a vowel. Some patterns following this model are found in Greek, yet they are of very low productivity, with only certain structures appearing in the lexicon. Table (1) Ablaut Patterns in Greek Pattern Example Meaning 'pafa-'pufa Sound of someone smoking a - u mats-muts Sound of kisses tik-tak Sound of a clock i - a din-dan Sound of a church bell tsiri'bim-tsiri'bom Silly or childish i - o tsiri'tri-tsiri'tro Sound of birds singing

However, the study of such structures for their potential reflection of processes found in natural speech is not without its criticism. Thun (1963) considers such segmental contrasts merely a matter of semantics rather than phonetics. In his analysis, a sharp perceptual contrast among the two vowels is essential for the formation of such structures ablaut structures. In other words, he suggests that a perceptual distance between the two vocalic segments can contribute to the semantic content of the structure. Minkova (2002) describes such English patterns (e.g. “riff-raff”) as complying to a phonological constraint INTEREST which ensures that the nuclei of the target syllables must be maximally different perceptually, thus avoiding mid vowels. Under Thun‟s (1963) and Minkova‟s (2002) analyses, accepting such cases as potential evidence for height and backness effects may be a somewhat far-fetched consideration. Ablaut reduplication freezes may form specific patterns in pursuit of an effect of

9 playfulness or alliteration (Ourn and Haiman, 2000), which may not particularly reflect speakers‟ spontaneous production of reversible binomials. Minkova (2002) suggests that their emergence and productivity are merely a function of stylistic and aesthetic effects. Furthermore, she notes that the members of this two-fold structure do not usually appear in isolation, in the manner that the constituents of other binomials, frozen or not, do. However, the utilization of these structures for the basis of phonological effects in binomials is widespread in the literature for their reflection of phonological contrast (Cooper and Ross, 1975; Parker, 2003), since their minimal pair structure cannot be easily found in binomial form in a natural language, especially one like Greek. Taking into account the vowel inventory of Greek, these coordinate structures may be a potential illustration of a backness effect, as Cooper and Ross (1975) and Pinker and Birdsong (1979) predict; a vowel of a lower F2 value is found in the second position, which can also be supported for Greek, considering F2 values found in table (2) measured for the five Greek vowels by Fourakis et al. (1999), and Nicolaidis (2003). It could be argued that an effect of backness is not enough to completely describe the patterns in table (1) since, in at least some of the patterns, there seems to be a height difference between the target vowels too, maximizing their perceptual contrast, reminiscent of Minkova‟s (2002) INTEREST constraint, but also Pordany‟s (1986:167) remarks on vowel contrasts for English and Hungarian binomials, for which she suggests that the backness rule only takes effect if the first constituent is host to a front high vowel. Table (2) Relative F1 and F2 values for the five Greek vowels (Hz) Vowel Fourakis et al. (1999) Nicolaidis (2003) F1 F2 F1 F2 i 331 2067 360.3 1892.3 e 477 1767 475.3 1671.6 a 715 1315 575 1452.7 o 492 1011 462 1202 u 344 961 376.7 1163.1

2.2.1.2 Length Another aspect frequently discussed in the literature is vowel length and tenseness. Since the vowel inventories of languages like English, German, and French are quite large and diverse, length and tenseness are two factors that cannot but be taken 10 into consideration. Indeed, the first parameter mentioned in Cooper and Ross (1975) regarding vowels is vowel length, with words containing longer vowels appearing in SLOT B. A significant role of length is described by Pinker and Birdsong (1979) for both French and English. Vowel length effects are also observed in a corpus study by Mollin (2012) but only when metrical and semantic constraints are excluded; however, in a similar corpus analysis, Benor and Levy (2006) do not detect a significant vowel length effect in a parallel condition, that is when excluding semantic and metrical constraints. In an even more detailed analysis of length, Oakeshott-Taylor (1984) reports phonetic and phonological length as important parameters in word ordering for English, German, and Afrikaans. His in-depth analysis delves into the phonetic properties of vowels, with intrinsic length being a highly ranked aspect for all three languages. His findings illustrate how length can be considered a continuum and not a categorically binary parameter, and are supported by Wright et al. (2005), who too report significant effects of vowel length in the same gradient manner. In her study, Minkova (2002) also suggests that length is an important aspect for the contrast of vowels in binomials as it further increases the perceptual difference between the two items. Although length was a feature of phonemic value in Ancient Greek (Revithiadou, 2004), the Greek vowel inventory does not include length contrasts (Kappa, 2002; Arvaniti and Baltazani, 2005), meaning that it is a phonologically inactive parameter for the scope of the current study. Phonetically, however, as Oakeshott-Taylor‟s (1984) findings suggest, the intrinsic length of the vowels could also be a potential parameter. Under his analysis a binary length distinction gives way to a gradience of length values, all related to the vowel inventory of the language tested. Under this analysis, intrinsic length could be an active parameter in the ordering of Greek binomials as well. Fourakis et al. (1999), and also Nicolaidis (2003), have respectively reported relative and absolute durations for the five Greek vowels in stressed positions, with high vowels being consistently durationally shorter than low vowels, but their reported values vary for [i] and [u]. Nicolaidis (2003), however, notes that her results on the high are not to be considered definitive, due to the limited instances of [u] in her data. Both the measurements of Fourakis et al. (1999) and Nicolaidis (2003) seem in line with Lehiste‟s (1970:18) crosslinguistic association of higher values of F1(height) with greater duration, which could, therefore, predict that if intrinsic vowel length is an important parameter in Greek binomial ordering, height effects should be expected to manifest themselves. 11

Table (3) Relative and Absolute vowel duration for the five Greek vowels (ms) Vowel Fourakis et al. (1999) Nicolaidis (2003) (Relative) (Absolute) i 61.31 69.1 e 80.94 80.9 a 90.96 85.4 o 82.79 78.4 u 84.41 59.8

Interestingly, some of these Greek patterns seem to violate the revised rule of height coming from Cutler and Cooper (1978) and Ross (1982), in the way Pinker and Birdsong (1979) predicted. Specifically, the a-u pattern is not in accord with the height parameter, since a low vowel occupies SLOT A, in favor of its agreement with the backness parameter (Pinker and Birdsong, 1979:506). In his analysis of English and Hungarian freezes, Pordany (1986) notes that [u] seems to never be followed by low front vowels, meaning that the height parameter is not really enough to explain patterns such as [a-u] in Greek, or „calm and cool‟ in English (Pordany, 1986:170). 2.2.2 Syllable Count The parameter mentioned first by Cooper and Ross (1975), also known as the “law of increasing constituents” (Behagel, 1909; in Mollin, 2012:85), or the “law of increasing members” (Abraham, 1950:282; in Parker, 2003:365), predicts that a given sequence of words will reflect a pattern of increasing syllables. Ryan (2019) terms it the Syllable- Count Effect, and notes that it is the parameter investigated the most in binomial studies and that its first ever mention was by ancient grammarian Pāṇini for Sanskrit dvandvas over 2,000 years ago. A trend of a short item followed by a longer one can be found in English, according to Malkiel (1959), while Bolinger (1962) also reports a tendency for a short item A and longer item B, which is lessened, but still very much observed, when there are longer stretches of unstressed syllables, or, in other words, effects of *LAPSE. In Pinker and Birdsong‟s (1979) experiment, the pattern is again observed, especially with native speakers and intermediate learners of French and English. Upon analyzing binomial ordering coming from a list of famous quotations, Kelly (1986) finds that there is also a tendency for short items to precede longer ones. In their analysis of corpus data, Benor and Levy (2006), Mollin (2012), but also Lohmann and Takada (2014), find significant effects of the parameter. The syllable count parameter is also confirmed in a small-scale experiment by Ryan (2019), in which he utilizes monosyllabic and 12 disyllabic proper names. He controls for both frequency and rhythm; the short names are deliberately chosen to be less frequent than the longer ones, while the longer names are stressed in the second syllable, thus violating *LAPSE, which proved to be an important criterion in Bolinger‟s (1962) results. Even with frequency and rhythm adjusted against the syllable count rule, the pattern that emerges reflects a significant short before long preference. The parameter is probably the first and most observed phonological parameter, with mentions coming also from Gustafsson (1975), Cooper and Ross (1975), Müller, (1997), and Wright et al. (2005). However, there have been instances where the results have been inconclusive (McDonald et al., 1993). Looking into natural language, in pursuit of a syllable effect and the obedience of structure to the Panini law, Kiparsky‟s (2009) study on Greek compound words can provide some interesting insights into the parameters partaking in the way these structures are formed. Kiparsky asserts that the word order within a compound structure is often dictated by aspects such as semantics and pragmatics, meaning that constituents appear in the logically expected order of the words they denote, a constraint he calls ICONICITY (Kiparsky, 2009:3). For example, a compound such as beno-vjeni „goes in and out‟ has this particular constituent order because it reflects the expected or „natural‟ order of the actions those constituents refer to. His analysis, however, includes a low- operating phonological constraint: SHORTEST-FIRST (p. 3). This constraint predicts that when there are no active semantic or morphological constraints, constituent order is affected by constituent length, with the shorter constituent appearing first, as in vixo- ftarnizete „coughs and sneezes‟ (p. 3). In other words, Kiparsky illustrates how word order is mostly affected by semantic factors, but also phonology, which operates on a lower level. Apart from Kiparsky‟s (2009) examples from Greek, there are other examples of the syllable count parameter, coming from freezes of idiomatic meaning with or without the Greek conjucation for „and,‟ [ce]: Table (4) Syllable effects: Evidence form Greek Expression Gloss Syllable Number Patterns pir ce ma'nia Very angry 1 and 3 'iζi ce 'eζima Customs (of a nation) 1 and 2 pe'tsi ce 'kokalo Very thin 2 and 3 bu'ca ce si'xorʝo Delicious 'xroɲa ce za'maɲa A very long time

13 i 'arta ce ta 'ʝanena Excessive 'tipos ce ipoɣra'mos Proper behavior 'enas ce monaði'kos One and only 2 and 4 apo'ro ce e'ksistame Expresses great surprise 'vios ce poli'tia Life and times 'ipa ce e'lalisa I said what I said u'e ce a'limono Exclamation of threat ri'ta ce katiɣorimati'ka Unequivocally 2 and 5 bros 'piso Back and forth 1 - 2 'vreksi ço'nisi Definitely 2 - 3 ma'ʎa kuvar'ʝa Mess

There are even some idiomatic multinomial freezes that seem to conform to the syllable effect: Table (5) Syllable effects in multinomials: Evidence form Greek Multinomial Syllable Literal Meaning Gloss Pattern pir ʝi.'ni ce 'ζa.la.sa 1+2+3 Fire, woman, and Dangerous things sea pa.'tris ζri.'sci.a 2+3+5 Homeland, religion, Nationalistic motto i.ko.'ʝe.ni.a family

These freezes illustrate an increasing length pattern that is in accord with the syllable count parameter. It is important to note that the majority of examples in table (4) illustrate the „2 and 3‟ and „2 and 4‟ patterns. 2.2.3 Codas Cooper and Ross‟s (1975) final consonant parameter refers to an item being more likely to occupy the second position of a binomial if it has a less complex final segment, or has a more sonorous final segment, in the case that both items have one final consonant. Coda complexity is not generally reported as a significant indicator by Benor and Levy (2006) and Mollin (2012), however, Pinker and Birdsong (1979) illustrate how their results favor a complex coda for the second item of the binomial. Ryan (2019:324) interprets complex coda first as an effect of vowel length, since a nucleic segment preceding a longer coda tends to be shorter “phonemically and durationally.” Regarding

14 the parameter of sonority, a similar argument of vocalic length is brought forward by Benor and Levy (2006:248) to explain asymmetries between voiced and voiceless codas in cases like “hit and hid,” since there is a tendency for nuclei to be affected by tautosyllabic consonants, generally accepted to be shorter or longer if adjacent to a voiceless or voiced consonant, respectively. Ad hoc tests on Bolinger‟s (1962) results conducted by Ryan (2019) indicate a strong effect of sonority in a gradient manner, with more sonorous coda segments having greater tendencies of occupying the second position of a binomial. Coda sonority effects are also obtained by the results on name ordering of Wright et al. (2005) and Mollin‟s (2012) corpus analysis. The significance of sonority could be interpreted in its relation to weight, as in some languages which exhibit weight effects only sonorous codas contribute to weight (Gordon, 2006:127). Bolinger (1962) draws a comparison of null to singleton codas, however, he finds that there is a 59% to 41% asymmetry disfavoring the item with the coda in the second position. Based on the relationship between weight and stress, Benor and Levy (2006) elaborate on Bolinger‟s (1962) results pertaining to the percentage where final stress is attested, noting that when the second item of a binomial has final stress, they would expect a preference for it to also have a coda. Such patterns would be in accord with the Weight-to-Stress principle (Prince, 1990), however, their expectations are not confirmed. Likewise, Mollin (2012) also reports a significant effect of an open main syllable appearing mostly as the first member of a binomial, upon excluding semantic and metrical constraints. It is fitting to note that vowels stand at the top of the sonority hierarchy (Selkirk, 1984; Parker, 2002), meaning that a reinterpretation of Cooper and Ross‟s (1975) “less obstruent final segment” could favor items of open syllables in the second position, since a final vocalic segment will always be less obstruent, and thus more sonorous than a consonant. A significant tendency for items in the second position not to end in a consonant is also indicated by the result of Wright et al. (2005), though not independently from the gender parameter. They rationalize this pattern by considering a binomial to be a prosodic unit complying to Prince and Smolensky‟s (1993) NOCODA constraint, according to which a syllable is optimal if it does not have a coda. The positioning of the coda-bearing unit in SLOT A could potentially result in re- syllabification, and thus not violate NOCODA (Wright et al., 2005:535). Ryan (2019) considers the effect of vowel length as an explanation for this tendency, especially in binomials of monosyllable words, which, when open, are reported to mainly contain long vowels or that could therefore influence the word ordering. This

15 tendency described by Ryan could also be the result of word minimality requirements for English, which is generally reported to be two morae. In Greek, codas have posed a puzzle for researchers, with many scholars, asserting that codas are generally unacceptable in the language. To begin with, the only coda segments that are considered legal word-finally (excluding loanwords) are /s, n/ (Kappa, 2002:116), but /r/ is also considered acceptable by some scholars, mainly due to its legality in Ancient Greek (Steriade, 1982), and probably the few words that survive in Greek bearing it (Holton, Mackridge, and Phillipaki-Warburton, 1997:12, Kappa, 2002:126). Word-medially, Kappa (1997a; in Malikouti-Drachman, 2002) considers [l] and [r] to be acceptable codas, unlike Setatos (1974), who rejects all word-medial codas. While Greek has been also analyzed as a strictly open syllable language (Malikouti- Drachman, 1984; Setatos, 1987), Kappa (2002) describes Greek syllabification as a process favoring onset maximization (Selkirk, 1984), while morphology usually operates on a secondary level, which makes some obstruent codas acceptable word- medially, as in the case that their syllabification to the next onset is restricted by phonotactic or morphological constraints. For example, in the compound word [pros+ka'lo] „invite,‟ we could argue in favour of a syllabification [pro.ska.lo], even though [pros] „towards‟ is a preposition and morpheme, the legality of a cluster [sk] triggers the resyllabification of [s] to the next onset. According to these constraints, onsets can only maximized if the emerging cluster is accepted as a legal onset word- initially, and thus the onset cluster can never include more than three segments; in the case of illegal clusters forming word-medially, irrespective of segments, morphological boundaries serve as syllable boundary indicators (Kappa, 2002). Such type of segmentation is mainly found in the event of a particle affixing at the beginning of words with complex clusters, making way for certain codas to appear word medially, as in [{ek}.stra'tia] „expedition‟ or [{ef}.spla'xnia] „mercy‟ (Kappa, 2002:137). These examples also pose a problem, since they illustrate how the so-called maximally complex onset in Greek (three segments long) can only have [s] as its first constituent (Holton, Mackridge, and Phillipaki-Warburton, 1997:15) thus violating the Sonority Sequence Generalization (Selkirk, 1984). For this reason, the syllabification of [s] plus consonant clusters has itself proven to be a matter of discussion. Goad‟s (2012, 2016) structural analysis of such clusters, has shown that, motivated by syllable well- formedness, [s] should be considered to be heterosyllabic and not a part of the onset. Goad‟s analysis is based on perception, and she maintains that for the greater part of s +

16 consonant cluster typology (including Greek), [s] should be syllabified as the coda of previous constituents (Goad, 2012, 2016). However, there has also been evidence for tolerance to codas within Greek. In a recent study by Gratsouni and Topintzi (to appear), when asked to syllabify words with internal obstruent-nasal clusters, native speakers showed a tendency to place a syllable boundary between the two consonantal segments, especially when they formed a cluster that is not accepted word initially by the language, and therefore accepting an obstruent coda word-medially. These results could indicate that native speakers are more tolerant of illegal codas, at least word-internally. Furthermore, the Greek lexicon seems tolerant of codas when it comes to certain loanwords. These loanwords come from various languages which allow a wider array of coda qualities (Swanson, 1958; Kappa, 2002). According to Kappa (2002) loanwords should not be considered definitively central to the Greek phonological system, since they constitute special cases that have infiltrated the Greek lexicon, and their foreign derivation is apparent. However, their incorporation into the Greek phonological system, with non-existent being adjusted to conform to the Greek inventory and processes (Setatos, 1974; Revithiadou, 1999; Kang, 2011), and many of them (mainly verbs) also acquiring productive suffixes (Ralli, 2018), indicates that there is no clear-cut boundary excluding words of foreign origin from the lexicon, since loanwords may go through various stages in their to the lexicon (Revithiadou, 1999). Finally, tolerance of codas can also be encountered in various regional dialects of Greece, which also allow for more types of coda segments to appear (possibly) word-medially and finally. These codas come to be mainly through a commonplace process among many Greek dialects, the deletion of unstressed high vowels (Joseph and Tserdanelis, 2003; Topintzi and Baltazani, 2012). Topintzi and Baltazani (2012) indicate that the emerging codas may comprise of multiple segments and even defy the Sonority Sequence Generalization (Selkirk, 1984). These codas may contain up to three consonants - also mentioned as the maximally complex coda for Greek by Kappa (2002), if loanwords are to be considered evidence for syllabic structure. In turn, these codas can also be problematic because, apart from their peripheral status in the language, complex codas violating sonority regulations have been considered by many to be extrasyllabic or extraprosodic under structural approaches (Goad, 2011; in Topintzi and Baltazani, 2012). It seems that there is great vacillation concerning the status of codas in Greek. While the existence of word-internal codas has been considered to be the function of a more prescriptivist set of syllabification rules, word-final codas are accepted with more 17 certainty, since they occur in a significant part of the lexicon (even to the exclusion of loanwords) (Holton, Mackridge, and Phillipaki-Warburton, 1997:12). For the purposes of this study, it would be safer not to take the existence of word-internal codas for granted, and codas are accepted at least word-finally in Greek. Word-finally, codas have shown interesting behaviors in relation to stress. Greek stress is usually lexically determined, but in any case, it must occur anywhere within the three final syllables of a word (Arvaniti, 1994; Nespor and Ralli, 1996; Drachman, et al., 1997:197), which has been termed the Stress Well Formedness Condition (Nespor and Vogel, 1986) or the three-syllable window (Malikouti-Drachman, 1994). There has been great debate regarding the default position of stress in Greek; the antepenultimate syllable has been proposed as the default locus for stress by Touratzidis and Ralli (1992), and Revithiadou (1999, 2007), while the penultimate is suggested by Malikouti- Drachman (2002), Kappa (2002), Protopapas et al. (2006) with a recent body of research indicating a preference for ultimate stress in non-lexical words, such as acronyms, especially upon the absence of morphologically familiar suffixes (Revithiadou et al., 2015) or coda segments licensed by the language (Topintzi and Kainada, 2012). Revithiadou et al. (2015) tested for stress preferences using novel Greek acronyms; their findings illustrate a tendency for ultimate and penultimate stress in items ending in vowels, but a clear-cut preference for ultimate stress when the items end in a consonantal segment or [e], which is not associated with a prominent morphological suffix. In a similar experiment, Topintzi and Kainada (2012) find that ultimate stress is the preferred option of native speakers, with regard to acronyms, especially when acronyms carry an illegal word-final coda. The pattern of stress being attracted to coda-bearing final syllables could be compared to Garcia‟s (2017) analysis of Portuguese gradient weight. In his analysis of nominal words, stress seems highly attracted to coda bearing syllables, with a gradient intensity fashion; ultimate syllables are more likely to attract stress than penultimate ones when bearing codas, with very few coda-bearing antepenultimate syllables being stressed. 2.3 The stress weight-interface and Prosodic End Weight (Ryan, 2019) In its first definition, End-Weight is associated with “complexity” (Quirk et al., 1972:944), or “grammatical weight” (Wasow, 1997:348), aspects which are realized in the form of longer constituents, syllable-wise. Considering cases of Heavy NP Shift and the English dative alternation, Anttila (2008) comes to the conclusion that there are significant prosodic motivations behind End-Weight manifestations. Within the framework of Selkirk‟s (1995) interface theory and formulation of the Nuclear Stress 18

Rule (Selkirk, 1995:562)1, according to which the most prominent syllable in a prosodic phrase is the stressed syllable that is closest to the right edge of said phrase, Anttila (2008) and Anttila et al. (2010) suggest that phrasal stress is one of the most important parameters for End-Weight. Since nuclear stress within a prosodic phrase is associated with the right edge (Selkirk, 1995), Anttila (2008:53) formalizes the Stress-to-Stress constraint, according to which stress on the phrase and word level must coincide. Remember that the environments in which End-Weight effects are observed have at least two possible manifestations (Ryan, 2016), meaning that there are at least two constituents, which are in competition to align with nuclear stress. Anttila‟s (2008) examples illustrate how longer constituents are likely to carry more word stresses, and thus satisfy the Stress-to-Stress constraint. This constraint assigns violation marks to lexical stresses that “fall outside of the phrasally stressed constituent” (Antilla, 2008:53). In (2), candidate (a) receives two violation marks as there are two lexical stresses in the constituent that does not align with position of the phrasal stress. Candidate (b) receives one violation mark as well for the lexical stress that falls outside the phrasally stressed constituent. Consequently, candidate (b) is the winner, because the constituent carrying the greater number of word stresses is also phrasally stressed. (2) Anttila (2008:53): Satisfying the Stress-to-Stress Principle Stress-to-Stress x x x x **! (a) Robertson gave [critical backing] [to Bush] x x x x * (b) Robertson gave [Bush] [critical backing]

Ryan (2019) suggests that the account on the importance of nuclear stress in End- Weight provided by Anttila (2008) and Anttila et al. (2010) cannot capture effects observed on the syllabic level. Focusing on the phonological aspects of the principle, he recognizes the significance of the influence of phonological weight. Especially for coordination, the original list of Cooper and Ross (1975) associates the second position of a binomial with a word containing features usually associated with increased phonological weight, such as more syllables, longer vowels, sonorous codas, and

1 Initially discussed by Chomsky and Halle (1968) 19 obstruent onsets. Ryan (2019) considers that manifestations of End-Weight pertaining to the syllabic level are an aspect of the interface between stress and weight (Gordon, 2006; Ryan, 2016), and the tendency of stress to be attracted to phonologically heavier constituents, at least in languages like English. He exemplifies his argument in (3) by providing the English freeze “trick or treat,” the item with the longer vowel is preferred in the second position, where it can align with nuclear stress, because it is heavier. (3) Ryan (2019:328): Satisfying the Stress-to-Stress Principle

x x x (a) treat or trick x x x (b) trick or treat

20

CHAPTER 3 EMPIRICAL STUDY 3.1 Parameters tested and Expectations The parameters tested in this study pertain to vowel quality (height, backness, intrinsic length), syllable count (disyllabic words compared to trisyllabic and tetra-syllabic ones), coda presence (null compared to singleton) while the latter was also tested in conditions of final and non-final stress, as well as, coda legality, sonority and . The research questions as listed in chapter 1. 1. Vowel quality:  Do words with vowels of further back and/or lower quality tend to appear at the right edge?  Is intrinsic vowel length a parameter that affects word order? 2. Syllable count:  Do words consisting of more syllables have greater tendencies to occupy the right edge? 3. Final Syllable:  Do words with final codas appear in the second position more often than words with open final syllables?  Does final stress have an effect? Do coda-bearing words appear in the second position more often when they are stressed word finally?  Do illegal codas influence word ordering?  Do more sonorous codas appear at position B more often than less sonorous codas do?

Table (6): Summary of parameters tested

Parameters Item B is expected to have

Vowels F2 vowels of lower F2

F1 vowels of higher F1

Intrinsic Length vowels of longer intrinsic duration

Word Length Syllable Count more syllables than A

Final Syllable Final Coda a closed final syllable

21

a closed final syllable, when it has Stress final stress a closed final syllable when coda is Legality illegal a closed final syllable when coda is Sonority sonorous

3.1.1 Vowels Based on the patterns seen in (1), it could be assumed that there is a backness tendency for Greek coordinate compounds, similar to the parameter listed by Cooper and Ross (1975), meaning that vowels of a lower F2 value would be expected to exhibit a tendency to occupy the second position of a binomial. However, since the vowels contained in these patterns are not front and back vowels of the same height, an influence of height cannot be dismissed completely. Intrinsic vowel length is also expected to produce significant results, since lower vowels in Greek tend to be longer (Fourakis et al 1999, Nicolaidis, 2003). 3.1.2 Syllable Count Taking into account that the syllable count parameter is listed as the most important phonological parameter by Cooper and Ross (1975), and the fact that this claim is supported empirically (Pinker and Birdsong, 1979, Benor and Levy, 2006; Ryan, 2019), Kiparsky‟s (2009) SHORTEST-FIRST constraint, and several binomial structures in Greek that illustrate a tendency for the longest constituent to appear second, effects of the parameter are expected to appear in the results of this study too. 3.1.3 Codas The final syllable parameter is a bit more challenging to make predictions for. Considering that in binomial ordering the greater body of research reports that there is tendency for coda bearing units to give way for open syllable to occupy the second slot (Bolinger, 1962; Cooper and Ross, 1975; Wright et al., 2005; Benor and Levy, 2006; Mollin, 2012), but also the marginal status of codas reported for Greek (Malikouti- Drachman, 1984; Setatos, 1987), coda bearing units should not appear in the second position of binomial structures. However, since studies focusing on stress and coda behavior in Greek specifically have illustrated a noticeable relationship between word- final codas – especially illegal ones – and stress (Topintzi and Kainada, 2012; Revithiadou et al., 2015), the results expected would be for coda-bearing units to be

22 preferred in the second position over items ending in open syllable, even more so when they are stressed or illegal. Finally, sonority is also a parameter for which predictions cannot be easily made either. One reason for this problem is that sonority has been studied mainly with the purpose of describing Greek syllabification under the Sonority Sequence Principle (Selkirk, 1984), and also the fact that there is great variability in the sonority scales that previous work has tested for the parameter. The most cited sonority hierarchy for Greek is by Malikouti-Drachman (1984), however, it does not include voiced oral stops [b, d, g] on account of their controversial autosegmental representation and inconsistency of distribution (Malikouti-Drachman, 2002; also see Arvaniti, 1999). This exclusion is problematic for the scope of this study, which also targets voiced stop [d] as a singleton coda. Steriade‟s (1982) hierarchy for ancient Attic Greek includes voiced stops, but lacks altogether, which are also targeted in this study. It is important to note that, in their respective hierarchies, the aspect of voice is parameterized by both Steriade (1982) – voiceless stops are less sonorous than voiced stops - and Malikouti-Drachman (1984) – voiceless fricatives are less sonorous than voiced fricatives. Therefore, Clements‟ (1990:296) simple obstruents < nasals < liquids hierarchy may not be enough to account for discrepancies within the Greek system. Instead, following Mollin (2012), Parker‟s (2002) fine-grain hierarchy in (4), which makes distinctions between voiced and voiceless phonemes, is utilized in this study, even though it was originally constructed for English. (4) Parker‟s (2002:69) sonority scale for English: p, t, k, tʃ < b, d, g, dʒ < f, ζ, x, ʃ < v, ð, ɣ, ʒ < m, n < l < r < j, w < h < vowels 3.2 General Points in Methodology Following Bolinger (1962), Pinker and Birdsong (1979), Oden and Lopes (1981), Oakeshott-Taylor (1984), Parker (2002), Wright et al. (2005), and Ryan (2019), this study makes use of forced choice tasks and aims at the elicitation of preferences. Two tasks were constructed to experimentally test the predictions in section 3.1 regarding phonological parameters in word ordering in Greek binomials. Target parameters appear in the stressed syllable of each word (Mollin, 2012), apart from the final coda parameter, for which the target segment is always word-finally irrespective of the stress. The consonantal segments that were tested in the final coda parameter are all alveolar [δ, d, ζ, t, l] but one [m]2, to match legal codas [s, n, r], but also due to its being the

2 However, in the second experiment segments [δ, ζ] were not targeted due to the scarcity of words, native or loan, bearing them in a final coda position. 23 richest natural class pertaining to place of articulation in Greek. The first task measures the parameters utilizing real words, while the second was constructed using nonce words. In both cases, the words were (or were constructed to resemble) nouns, adjectives, or verbs. The tasks were distributed through a web-based software (see Ryan, 2019), namely Google Forms. The utilization of such software reduces the risk error of manual data collection, accelerates the process of the collection (Ayoun, 2000), while reaching a greater number of participants faster. Participants were expected to fit a set of criteria, namely have Greek as their mother tongue, have virtually no experience with linguistics (meaning that they were either first year linguistics students, or had never taken a linguistics university course), and be between the ages of 18 and 30 for reasons of consistency. 3.3 First experiment A list of 68 sentences was presented to 85 anonymous participants. At the end of each sentence there was a non-freeze binomial structure [A and B], using the Greek conjunction [ce] „and.‟ This decision was made on the basis that corpus studies, such as the ones of Gustafsson (1975), and Benor and Levy (2006) report that the majority of binomials are found phrase finally. Participants were to read each sentence to themselves and indicate which order of constituents they thought sounds more “natural” (Bolinger, 1962; Pinker and Birdsong, 1979; Oakeshott-Taylor, 1984; Parker, 2002). The order of the sentences, as well as the two possible answers provided for every sentence appeared in a random order generated by the Google forms software. 3.3.1 Item Selection Item selection for the first task proved challenging due to the need for controlling non- phonological parameters that are reported to be significant, such as frequency. In all sentences, the pairings of words had to be new and somewhat random, so as to be distinct and different from any frozen forms already existing in Greek. Furthermore, the two items within a binomial were matched for frequency, using the Hellenic National Corpus (HNC) (Hatzigeorgiou et al., 2000). After Benor and Levy (2006), only the exact forms of the words utilized were searched, so as to exclude derived forms. The words were also matched in part of speech and number, and an endeavor was made so that they match in grammatical gender, have a similar phonetic form and semantic content, avoiding hyponymic or agentive relations, aiming at a balance similar to Malkiel‟s (1959) definition of binomials. Finally, the items of the first two parameters were also matched in number of syllables.

24

Natural language, however, imposes certain limits, thus, out of the 68 pairs, seven were slightly misaligned in frequency. Out of these seven pairs, six comprised of very low frequency words (

2+4 syllables 'daʎes ce nera'gules „dallias and buttercups‟ fa'ci ce faso'laða „lentils and beans‟ 'roði ce porto'kali „pomegranate and orange‟ mu'do ce melaŋxoli'ko „gloome and dreary‟ Final Syllable Final stress pal'to ce bu'fan „coat and jacket‟ 'fimi ce i'sçis „reputation and clout‟ aɣo're ce perma'nad „pixie cut and permed‟

3 Vowels were categorized for F1 and F2 values based on measurements by Fourakis et al. (1999) and Nicolaidis (2003), as shown in table (2). 25

Non-final stress bi'ʎarðo ce 'badmidon „billiard and badminton‟ ba'leto ce 'mjuzikal „ballet and musical‟ 'dzudo ce 'kricet „judo and cricket‟

No filler items were used since, apart from such items not appearing in the greater extent of similar experiments (Bolinger, 1962; Pinker and Birdsong, 1979; Oakeshott- Taylor, 1984; Parker, 2002), the sheer number of parameters tested allows for greater variation among items. (5) First experiment. Item example: sentence #7, targeting [i] versus [u] Σα γιπθά γηα ην ζρνιηθό πάξηη δελ πξέπεη λα πεξηέρνπλ μεξνύο θαξπνύο όπσο… ta ɣli'ka ʝa tν sxνli'ko 'parti ðen 'prepi na peri'exun ksi'rus kar'pus, 'opos… The deserts for the school party should not contain nuts such as… 1. θηζηίθη θαη θνπληνύθη. fi'stici ke fu'duci. peanut and hazelnut.

2. θνπληνύθη θαη θηζηίθη. fu'duci ke fi'stici. hazelnut and peanut.

3.3.3 Results and Discussion Following Oden and Lopes (1984:676), Benor and Levy (2006:251), and Mollin (2012:93), binomial4 distribution tests were used to determine whether there is a significant tendency for words of certain properties to appear in the second position of a binomial structure [A and B] more often. The test proportion percentage was set to 50%, which is translated to the null hypothesis, meaning that that there is an equal and balanced distribution between the two possible responses for every parameter tested. Success rates for each parameter and sub parameter can be found in tables (8) and (9). The first column includes the parameters, as seen in table (8), which were tested with a binomial distribution test. The second column refers to the total number of binomial structures tested, and the third includes the number of those binomials that exhibit the expected order, and the fourth column shows this number in a percentile form. Finally, the last column shows the p value for every parameter tested. The level of

4 As mentioned in Mollin (2012:92), the name of the statistical test utilized and the linguistic structure tested coincide, but are not related. 26 significance is set at p<0.05. Parameters found to be statistically significant are indicated by shaded rows. Table (8) First experiment: Results per parameter Parameter N Expected Order % p (1-sided) F2 1700 867 51% 0.424 F1 1700 813 48% 0.077 Intrinsic Length 1700 823 48% 0.199 Syllable Count 1700 919 54% 0.001 Final Coda 2380 1259 53% 0.005

To begin with, none of the vowel parameters were found to be significant. The numbers in table (8) also illustrate how the distribution of vowels between those two positions is at chance level. Remember that the basis for the argument of vowel effects, and especially backness, were fixed binary structures that are not particularly productive. It seems that this presumed backness pattern does not translate to real word ordering, much like Thun (1963) and Minkova (2002) suggest. Since these structures may constitute a special part of the vocabulary and do not appear in isolation (Minkova, 2002), it is probably the case that, even though these patterns may follow certain rules, natural language does not (Minkova, 2002:165).

Table (9) First experiment: Number of times that vowels appear in slot B Vowel N out of 676 % i 339 50% e 337 50% a 314 46% o 343 51% u 347 51%

A statistically significant tendency was found, however, pertaining to the syllable count and the coda parameters, meaning that items in the second position are more likely to have more syllables than the first item. This metrical constraint is probably the most widely reported and perhaps the most important phonological constraint in word ordering (Benor and Levy, 2006). In order to find out whether tetra-syllabic words have greater tendency to occupy the second position than trisyllabic words, a Wilcoxon

27 signed-rank test was conducted, but it did not yield a statistically significant result, Z=- .213, p=0.831. The final coda parameter also produced significant results, as seen in table (8) with the second position being filled with the coda-bearing word. In order to test for the sub-parameters having to do with coda behavior, namely stress and legality, a set of Wilcoxon signed-rank tests were conducted, following the methodology of Wright et al. (2005). The results can be found in table (10). Table (10) First experiment: Wilcoxon’s results per Final Coda sub-parameter Parameter N Mean (SD) Z p Stressed 85 0.4995 (0.1377)1 Stress -1.132 0.258 Unstressed 85 0.5327 (0.1669) Illegal 340 0.5341 (0.2143) Legality -2.573 0.010 Legal 340 0.4868 (0.3631) 1 a mean value closer to 1 indicates that results are in agreement with the expectations of the study.

While an initial Wilcoxon test yielded a significant effect of stress on word ordering, it could hardly be considered reliable. The imprecision lies in the fact that the unstressed parameter initially included real words, which were stressed on the penultimate or antepenultimate syllable. This is problematic, since antepenultimate stress is not considered to be more marginal, as it has been heavily disfavored in stress assignment tasks by Protopapas et al. (2006) and Revithiadou et al. (2015), but also because this asymmetry may not explain stressing preferences clearly. When the words that were stressed in the antepenultimate were excluded, a small tendency emerged for words of penultimate stress to occupy the second slot over words of ultimate stress (as can also be inferred from the Mean values), but did not yield significant results (Z=- 1.132, p=0.258). As can also be seen in table (10), the legality condition produced results that are compatible with the expectations of the study; as can be inferred from the mean values the number of times that words with illegal codas appear in the second slot (M=0.5341) is statistically higher than the one of words with legal codas (M=0.4868), Z=-2.573, p=0.010. The reliability of these findings, however, is also questionable. In order to target coda segments, especially such as [l, m, t, d] which are not acceptable as native codas word finally, the real words that contained them were all loanwords. At the very

28 least, this asymmetry could produce unreliable results pertaining to the legality parameter. In order to identify differences between Greek words and loanwords bearing codas, the binomials tested for this parameter were categorized into three groups; the first group contains binomials, the items of which are both Greek words (GG), the second contains binomials that include a Greek word and a loan word (GL), which also bears the coda, and finally, the third group contains binomials of two loanwords (LL). Binomial distribution tests were conducted so as to assess whether there is a significant divergence from the null hypothesis (a balanced distribution). The figures5 in table (12), illustrate a pattern. When a binomial contains words that belong to the Greek native lexicon, the coda-bearing unit is actually preferred in the first position. In the case that the two words are both loanwords, the preference is at chance level, with an almost balanced distribution. Finally, when the binomial consists of a Greek word (open final syllable) and a word of foreign origin (closed final syllable), the coda-bearing word is preferred in the second position. Table (12) First experiment: Binomial distributions per group Group N Expected order % p (1-sided) GG 425 176 41% <0.001 LL 1190 602 51% 0.706 GL 765 481 63% <0.001

As supported by Kappa (2002), Greek syllabification favors onset maximization, meaning that, if put phrase-medially, the consonantal segment could potentially re- syllabify and attach to the next syllable (Wright el al., 2005), becoming part of its onset.

5 Furthermore, a Friedman‟s test yielded statistically significant results and illustrated that the tendencies that coda-bearing words have to occupy the second position of a binomial are significantly different based on binomial type, ρ2(2)=30.494, p<0.001. These results are further supported by post-hoc Wilcoxon signed-rank tests, which were conducted to determine differences between the three types of binomials, with the Bonferroni correction setting the new significance level at p<0.017 (with the initial level of significance set at p<0.05). The Median (IQR) for the GG, GL, and LL group were 0.40 (0.20 to 0.60), 0.66 (0.55 to 0.78), and 0.50 (0.43 to 0.57), respectively. The differences among the three groups were all statistically significant: GG and GL (Z=-5.834, p<0.001), GL and LL (Z=-4.693, p<0.001), and LL and GG (Z=-3.211, p=0.001).

Table (11) Friedman’s test mean values per group Group N Mean (SD) GG 85 0.4141 (0.2018) LL 85 0.5059 (0.1300) GL 85 0.6288 (0.1728) A mean value closer to 1 means that the coda bearing item is preferred in the second position.

29

The possible clusters emerging from such process can be [sc], [nc], [rc], since the following word is always the Greek conjunction [ce] „and.‟ Syllabification placed aside, they are all attested clusters, at least word medially, as seen in (6), in contrast to [mc], t, d]. It is important to note that [sc] is also attested word-initially, but can arise as a cluster word-medially in compound words, and, though not attested word-initially, [nc] clusters can occur word-medially as heteromorphemic segments, but surface as [ɉ] due to the phonotactic rules of the language (see Arvaniti, 1999, 2000). (6) Legal word-medial clusters in Greek sc pros+cefali prosce'fali bedside nc (ɉ) en+ceros e'ɉeros on time rc arce'ta enough

This tendency could be interpreted as an attempt to avoid a word-final, and in this case also phrase-final, coda is in accord with supporters of Greek being a strictly open syllable language (Malikouti-Drachman, 1984; Setatos, 1987), At the opposite edge of the “spectrum,” when placed in a binomial along with a Greek word ending in a vowel, the coda-bearing loanword is more likely to appear in the second position. This tendency cannot positively be attributed to the presence of an illegal coda, as in the group of binomials that consist of two loanwords (and therefore, also a comparison between an open final syllable and an illegal word-final coda) there is no significant word order preference. This asymmetry can be explained in the sense that the pairs of this group do not really conform to the definition given for binomials by Malkiel (1959), as these words cannot be really considered to belong to the same word class, for loanwords have been reported to have their own phonological patterns, as illustrated by Ito (2008, 2014) for Yanbian Korean. Finally, even though sonority has been reported by many scholars (Bolinger, 1969; in Ryan, 2019; Cooper and Ross, 1975; Wright et al., 2005; Mollin, 2012) to be a significant indicator in word ordering, it did not yield any statistically significant results in this experiment. It is important to note that in its original formulation by Cooper and Ross (1975) the sonority parameter is only significant when both words have a singleton coda, which is not the case for the items tested here. Its significance has been mainly reported for weight-sensitive systems, due to the association of higher coda sonority with greater syllable weight (Gordon, 2006), but it does not seem to have any

30 significant effect in Greek6. Weight distinctions aside, the insignificance of sonority is really not surprising for a language like Greek, for which sonority is a matter of disagreement. Discussed mainly as an aspect of syllabification, sonority scales have failed to predict certain patterns found in Greek, such as the syllabification of word medial obstruent plus nasal clusters (Gratsouni and Topintzi, to appear) 3.4 Second experiment The second task was designed with the purpose of remedying some of the imprecisions that may have impeded the production of more reliable results in the first experiment. The second experiment utilized a similar methodology to the first one, only in this task, the binomials were made up of nonce words. Nonce words have proved to be a useful tool in eliminating any influence of frequency and semantic impact on the results that might be found in the first task, while also providing the opportunity to control for phonological effects through the creation of minimal pairs, and control of stress (Oakeshott-Taylor, 1984:227; Parker, 2002:370). It consisted of 86 sentences and was completed by 86 anonymous participants through the same online platform. 3.4.1 Nonce word construction The nonce words appeared within a sentence much like in the first task and similarly to the task administered by Pinker and Birdsong (1979). The nonce words were presented in orthographical forms, since the Greek orthographical system is quite transparent, with a somewhat balanced relationship between graphemes and phonemes (Porpodas, 2006). The use of orthography for the utilization of nonce words is found in binomial order studies (Pinker and Birdsong, 1979; Oakeshott-Taylor, 1984; Parker, 2002), but also quite widespread in Greek studies as well, especially by Protopapas (Protopapas, 2006; Protopapas and Vlahou, 2009; Grimani and Protopapas, 2017). Nonce words used for this experiment were constructed in such a way as to follow the phonotactic rules of the language, but not resemble existing Greek words appearing in fixed binomial structures. Basic CVCV structures7 were utilized for the construction of nonce words and

6A Kruskal-Wallis test indicated that, among the number of times that coda-bearing items appear in the second position, there is no effect of sonority, H(5)=5,893, p=0,317. 7 Vowels: (CV)CVCV(C) structures were utilized for the construction of the nonce minimal pairs with the target sound being placed in the nucleus of the penultimate syllable (bold). They were all stressed on the penultimate. Syllable count: CVCV(C), CVCVCV(C), CVCVCVCV(C), structures were created for the disyllabic, trisyllabic, and tetra-syllabic words respectively. The vowels of the ultimate and the penultimate syllable were identical within every pair. They were also all stressed in the penultimate. Final Syllable: simple (CV)CVCV and (CV)CVCVC were also used here, however there was an asymmetry, with only one of the pairs being a coda-bearing unit. Since final and non-final stress was a parameter for this category of words, stress was on the penultimate for half of the pairs, and on the ultima for the rest. 31 complex clusters were avoided so as to avoid inconsistencies in syllabification by the participants, which could introduce potential effects of onset or coda complexity. For the current design, the target sounds were always placed in the stressed syllable (Mollin, 2012), for a stressed syllable in Greek is more prominent, higher in amplitude and pitch (Arvaniti, 1994). For the present experiment, and when stress in not targeted as an experimental parameter (i.e. final stress in the coda parameter), all nonce words are accented on the penultimate syllable, to achieve consistency. Another reason for penultimate stress was, the fact that Greek is reported to have fusional morphology (Apoussidou, 2003), meaning that a great deal of semantic content, such as gender and number, lies on the inflectional suffixes of the words, and grammatical gender had to be matched within binomials, so as to avoid effects of gender (Wright and Hay, 2002, Wright et al., 2005). Since the items within a binomial were to be matched for grammatical gender and number, not much variation was possible for at least the rime of the final syllables of the words, eliminating final syllables as a target position. For the vowel and the syllable count parameter, the rimes of final syllables of the words constructed resembled Greek nominal (e.g. -os, -a, -i) and verbal suffixes (eg. -o, -is) but they were always matched within the same binomial. Productive suffixes like adjectival [-ikos] or verbal [-evo], were excluded. For the coda parameter, the nucleus of the final syllable was also matched, to avoid any vowel effects. 3.4.2 Stimuli The 86 sentences included a binomial each. First, thirty binomials targeted vowels. The nonce words constructed for this part, were all minimal pairs, with alternations in the vowel of the penultimate syllable. There were three pairs for every possible combination of the five Greek vowels. For the syllable parameter, twenty binomial structures targeted the syllable parameter; ten tested tor the 2+3 pattern, and 10 for the 2+4 pattern. In this case, there was more variation, but the vowels of the penultimate and ultimate syllable were always matched within each pair. Finally, for the final coda parameter, there were thirty-six pairs; four binomials for every phoneme targeted [s, n, r, l, m, t, ζ, d, ð], two for each stress condition (final and non-final stress). Table (13) Second experiment: Examples of binomials utilized Parameter Target Binomial Vowel Quality [e] and [a] 'tesa ce 'tasa [e] and [u] 'ɉeζis ce 'guζis [o] and [u] po'toɲa ce po'tuɲa

32

Word Length 2+3 syllables 'pemi ce pe'teci 'gaζas ce ka'vadas 'koɣon ce ka'ðokon 'foba ce ta'voka

2+4 syllables 'bosa ce meku'pota 'doci ce boka'lovi 'ɉemis ce dova'metis 'gara ce kuta'nala Final Syllable Final stress ce'ҫi ce te'vis ko'ða ce to'ɣar go'ni ce bo'miζ

Non-final stress ta'diva ce ka'diðal 'dafo ce 'gaxod 'ceno ce 'temor

(7) Second experiment. Item example: sentence #13 targeting [e] and [a]. Ζ αιπζίδα είρε ζθνπξηάζεη εληειώο, είρε γίλεη ζθέηε… [i ali'siða 'içe skur'ʝasi ede'los, 'içe 'ʝini 'sceti] „The chain had rusted completely, it had become… 1. ηέζα θαη ηάζα. ['tesa ce 'tasa] tesa and tasa. 2. ηάζα θαη ηέζα. ['tasa ce 'tesa] tasa and tesa.

3.4.3 Results and Discussion As in the first experiment, results for the first parameter were obtained through binomial distribution tests, with the test proportion value being set at 50%. The results are summarized in table (14). The first results of this experiment are comparable to the result of the first, however, in this experiment, vowel backness seems to also have produced a statistically significant result. It appears that vowel backness being a 33 significant indicator in word ordering is contradictory to the results of the respective parameter in the first experiment. One possible explanation for this result lies in the very construction of the task. The use of nonce words eliminates any effects of semantics, but since there is no lexical specification for these words, they are not processed in the way that a real word would (Protopapas, 2006).

Table (14) Second experiment: Results per parameter Parameter N Expected Order % p (1-sided) F2 2580 1355 53% 0.011 F1 2580 1339 52% 0.056 Intrinsic Length 2580 1323 51% 0.201 Syllable Count 1720 1104 64% <0.001 Final Coda 3096 1613 52% 0.020

Protopapas (2006:176) suggests that if nonce words resemble real words, then they may be read “by analogy,” meaning that the minimal pairs used in the second experiment, which are very comparable to the structures seen in table (1), could be perceived as indeed resembling them, and therefore, expected to comply to the same back template. Table (15) illustrates this tendency of items containing vowels with a lower F2 to appear in the second position of a binomial in a somewhat gradient pattern.

Table (15) Second experiment: Number of times that vowels appear in slot B Vowel Number out of 1.032 %

i 505 50% e 496 43% a 561 54% o 717 69% u 731 71%

As in the first experiment, the syllable parameter produced significant results. A Wilcoxon signed ranks test was conducted to determine if longer words illustrate greater tendencies to appear at the second position of a binomial structure. It seems that, in this task, the longer a word is, the less likely it is to appear in the second position, with trisyllabic words having greater tendencies to appear in the second position than

34 tetra-syllabic ones8, Z=-2.864, p=0.004. It is important to note that, in this parameter, all words were stressed in the penultimate no matter their length, consequently, a disyllabic word preceding a tetra-syllabic one could produce lapse9 effects. Violations of *LAPSE could be ameliorated if the tetra-syllabic word occupied the first slot. *LAPSE is violated two times in (8a), for every unstressed syllable not adjacent to a strong beat, while only one in (8b) since there is only one syllable not adjacent to a strong beat. (8) Violations of lapse a. 'feci ce fiko'reti b. fiko'reti ce 'feci The effect tetra-syllabic words present was not observed in the real words task, maybe because stress is not controlled and there is individual variation in the stress patterns that each binomial produces. Even though lapse might be a definitive factor in word ordering for some languages such as English (Bolinger, 1962; Benor and Levy, 2006; Mollin, 2012), even without the control of sematic constraints, its potential significance in Greek is debatable. Greek is saturated with lapses on account of its abundance of polysyllabic words, and the limitation of one lexical stress per word (Arvaniti, 1994; Revithiadou, 1999). Arvaniti (1994) asserts that the language is very tolerant of lapses, since, even though it resolves clashes of stress by the means of vowel lengthening, it does not usually remedy long stretches of unstressed syllables, as stress shifts are banned (except for when the Stress Well Formedness Condition is violated or a stress shift is dictated by morphological conditions, also see Revithiadou, 1999; Arvaniti, 2007; van Oostendorp, 2012), and thus, at least phonetically, lapse is not ameliorated in any way in the Greek language (Arvaniti, 1994, 2007). One limitation of this study is that the metrical constraints of lapse and non- finality have not been at studied in detail for the current experiments. Ryan (2019) suggests that even though such constraints are often considered and tested in End- Weight studies, they are not indeed definitive factors in End-Weight manifestations, on the basis that they might be overridden by the syllable count parameter, while they also

8 Even though tetra-syllabic words do not appear in the second position as often as trisyllabic words do, further binomial distribution tests indicated that there is still a significant tendency for them to appear in the second position with an observed proportion of 61%, p<0.001 (1-sided). 9 Elenbaas and Kager (1999:282) formalized the constraint in the framework of Optimality Theory: *LAPSE Every weak beat must be adjacent to a strong beat or the word edge 35 fail to capture the influence of phonological parameters that pertain to the sub-syllabic level, such as vowel quality or onset/coda complexity (Ryan, 2019:346). The final coda parameter has also produced a significant main effect in the binomial test, as coda-bearing items appear in the second position of a binomial, yet with a relatively low proportion of 52%, p=0.02. Even though, the results of the first experiment for the corresponding parameter indicated that it is statistically significant, recall that there were great discrepancies in the results according to binomial type. These discrepancies are not relevant in this case, due to the use of nonce words. As done in the first experiment, the results for the coda parameters pertaining to stress, legality, and voice, were elicited through Wilcoxon‟s signed-rank tests. A summary of the results can be found in table (15). Table (16) Second experiment: Wilcoxon’s results per Final Coda sub-parameter Parameter N Mean (SD) Z p Stressed 172 0.5092 (0.1877) Stress -1.741 0.082 Unstressed 172 0.5297 (0.1839) Illegal 344 0.5213 (0.2175) Legality -0.099 0.921 Legal 344 0.5203 (0.2972)

In this experiment, the Wilcoxon‟s signed-rank tests did not produce any significant results regarding the influence of final stress or coda legality on word ordering. One interesting finding is that, for the legality parameter, legal and illegal coda segments seem to have the same behavior; these results are very similar to the ones of the first experiment, specifically for the LL group. Considering that loan words are more peripheral to the native lexicon (Kang, 2011), maybe nonce are also peripheral, mostly in terms of their phonological structure. Finally, through a Kruskal-Wallis test, coda sonority was not found to have a statistically significant effect in word ordering, H(6)=3.305, p=0.770, which is quite comparable to the respective results of the first experiment.

3.5 Summary of Results The findings of the two experiments illustrate how Greek does indeed exhibit End- Weight effects. The expectations of this study were partially confirmed. Firstly, in terms of vowel quality, vowel height did not prove to be a significant parameter. The parameter of intrinsic length did not yield any significant results as expected. Vowel

36 backness, on the other hand, produced significant results as expected, but only in the nonce word task. Regarding the Final Coda parameter, both experiments produced statistically significant results for the final coda condition, with coda bearing units appearing in the second position significantly more often. However, this is considered an unreliable result for the first experiment, due to asymmetries arising from the distribution of native Greek words and loan words within the binomials. This is also the reason why the sub-parameter of legality initially produced significant results in the first task. The following table (16) summarizes the findings of the two experiments. The findings are quite consistent within the two tasks, with the syllable count and the final coda parameter yielding significant effects in both tasks. An asymmetry can be seen regarding vowel backness (F2), which only produced significant effects in the nonce word experiment. A second asymmetry can be seen regarding the parameter of Legality, which appears to be significant only in the real word experiment.

Table (17) Summary of Results Real words Nonce words F2 * F1 Intrinsic Length Syllable Count * *! Final Coda * * Stress Legality (*) Sonority *! 2-4 > 2-3 longer words have greater End Weight propensity * p<0.05, (*) p<0.05 against the expected result, but post-hoc tests proved it unreliable

The first asymmetry could be explained on the basis of phonology being overridden by factors such as semantic and cognitive relations among words (Pordany, 1986; McDonald et al, 1993). Benor and Levy (2006) and Mollin, (2012) found that phonological factors only yield significant results when semantic parameters and metrical constraints are controlled. Even though in the first task an attempt was made to

37 control semantic parameters, the discrepancy of results regarding vowel backness indicate that they were probably not successfully controlled. Semantics aside, the items of the first task were not completely controlled for metrical parameters either, since, even though words had the same number of syllables, the variation of stress between them made them subject to lapse. On the other hand, in the nonce word experiment, the influence of semantics was bypassed altogether, and metrical predictors were controlled, as words within a binomial had the same number of syllables and the same penultimate stress. Within this context, the parameter of vowel backness seems to be a significant predictor in word ordering, a result that is in accord with the findings of Benor and Levy (2006) and Mollin, (2012).

38

CHAPTER 4 GENERAL DISCUSSION This section constitutes an endeavor in interpreting the results of this study focusing on discussing two popular accounts of End-Weight motivations that can potentially capture all of the findings of the current study; namely the effects of frequency and the stress- weight interface. Even though, at least in this study, Greek seems to have certain End- Weight manifestations, the motivation behind the tendencies found remains vague, as neither account can fully explain the findings of this study. Given that the parameters of word ordering as stated in the principle of End-Weight have been discussed mainly for weight-sensitive systems, the observation of its effects in Greek could be interpreted in two manners. First, maybe the patterns observed are not exclusive to weight-sensitive languages, and the regulation of their emergence is motivated by cognitive aspects (Cooper and Ross, 1975; Fenk-Oczlon, 1989; Benor and Levy, 2006). A second interpretation could be that maybe the Greek language presents itself as a type of weight-sensitive system with an active stress-weight interface, after all (Garcia, 2017, 2019; Ryan, 2019). 4.1 Frequency, Markedness, and Accessibility An explanation for the manifestation of End-Weight phenomena is the principle that the second position is filled with the item that is not as central to the lexicon, due to its low frequency and high markedness, which in turn translates to increased requirements in cognitive processing. In other words, infrequent words are not as easily accessible and are thus considered marked due to their peripheral status in the lexicon. Frequency is reported by Fenk-Oczlon (1989) to be the most accurate and significant indicator in word ordering. She proposes that both the phonological and the semantic predictor listed by Cooper and Ross (1975) and then revised again by Ross (1980) can be explained in terms of high before low frequency which she summarizes in the constraint HF>LF (Fenk-Oczlon, 1989:251). The main argument in favor of the frequency interpretation is that frequent linguistic items are more readily accessible, but they also lift the cognitive load, as faster recognition of words is facilitated by their high frequency. Fenk-Oczlon (1989) asserts that her collective frequency constraints can accurately predict the correct word order of 84% of her corpus of frozen binomials, while also maintaining that frequency effects can only be applied to the word level, However, experimental evidence indicates that frequency effects can also be relevant to “sub-lexical, phrasal, and clausal levels” (Siyanova-Chanturia et al., 2011:776). Testing

39 for reaction times in the process of binomials, Siyanova-Chanturia et al. (2011) report the significance of a phrasal frequency in faster , rather than item frequency, and argue that the entrenchment in a language generates a dynamic statistical grammar, which predicts that the most preferred linguistic unit or structure is the most frequent. Fenk-Oczlon (1989) maintains that vowel effects in ablaut binomials (e.g. tick- tack) cannot be explained under her frequency analysis, because, as she supports, the items of these structures do not appear in isolation, meaning that they are not really binomials, but more so words in their own right. However, since the alternation in the structures revolves around one phoneme, maybe phoneme frequency is a reason for the patterns arising in this study. In their corpus analysis, Nicolaides et al. (2003) report that front vowels are more common than back vowels in Greek, which can potentially explain the backness effect reported for the vowel parameter of the second experiment; since back vowels are more infrequent, the frequency account correctly predicts that the words that contain them appear in the second position more often. An interesting finding in this study is that, particularly, [u], the vowel with the lowest F2 value for Greek reported by Fourakis et al. (1999), but also Nicolaidis (2003), appears in the second slot of a binomial the highest number of times in both experiments10, which can further support the frequency account as this phoneme is reported as the most infrequent Greek phoneme in measurements of spontaneous speech by Nicolaides et al. (2003)11. This argument, however, cannot be supported by other studies focusing on frequency and reaction times; for example, Siyanova-Chanturia et al. (2011) note that the lower component for which frequency effects can be observed is the morpheme, not the phoneme, meaning that maybe the infrequency of back vowels in Greek cannot motivate the backness effect found in this study. Ryan (2019) also mentions that the most well-known aspect of word ordering, the syllable count parameter may also be related to frequency, (also supported by Fenk- Oczlon, 1989; McDonald et al., 1993; Wright et al., 2005), in the sense that shorter words may also be more frequent, as also supported by Fenk-Oczlon (1989:522), who asserts that words of more syllables are negatively correlated to frequency of use cross- linguistically. This study draws a comparison between short and long words, and short words were selected and constructed to be two syllables long in both the real and nonce

10 Out of all of its appearances in each task separately, [u] appears in the second slot of a binomial with a frequency 51% for the first task – with [o] being a very close second – and an impressive 71% in the second task. 11 With [o] also being a close second, which further supports a vowel backness argument based on frequency for word ordering in Greek binomial structures. 40 word task respectively. The argument in favor of frequency could also explain the significant tendency of short words to appear first within a binomial in Greek; a quick overview of the HNC (Hatzigeorgiou et al., 2000) showed that within the five most frequent nouns, adjectives, and verbs (i.e the types of words used in this study) there is only one trisyllabic word, while the rest are all disyllables. McDonald et al. (1993) also mention the impact of lexical accessibility on word ordering, with the argument that the phonological form of the word is easier to retrieve when it is more compact, and has fewer syllables. It seems that the argument regarding the significance of a word‟s cognitive load is also relevant to the effects of the syllable count parameter. First, Benor and Levy (2006) discuss derived forms being usually longer, as they consist of more morphemes, more infrequent, and therefore more marked. In general, derived forms are considered to be more complex on account of their higher morphemic content, even in word length is controlled (Lohmann 2013:130, 136). Longer or more complex items are found to be preceded by longer pauses (Zvonik and Cummins, 2003) which could point to their taking longer to process. In a picture-word association experiment by Kelly et al. (1990), a significant tendency is reported for complex shapes to be associated with longer words but also cognitively complex notions. Put differently, longer words tend to be associated with a greater degree of complexity, perceptually and formally, and they are, therefore, more likely to appear in the right edge of a prosodic phrase, illustrating “heavy-to-periphery phenomena” (Shih and Zuraw, 2017:e323). Cooper and Ross‟s (1975:67) cumulative constraint MEFIRST, dictates that the item that occupies the first position of a binomial has the most unmarked form and content of the two items and “refers to those factors which describe the prototypical speaker.” Fenk-Oczlon (1989) also discusses perceptual markedness relating to frequency, suggesting that infrequent words are less likely to undergo reduction and assimilation processes, and more likely to be pronounced more emphatically, which makes them stand out perceptually (also see Phillips, 1984). The notion of markedness is formalized into two constraints for word-ordering by Benor and Levy (2006:236-7). The first constraint, perceptual markedness, pertains to the cognitive aspect of markedness, in the manner Cooper and Ross‟s (1975) describe. The second markedness constraint, formal markedness, refers to aspects such as broader meaning and simpler form (as also discussed by Battistella, 1990). The first markedness constraint could only be considered relevant to the first experiment, where the items have semantic content, but were matched within a binomial in terms of animacy, number, and concreteness, tackling the main aspects listed by Battistella (1990). The second constraint is relevant 41 to both experiments in terms of formal complexity, and seems to be active as longer words were found to have a statistically significant tendency to appear at the second position. The conclusion that can be drawn is that, if words with more syllables are considered to be more infrequent and inherently denote more complex ideas, as well as be more marked in terms of form, and thus more challenging to retrieve from memory, then, when put in a binomial context along with a word of fewer syllables, the longer word is expected to appear in the second position, as can be observed in the results of this study, but also in the majority of studies dealing with word ordering in binomials. Turning to the final-coda parameter, the frequency and markedness argument seems to fail to account for the behavior of codas in Greek binomials. Certain consonantal segments are found word-finally in Greek, namely [s, n, r], but reflect a small part of the lexicon (Holton, Mackridge, and Phillipaki-Warburton, 1997:12). A great deal of other consonantal qualities can also be found in words of foreign origin12. In general, since word-final codas, especially illegal ones are not very frequent, they are expected to constitute marked structures, and therefore appear in the second position of a binomial. This prediction is consistent with the initial results of both experiments. However, a closer examination of the results suggests that the infrequency of word-final coda segments cannot be a definitive indicator of word ordering in Greek. The utilization of loanwords in the first experiment was essential in order to target word- final codas that are otherwise unacceptable in the native lexicon. The results of the first experiment illustrate that coda-bearing words behave differently according to their native or non-native status in the lexicon. First, the GG group yielded some interesting results, as words with codas illustrated a tendency to appear at the first position. Furthermore, when a binomial contains a Greek word and a loanword ending in a consonant, there is a high preference for the loanword to appear in the second position. Kappa (2002) notes that loanwords constitute special units of the vocabulary, and that their foreign origin is recognizable, meaning that they constitute a more peripheral stratum within the Greek system. The results of the GL group indicate that maybe coda- bearing words appear in the second position due to the markedness of the loanwords themselves rather than a word-final consonantal segment. This argument is further corroborated by the results of the LL group; when a binomial consists of two loanwords, the choice seems to be at chance at an almost balanced distribution, meaning that when both words are equally marked in terms of origin, there is really no indication that the

12 Swanson (1958) provides a comprehensive list of loanwords of the Greek lexicon, mostly of English and French origin. 42 coda renders the word further marked. The second experiment included more uniform items, since they were all nonce words. This experiment produced significant results for the final coda parameter, as coda-bearing units were found to have a statistically significant tendency to appear in the second position of a binomial. This finding seems in line with the frequency account. However, an interpretation of frequency and markedness should not be considered successful. Even though, word-medial codas have been reported to be marginal and therefore marked within the Greek system, word-final codas are somewhat more acceptable (Malikouti-Drachman, 1984; Setatos, 1987). A frequency account would predict that illegal coda segments should be even more marked due to their infrequency and illegal (thus, marked) status and, therefore, appear in the second position more often. This prediction is not confirmed as the sub-parameter of coda legality produced null results. This means that, at least in this experiment, frequency and markedness cannot be considered reliable indicators of the behavior of coda-bearing units. To sum up, an explanation of frequency, markedness, and accessibility seems to be able to partially capture the End-Weight effects observed in this study. Namely, while it can explain syllable effects, it fails to account for the aspect of vowel quality in the second experiment, which cannot be positively interpreted as a function of lexical frequency, but also certain aspects of final codas effects. 4.2 Phrasal Stress and Weight More recent studies on End-Weight suggest that phrasal stress is a highly significant predictor in constituent ordering. Anttila (2008) and Anttila et al. (2010), extensively discuss the importance of phrasal stress on instances of English dative alternations. Focusing on the syllabic level and considering the list of Cooper and Ross (1975) on phonological parameters in word ordering, Ryan (2019) also regards phrasal stress to be an elementary component of End-Weight in coordination. By assuming that the default locus of nuclear phrasal stress is closer to the right edge of the prosodic phrase (Selkirk, 1995), he suggests that this position is expected to be associated with constituents of greater prosodic weight, at least for weight-sensitive languages that manifest End- Weight effects. His argument pertaining to the relevance of End-Weight manifestations to phrasal stress lies in the extensively discussed interface between stress and weight, and their tendency to attract each other (Gordon, 2006; Ryan, 2016). In other words, the phonological parameters associated with the second position of a binomial (Cooper and Ross, 1975), are associated with greater phonological weight, and are consequently

43 better candidates to coincide with nuclear phrasal stress. Therefore, Ryan‟s (2019:328) main argument is that “[prosodic end weight] reflects phrasal stress.” On the other hand, Greek is widely regarded as a weight-insensitive system (Kappa, 2002), which means that the effects observed in the two experiments of this study cannot seemingly be explained under the phonological weight and nuclear stress interpretation. However, new types of weight-sensitive systems have been proposed, which deviate from categorically binary distinctions and reflect gradient hierarchies (Gordon, 2006; Ryan, 2014, 2016), or even illustrate weight effects only on a limited part of the lexicon (Garcia, 2017), which could be relevant to effects reported for Greek. One particularly interesting example of an alternative weight-sensitive system is (as discussed by Garcia, 2017). Similarly to Greek, this language allows a trisyllabic window for stress, while the most common locus of stress for non- verbal forms is the penultimate syllable. Brazilian Portuguese weight effects manifest in non-verbal forms, as syllables containing diphthongs, nasal vowels, or a coda, pattern as heavy and, therefore, attract stress. Verbal forms are excluded from weight effects on account of their morphological marking, which influences stress assignment, inhibiting weight-stress effects from emerging. What further singles out Brazilian Portuguese form other languages reported to illustrate stress-weight effects is that, in this case there is a gradient pattern of decreasing intensity of weight from the ultimate to the antepenultimate syllable. In other words, the closer a heavy syllable is to the right edge of the word, the more likely it is to attract stress. Garcia (2017) emphasizes that there is variance in the lexicon, meaning that the weight effects he reports are not categorical, but rather significant tendencies. In Garcia (2019) this variance is addressed and the influence of morphology on stress assignment is made clearer, and thus, in search of more systematic patterns, nonce words experiments are used. Their results illustrate that speakers seem to apply gradient weight generalizations in nouns across the board in a more systematic way when morphology is controlled. Garcia (2019) therefore argues that in the absence of morphological cues, more consistent patterns emerge and he even argues that the patterns observed in nonce words are actually more reliable in explaining phonological properties that may otherwise not manifest as extensively in the language due to morphological interception. In his study, Garcia (2019) actually showcases how the phonological grammar of a language and its operations are actually learned by speakers, as well as how the use of nonce words can uncover that when free of lexical exceptions and morphological cues, these operations take effect.

44

Recent studies on Greek stress assignment have illustrated tendencies that are comparable to Garcia‟s (2017) report on Brazilian Portuguese weight effects, and have revealed some interesting correspondences between stress and coda bearing final syllables. Testing for stress assignment in novel acronyms, Topintzi and Kainada (2012) have found that the final syllable attracts stress, especially when it also includes a consonantal coda in the rhyme. In a similar experiment regarding stress assignment in novel acronyms, Revithiadou et al. (2015) report different stressing patterns for their items depending on the morphological information they contain in the last syllable. Items ending in frequent morphological suffixes showed a tendency for penultimate stress, while ultimate stress is preferred for items ending in consonants13. What these studies have in common with Garcia (2019) is that they incorporate the study of non- words, namely novel acronyms, to the study of stress assignment for Greek, therefore, controlling for the influence of morphology14. Excluding items with morphological information in the last syllable, the findings of the experiments by Topintzi and Kainada (2012) and Revithiadou et al. (2015), have illustrated tendency of final stress in words ending in consonants, which is particularly interesting considering a growing body of evidence indicating that the penultimate syllable is the potential default of Greek stress, especially for specific noun classes (Protopapas et al., 2006; Apostolouda, 2012; Revithiadou et al., 2012, 2013; Revithiadou & Lengeris, 2016). The two main similarities between the two languages lie in the fact that morphology is a highly significant indicator of stress placement, and that the penultimate syllable is reported to be the most preferred locus of stress for nominal forms15. A third similarity arises in the preference for final stress in the presence of a, crucially, word-final coda, as reported for Greek by Topintzi and Kainada (2012), and Revithiadou et al. (2015). Even though neither of these studies makes such claims, it

13 Ultimate stress is also reported for acronyms ending in [e], however, apart from the interpretation of morphological conditioning, this finding could also be interpreted as a frequency effect, as many Greek acronyms ending in [e] are also stressed in the ultima; [otè] orɣanizmòs tilepicinoniòn elàðos „Hellenic Telecommunications Organisation‟ [oiè] orɣanizmòs inomènon eζnòn „United Nations‟ [osè] orɣanizmòs siðiroðròmon elàðos „National Railway Infrastructure‟

14 Morphology is considered to be a highly significant indicator of stress assignment in Greek (Arvaniti, 2007). This claim is also empirically confirmed by the experiments of Protopapas et al. (2006), who illustrate that even explicit stress diacritics on pseudowords is not as effective as the presence of inflectional suffixes for the assignment of stress, but also, Grimani and Protopapas (2017), who showcase that, even in pseudowords, derivational suffixes are very closely associated with fixed stress positions, overriding parameters such as the absence of a stress diacritic or even its explicit placement on an unexpected position. 15 Based on Protopapas et al. (2006), Apostolouda, (2012), Revithiadou et al. (2012, 2013), Revithiadou and Lengeris (2016) 45 could be the case that the preference for final stress in the presence of a word-final coda could be a function of the coda rendering the word-final syllable heavy, and, therefore, attracting stress, in the manner that Garcia (2017) reports for Brazilian Portuguese. The results of this study could also add to the argument of a word-final coda contributing to weight, on account of coda-bearing items appearing in the second position of a binomial, especially in the second experiment16. If a gradient weight system were to be considered a possibility in Greek, then the stress-weight interface could in turn be the motivation behind End-Weight manifestations in the language, in the manner Ryan (2019:328) describes; the heaviest constituent is attracted to the end of the prosodic phrase, which is the position of nuclear stress, i.e. the “locus of greater stress” within the prosodic phrase. Under this interpretation, the Greek language seems highly comparable to a gradient weight system, such as Brazilian Portuguese However, the argument of Greek being relevant to a weight-sensitive system, and more so a gradient weight one, cannot be definitive, especially based on the findings if this study. First, the definitive indicator of the stress-weight interface being active in Brazilian Portuguese is the attraction of stress to the heaviest syllable. In Brazilian Portuguese, heaviness is defined (among other parameters) by the presence of a coda, and is attested in the three final syllables of non-verb words. With the exception of the final coda parameter, codas in different positions within the trisyllabic window and their behavior are not explored in this study. In order to determine the pertinence of Greek to such systems, coda behavior should be studied in all positions of the trisyllabic window, not just in the ultima. Furthermore, the construction of a theory of Greek phonological weight would need to account for all effects observed in the language associated with the principle of End-Weight, which for this study would amount to the syllable count parameter, observed in both experiments, and the vowel backness effect observed in the second experiment. A weight-sensitive status for Greek could also explain the significant effects of the syllable count parameter, as more syllables, and therefore, more nuclei, are associated with greater weight within a word. On the other hand, the vowel backness effect found in the second experiment cannot really be interpreted as a result of the stress-weight interface, at least according to the analysis of Ryan (2019), who rejects vowel backness as an important parameter for weight-sensitive systems. He reaches this assumption based on Ross‟s (1982) reformulation of the original backness rule (Cooper and Ross, 1975), which recognizes backness only being significant when

16 It is important to note that for the final coda parameter, all words were non-verbs in both experiments. 46 height is held constant. In reality, the only vowel parameters discussed in Ryan (2019) is vowel height and intrinsic length, which, however, produced null results in this study. With regard to Greek particularly, Ryan (2019) cites Jespersen (1961) and reports vowel height effects in reduplicative structures, which he interprets as a function of intrinsic length (see Lehiste, 1970) which he, in turn, links to weight on account of vowel length correlating with weight in weight-sensitive systems. Another contradictory aspect pertains to the null results produced for sonority for both experiments. One limitation of this study is that the manner in which sonority was examined does not completely align with the rule of Cooper and Ross (1975), which predicts that when both words end in a single consonant, the word with the less obstruent segment is preferred in the second position. However, systems with stress- weights effects very often illustrate patterns of non-binary weight distinctions, on account of coda sonority, as more sonorous codas render a syllable heavier 17 (Gordon, 2006; Ryan, 2016), thus more likely to attract stress. As expected, Brazilian Portuguese illustrates such patterns, as coda-bearing syllables show greater tendencies for stress when their coda is sonorous (Garcia, 2017:135). If Greek were to be relevant to such a system, then it would be expected for syllables with sonorous codas to pattern as heavier, and thus appear in the second position more often than items with non- sonorous codas. Of course sonority partaking in stress-weight effects is a common occurrence in weight sensitive languages but it does not constitute the norm or a prerequisite for a language to be classified as one. A similar argument could hold for the null results produced in the final/non-stress condition of the coda parameter. This sub- parameter was expected to illustrate how final stress on coda-bearing words would make them better candidates for the second position, so that the potentially heavy final syllable can coincide with nuclear stress in the way Ryan (2019) predicts for weight- sensitive languages. Finally, an argument in favor of weight-sensitivity for Greek cannot motivate the discrepancies found in the final coda parameter based on binomial type found in the first experiment, specifically the three different combinations of native and loan words, as final codas rendering final syllables heavy should illustrate a uniform and consistent behavior. However, this effect can be considered to be a function of pragmatic and morphological interference. Garcia (2019) illustrates how parameters such as

17 Gordon (2006:127) reports that, with respect to rime, the weight hierarchy is defined as VV>VR>VO>V, with a rime type being “equivalent in weight to a rime to its left but never heavier.” He also discusses the binary distinction of light and heavy syllables against the sonority status of the coda, reporting that in languages such as Kwakw‟ala (122), only sonorant codas render a syllable heavy. 47 morphology and lexical frequencies can overrule phonological operations within the real lexicon, which again points to the unreliability of the first experiment with regard to phonological findings, due to the utilization of real components of the lexicon. Simply put, he makes a clear distinction between components such as phonology and morphological and lexical frequency. It seems that even though Greek shares a number of prosodic features with Brazilian Portuguese, a language with a unique weight-sensitivity status, the similarities are not enough to argue in favor of an analogous status for Greek, at least not within the scope of this study. Such a status could account for End-Weight manifestations in the language, as a function of phrasal end-stress as analyzed by Ryan (2019) however, this explanation fails to describe many of the findings reported for Greek binomials in the two experiments of this study. Nevertheless, the possibility of an active stress-weight interface in Greek should not be completely disregarded. This study, having not focused on exploring that aspect of the Greek system, explores only a small part of the potential components of such interface, and cannot, therefore, make definitive assertions on the matter. 4.3 Limitations The current study has a number of shortcomings in its exploration of the End-Weight principle. Word ordering is reported to mostly be a matter of semantic and cognitive parameters, while phonology concerns lower level constraints, which only manifest themselves under certain circumstances (Pordany, 1986; McDonald et al., 1993; Benor and Levy, 2006, Mollin, 2012). We could assume that Greek too is a language for which phonology operates on a lower level, since the results of this study yield such low percentages. It is important to note that the parameter which reached the highest success rate (64%), the syllable count parameter in the nonce word experiment, is also the only phonological constraint noted for word ordering in compounds by Kiparsky (2009), who highlights that it can only be active if semantic constraints are not. All other parameters found significant (namely the syllable, and the final coda parameter in the first experiment, and the vowel backness and coda parameter in the second) reached success rates between 51% and 54%. Even though these percentages are low, they produced a statistically significant result, most likely due to the high number of binomials tested. This goes to show that, with the exception of the syllable count parameter in the first experiment, the effects reported are in reality fine-grain tendencies, which are impeded by higher-ranked semantic constraints or even syntactic parameters such as the level of coordination in binomials. Another indication that 48 speakers do not rely on phonological aspects is the feedback provided by some of the participants that when they were unsure of which word order to choose, they resorted in choosing the option in which the items would be arranged in alphabetical order, a parameter also discussed in Mollin (2012) and Benor and Levy (2006), however only the latter report it as a significant indicator. A second limitation concerns the items utilized in the two experiments of this study. In the first experiment the use of real word may be considered problematic. Since real words carry semantic content, an exhaustive control of semantic and pragmatic parameters is virtually impossible. An example of this complication is the unreliable results produced in the final coda parameter of the first experiment. Another drawback in using real words in such an experiment is related to the notion of the prototypical speaker which is the base of Cooper and Ross‟s (1975) MEFIRST constraint. According to Siyanova-Chanturia et al. (2011), language can be viewed as an accumulation of experiences, which can be different for every person. Iliev and Smirnova‟s (2016) study on binomial ordering and speaker profiles challenges the notion of prototypicality and indicates that even microvariations among participants such as religion and political views can affect the order of items within a binomial. On the other hand, the use of nonce words also comes with certain disadvantages. Nonce words can eliminate any semantic factors, but, especially when targeting aspects that are not central within the native lexicon, such as word-final codas, they may appear non-native. The Greek lexicon is infused with a great deal of loanwords which end in a variety of consonantal clusters. Loanwords are considered to be part of the peripheral strata of the lexicon, as they usually lack the morphological configurations that are associated with the language (Kiparsky, 1968). Words that are part of peripheral strata of grammar can be very tolerant of structures that are otherwise not accepted (Pinta, 2013), which means that they may not accurately represent the core phonological grammar of the language (Ito, 2008, 2014). The existence of peripheral strata of phonology can potentially confound speakers in their judgments, as the distinction between native and non-native cannot be definitive. For example, this study did not produce any significant effects for vowel height, even though Jespersen (1961) reports height-low contrast in Greek ablaut pairs. These high-low ablaut structures, although fairly common (i.e. tik-tak), are not nearly as frequent in the language as the [a-u] pattern (i.e. 'pafa-'pufa), a pattern showcasing a low-high sequence. This low-high structure, however, is quite comparable to [a-u] ablaut structures reported for Turkish by Marchand (1952). It is well known that the Greek lexicon contains a fair number of Turkish loans, and such ablaut structures could 49 have infiltrated the lexicon as well as loaned word-play patterns. Their frequency in casual speech might override or distort any relevant vocalic alternation rules of the Greek phonological grammar. In this case, it would be very unclear whether speakers are conscious that an [a-u] alternation pattern is a loan. Another limitation is that the principle of End-Weight is a phenomenon, the phonological motivations of which have been mainly discussed for weight-sensitive languages, not so for ones that illustrate gradient weight patterns. Ryan (2019) considers it to be a function of the stress-weight interface usually associated with weight-sensitive languages. Since Greek is not considered to be a weight-sensitive system, an attempted interpretation of the results of this study under a similar framework might be a far- fetched venture. However, Ryan (2019) notes that End-Weight exhibits variations among languages which observe it. For example, Pinker and Birdsong (1979) have found an effect which they interpret as a universal tendency regarding vowels, but, English and French are two languages that could be considered similar and also closely related on account of their Indo-European ancestry. The same argument could also hold for Oakeshott-Taylor (1984) and his tests of English, German, and Afrikaans, as the vowel inventories of these languages are very similar. Furthermore, languages such as Turkish, Bengali, and Tamil are regarded as having left oriented phrasal stress (at least in the p-phrase) and expected to behave in an opposite manner for phenomena similar to End-Weight (as in the [a-u] patterns in Turkish) (Ryan, 2019), this being the reason why such principles cannot be considered universal, but rather as manifesting in distinct ways among languages. These languages are reported to have a prosodic head on the left side, and thus possibly exhibit Beginning-Weight effects (Ryan, 2019). Consequently, the great variation in the systems that exhibit End-Weight patterns means that maybe some aspects of its manifestations have not been considered or studied yet, and that languages which lack categorical weight can also illustrate such patterns. A final limitation of this study is its failure to also control and account for metrical aspects. While the second experiments controls for word accent, the first one does not; metrical aspects such as avoidance of lapse, clash, and final stress have been reported to be highly significant (Bolinger, 1962; Benor and Levy, 2006; Mollin; 2012; Lohmann, 2013; Ryan, 2019), this study does not specifically target them. This omission is due to the fact that rhythm is not a particularly significant indicator in phonological operations in Greek (Arvaniti, 1994), but also due to the fact that rhythmic effects were not prominent in the data that constituted the basis for the hypothesis of

50 this study. However, their significance is not dismissed, rather undetermined, so they could constitute a potential ground for further study.

51

CHAPTER 5 CONCLUSION This study experimentally explores manifestations of the phenomenon of End-Weight in Greek binomial structures. The two experiments of this study focus on three phonological parameters; vowel quality, syllable count, and coda status. The first experiment produced significant results pertaining to the syllable count, and the final coda parameter. However, a closer examination of the findings of the final coda parameter revealed certain asymmetries, on account of discrepancies in the etymological origin of the items. The second experiment utilized nonce words, seeking to completely eliminate such factors. It produced significant results in the parameters of vowel backness, syllable count, and final coda. The interpretation of these results was attempted to be made through two prevailing theories: the theories of accessibility and phrasal stress. This study has illustrated that the manifestations of End-Weight effects in Greek binomials cannot be completely described through these accounts. While both approaches can successfully explain particular aspects of the findings reported in this study, certain aspects cannot be described. While an interpretation of frequency and accessibility can successfully explain why a shorter word is more likely to appear first, it fails to account for the findings regarding the parameter of vowel backness in the second experiment, as well as, the greater part of the results of the coda parameter. On the other hand, an attempt to explain the findings under Ryan‟s (2019) interpretation can also account for syllable and final coda effects, but also falls short in accounting for effects of vowel backness found in the second experiment. It seems that the results of this study cannot be interpreted based on current means as both theories lack basic foundations in Greek. The frequency and markedness account is not reliable due to the irresolution regarding the status of codas in the language, as well as the lack of a comprehensive theory of lexical strata, that could determine what aspects and structures lie in the core or the periphery of the grammar. As for the second theory, its relevance to the manifestations of End-Weight effects in Greek binomials would entail that Greek is a weight-sensitive system. While studies such as Topintzi and Kainada (2012) and Revithiadou et al. (2015) report tendencies that seem quite comparable to effects observed in weight- sensitive systems, especially one similar to Brazilian Portuguese, an argument in favor of a weight-sensitive system in Greek remains far-fetched. Of course, we cannot eliminate the possibility that the findings of this study are a function of a number of

52 parameters, or even aspects that have not been considered or controlled for in this study18. Even though this piece of research has kept a restricted scope, it contributes to the study of word ordering and coordination. The study of binomials has mainly been relevant in weight-sensitive systems; the study of End-Weight manifestations in a language which is generally believed to be weight-insensitive can initiate a discussion about End-Weight being relevant to such systems, or even illustrate that weight, along with other phonological aspects, can be trumped by factors such as semantics and morphology. Furthermore, this study also contributes to the study of the prosodic right edge, which generally seems to be the locus of great interest. Apart from its status as prominent to perception (Shih and Zuraw, 2017), the right edge has also been significant in the study of poetic meters Ryan (2013) considers it to be a strict position which does not allow much flexibility, an effect also found in the Greek traditional poetic meter, dekapentasyllavo (Topintzi and Versace, 2015) In light of the results pertaining to the final coda parameter, studies on lexical strata and loanword phonology in Greek could reveal new phonological tendencies, as well as degrees of phonological tolerance in the language. Relating to the results of the syllable count parameter in the second experiment, further studies on word ordering can also focus on other metrical aspects, such as avoidance of lapse, as, while Greek is reported not to repair lapses phonetically (Arvaniti, 1992, 1994), potential lapse amelioration through constituent reordering is not established in Greek. Secondly, an alternative study on binomials could utilize data from corpora so as to focus on the patterns of novel binomials in free speech, as opposed to the forced choice experiments of this study. As this study has only focused on binomial patterns, End-Weight effects in other domains remain undetermined for Greek. Furthermore, in order to further examine the relevance of Greek to a weight sensitive system, and whether a coda can render syllables heavy, future pieces of research should consider testing with stress assignment tasks while controlling for morphological information. Finally, a corpus study utilizing Bayesian statistics, such as the one Garcia (2017) carried out for Brazilian Portuguese, could bring forward more robust tendencies of Greek stress behavior.

18 For example, even though order of appearance was randomized, the software of Google Forms does not allow for a view of the orders generated, meaning that we cannot eliminate the possibility of presentation being a significant parameter or affecting participant preferences. 53

REFERENCES Alexiadou, A. 2002. Word order patterns in Greek nominals: aspects of diachronic change. ZASPI 27: 91-108. Allan, Keith. 1987. Hierarchies and the choice of left conjuncts (with particular attention to English). Journal of Linguistics 23: 51–77. Anttila, A. 2008. Phonological constraints on constituent ordering. In Proceedings of the 26th west coast conference on formal linguistics (pp. 51-59). Somerville, MA: Cascadilla Press. Anttila, A., Adams, M., and Speriosu, M. 2010. The role of in the English dative alternation. Language and Cognitive Processes 25(7–9): 946–981. Apostolouda, V. 2012. Ο ηνληζκόο ησλ νπζηαζηηθώλ ηεο Διιεληθήο: κηα πεηξακαηηθή πξνζέγγηζε [Nominal stress in Greek: An experimental approach]. ΜΑ Dissertation, Aristotle University of Thessaloniki. Apoussidou, D. 2003. Irregular accent patterns in Correspondence Theory. Selected papers on theoretical and applied linguistics 15: 228-238. Arvaniti, A. (1992). Secondary stress: evidence from Modern Greek. In G. J. Docherty and D. R. Ladd (eds.), Papers in Laboratory Phonology II: Gesture, Segment, Prosody, pp 398-423. Cambridge: Cambridge University Press. Arvaniti, A. 1994. Acoustic features of Greek rhythmic structure. Journal of phonetics 22(3): 239-268. Arvaniti, A. 1999. Standard Modern Greek. Journal of the International Phonetic Association 29(2): 167-172. Arvaniti, A. 2000. The phonetics of stress in Greek. Journal of Greek Linguistics 1(1): 9-39. Arvaniti, A. 2007. Greek phonetics: The state of the art. Journal of Greek linguistics 8(1): 97-208. Arvaniti, A., and Baltazani, M. 2005. Intonational analysis and prosodic annotation of Greek spoken corpora. Prosodic typology: The phonology of intonation and phrasing 84: 117-169. Ayoun, D. 2000. Web-based elicitation tasks in SLA research. Language Learning & Technology 3(2): 77-98. Battistella, E. L. 1990. Markedness: The evaluative superstructure of language. SUNY Press. Benor, S. and Levy, R. 2006. The chicken or the egg? A probabilistic analysis of English binomials. Language 82(2): 233–278. 54

Bolinger, D. L. 1962. Binomials and pitch accent. Lingua 11: 34–44. Chomsky, N. and Halle, M. 1968. The sound pattern of English. New York: Harper and Row. Clements, G. N. 1990. The role of the sonority cycle in core syllabification, in J. Kingston & M. E. Beckman (eds.), Papers in Laboratory Phonology I, pp. 283- 333. Cambridge: Cambridge University Press. Cooper, William E., and John R. Ross. 1975. World order. In R. Grossman, L. J. San, and T. Vance (eds.), Papers from the parasession on functionalism, pp. 63–111. Chicago: Chicago Linguistic Society. Cutler, A., and Cooper, W. E. (1978). Phoneme-monitoring in the context of different phonetic sequences. Journal of Phonetics, 6: 221–225. Drachman, G., Kager, R. W. J., and Malikouti-Drachman, A. 1997. Greek allomorphy: an optimality account. Greek Linguistics 95(1): 151-160. Drachman, G. and Malikouti-Drachman, A. 1999. Greek word accent. In H. van der Hulst (ed.), Eurotyp, 4, Word Prosodic Systems in the Languages of Europe, pp. 897-946. Berlin, Boston: De Gruyter. Elenbaas, N., & Kager, R. 1999. Ternary rhythm and the lapse constraint. Phonology 1(3): 273-329. Fenk-Oczlon, Gertraud. 1989. Word frequency and word order in freezes. Linguistics 27(3): 517–556. Fourakis, M., Botinis, A., & Katsaiti, M. 1999. Acoustic characteristics of Greek vowels. Phonetica 56(1-2): 28-43. Garcia, G. D. (2017). Weight gradience and stress in Portuguese. Phonology 34(1): 41- 79. Garcia, G. D. 2019. When lexical statistics and the grammar conflict: Learning and repairing weight effects on stress. Language 95(4): 612-641. Goad, H. 2012. sC clusters are (almost always) coda-initial. The Linguistic Review 29(3): 335-373. Goad, H. 2016. Phonotactic evidence from typology and acquisition for a coda+onset analysis of initial sC clusters. In K. Kim, P. Umbal, T. Block, Q. Chan, T. Cheng, K. Finney, M. Katz, S. Nickel‐Thompson, & L. Shorten (eds.), Proceedings of the 33rd West Coast Conference on Formal Linguistics pp. 17–28. Somerville, MA: Cascadilla Proceedings Project. Gordon, M. 2006. Syllable weight: Phonetics, phonology, typology. New York: Routledge Press. 55

Gratsouni, D., & Topintzi, N. to appear. On the Syllabification of ON (Obstruent-Nasal) Clusters in Greek. Proceedings of the 40th Meeting of the Department of Linguistics (AMGL40). Thessaloniki: Aristotle University of Thessaloniki. Grimani, A., & Protopapas, A. 2017. Derivational suffixes as cues to stress position in reading Greek. Journal of Research in Reading 40: S23-S41. Gustafsson, M. 1975. Binomial expressions in present-day English: A syntactic and semantic study. Turku: Turun Yliopisto. Gustafsson, M. 1976. The frequency and „frozenness‟ of some English binomials. Neuphilologische Mitteilungen 77: 623–37. Hatzigeorgiu, N., Gavrilidou, M., Piperidis, S., Carayannis, G., Papakostopoulou, A., Spiliotopoulou, A., Vacalopoulou, A., Labropoulou, P., Mantzari E., Papageorgiou, H., & Demiros, I. 2000. Design and implementation of the online ILSP Greek Corpus. In M. Gavrilidou et al. (eds.), Proceedings of the LREC 2000 Conference pp. 1737 – 1742. Athens. Hegarty, P. 2014 Ladies and gentlemen: Word order and gender in English. In Corbett, G. (Ed.) The Expression of Gender. Berlin, Boston: De Gruyter Mouton. 69 – 86. Hegarty, P., Watson, N., Fletcher, K., and McQueen, G. 2011. When gentlemen are first and ladies are last. Effects of gender stereotypes on the order of romantic partners‟ names. British Journal of Social Psychology 50: 21–35. Holton, D., Mackridge, P., and Philippaki-Warburton, I. 1997. Greek: A comprehensive grammar of the modern language. Routledge. Horrocks, G. 1983. The Order of Constituents in Modern Greek. In G. Gazdar, E. Klein & G. Pullum (eds.), Order, Concord and Constituency Pp. 95-111. Dordrecht: Foris. Iliev, R., and Smirnova, A. 2016. Revealing word order: Using serial position in binomials to predict properties of the speaker. Journal of psycholinguistic research 45(2): 205-235. Ito, C. 2008. Analogical changes in the accent of Sino-Korean words in Yanbian Korean. In Proceedings of the 27th west coast conference on formal linguistics, pp. 238-246. Cascadilla Proceedings Project Somerville, MA. Ito, C. 2014. Compound tensification and laryngeal co-occurrence restrictions in Yanbian Korean. Phonology 31(3): 349-398. Jespersen, O. 1961. A modern English grammar on historical principles. Part 6, Morphology. London: George Allen and Unwin Ltd.

56

Joseph, Brian D. and Tserdanelis, G. 2003. Modern Greek. In Roelcke, Thorsten (ed.), Variationstypologie / Variation Typology. Ein sprachtypologisches Handbuch der europäischen Sprachen in Geschichte und Gegenwart / A Typological Handbook of European Languages, pp. 823-836. Berlin, New York: Walter de Gruyter. Kang, Y. 2011. Loanword Phonology. In M. van Oostendorp, C. Ewen, E. Hume and K. Rice (eds.), Companion to Phonology, pp 2258-2282. Malden, MA: Wiley Blackwell. Kappa, I. 2002. ΢πιιαβηθή δνκή [Syllabic structure]. Ms. https://student.cc.uoc.gr/uploadFiles/177%CE%93%CE%9B%CE%A9%CE%A6 153/3%CE%A3%CE%A5%CE%9B%CE%9B%CE%91%CE%92%CE%99%CE %9A%CE%97%20%CE%94%CE%9F%CE%9C%CE%97.pdf Kelly, M. H. 1986. On the selection of linguistic options. PhD dissertation, Cornell University. Kelly, M. H., Springer, K., and Keil, F. C. 1990. The relation between syllable number and visual complexity in the acquisition of word meanings. Memory & cognition 18(5): 528-536. Kiparsky, P. 1968. Linguistic universals and language change. In E. Bach & R. Harms (eds.), Universals in Linguistic Theory, pp. 170-202. Holt, Rinehart and Winston, New York. Kiparsky, Paul. 2009. Verbal co-compounds and subcompounds in Greek. MIT Working Papers in Linguistics 57. Lascaratou, C., and Georgiafentis, M. 2006. Βαζηθά ραξαθηεξηζηηθά ηεο ζεηξάο ησλ πξνηαζηαθώλ όξσλ ζηελ ειιεληθή θαη ζηελ ηνπξθηθή [Main properties of the word order in Greek and Turkish]. In S. Moschonas (ed.) Η ΢ύληαμε ζηε Μάζεζε θαη ζηε Δηδαζθαιία ηεο Ειιεληθήο ωο Ξέλεο Γιώζζαο [The syntax in the learning and teaching of Greek as a foreign language] pp. 11-61. Athens: Patakis. Lehiste, I. 1970. Suprasegmentals. Cambridge: MIT Press. Lohmann, A. 2013. Constituent order in coordinate constructions: a processing perspective. PhD dissertation, University of Hamburg. Lohmann, A., and Takada, T. 2014. Order in NP conjuncts in spoken English and Japanese. Lingua 152: 48–64. Malikouti-Drachman, A. 1984. Syllables in Modern Greek. In W. U. Dressler, H. Luschützky, O. Pfeiffer and J. Rennison (eds.) Phonologica 1984, pp. 181–187. London: Cambridge University Press.

57

Malikouti-Drachman, Angeliki. 1994. New approaches to some problems of Greek Phonology. In I. Philippaki-Warburton, K. Nikolaidis and M. Sifianou, Themes in Greek Linguistics, pp. 33–44. Amsterdam & Philadelphia: Benjamins. Malikouti-Drachman, A. 2002. Greek phonology: A contemporary perspective. Journal of Greek Linguistics 2(1): 187-243. Malkiel, Y. 1959. Studies in irreversible binomials. Lingua 8: 113–160. Marchand, H. 1952. Alliteration, Ablaut und Reim in den Türkischen Zwillingsformen. Oriens 5: 60–69. McDonald, J. L., Bock, K. and Kelly, M. H. 1993.Word and world order: Semantic, phonological, and metrical determinants of serial position. Cognitive Psychology 25: 188–230. Minkova, D. 2002. Ablaut reduplication in English: The criss-crossing of prosody and verbal art. English Language and Linguistics 6: 133–169. Mollin, S. 2012. Revisiting binomial order in English: Ordering constraints and reversibility. English Language and Linguistics 16(1): 81–103. Morgan, E., and Roger Levy. 2016. Abstract knowledge versus direct experience in processing of binomial expressions. Cognition 157: 384–402. Müller, G. 1997. Beschrankungen für Binomialbildungen im Deutschen. Zeitschrift für Sprachwissenschaft 16(1/2): 5–51. Nespor, M., and Ralli, A. 1994. Stress domains in Greek compounds: A case of morphology-phonology interaction. Amsterdam Studies in the Theory and History of Linguistic Science Series 4, 201-208. Nicolaidis, K. 2003. Acoustic variability of vowels in Greek spontaneous speech. In Proceedings of the 15th international congress of phonetic sciences, pp. 3221- 3224. Nicolaidis, K., Edwards, J., Beckman, M., and Tserdanelis, G. 2003. Acquisition of lingual obstruents in Greek. In Proceedings of the 6th International Conference of Greek Linguistics, Rethymno, Crete. Oakeshott-Taylor, J. 1984. Phonetic factors in word order. Phonetica 41: 226–237. Oden, G. C., and Lopes, L. L. 1981. Preference for order in freezes. Linguistic Inquiry 12: 673–679. Ourn, N., & Haiman, J. 2000. Symmetrical compounds in Khmer. Studies in Language. International Journal sponsored by the Foundation “Foundations of Language” 24(3): 483-514.

58

Parker, S. 2002. Quantifying the sonority hierarchy. PhD dissertation, University of Massachusetts, Amherst. Parker, S. 2003. The psychological reality of sonority in English. Word 54(3): 359-399. Phillips, B. S. 1984. Word frequency and the actuation of sound change. Language 60: 320-342. Pinker, S., and Birdsong, D. 1979. Speakers‟ sensitivity to rules of frozen word order. Journal of Verbal Learning and Verbal Behavior 18(4): 497–508. Pinta, J. D. 2013. Lexical strata in loanword phonology: Spanish loans in Guarani. MA dissertation, University of North Carolina at Chapel Hill. Pordany, L. 1986. A comparison of some English and Hungarian freezes. Papers and Studies in Contrastive Linguistics 21: 117–125. Porpodas, C. D. 2006) Literacy acquisition in Greek: Research review of the role of phonological and cognitive factors. In R. M. Joshi and P. G. Aaron (Eds.), Handbook of orthography and literacy, pp. 189-199. Mahwah, NJ: Erlbaum. Prince, A. 1990. Quantitative consequences of rhythmic organization. Chicago Linguistic Society 26(2): 355–98. Prince, A., and Smolensky, P. 2004. Optimality theory: Constraint interaction in generative grammar. Malden: Blackwell. Protopapas, A. 2006. On the use and usefulness of stress diacritics in reading Greek. Reading and Writing 19(2): 171-198. Protopapas, A., Gerakaki, S., and Alexandri, S. 2006. Lexical and default stress assignment in reading Greek. Journal of Research in Reading 29(4): 418-432. Protopapas, A., and Vlahou, E. L. 2009. A comparative quantitative analysis of Greek orthographic transparency. Behavior research methods, 41(4): 991-1008. Quirk, R., Greenbaum, S., Leech, G., and Svartvik, J. 1972. A grammar of contemporary English. London: Longman. Ralli, A. 2018. Morphological theory and synchronic variation. In J. Audring & F. Masini (eds.), The Oxford Handbook of Morphological Theory. Oxford University Press. Revithiadou, A. 1999. Headmost accent wins: Head dominance and ideal prosodic form in lexical accent systems. PhD dissertation, Leiden University. Revithiadou, A. 2004. The iambic/trochaic law revisited. Leiden papers in Linguistics 1: 37-62.

59

Revithiadou, A. 2005. Prosodic phrasing and focus in Greek declaratives. In Proceedings of the 6th International Conference of Greek Linguistics, pp. 64-74. Rhethymnon: University of Crete. Revithiadou, A. 2007. Colored turbid accents and containment: A case study from lexical stress. Freedom of analysis 149: 174. Revithiadou, A., & Lengeris, A. 2016. One or Many? In Search of the Default Stress in Greek. In J. Heinz, R. Goedemans, and H. van der Hulst (Eds.), Dimensions of Phonological Stress, pp. 263–290. Cambridge: Cambridge University Press. Revithiadou, A., Lengeris, A. and Ioannou, D. 2012. Is there Default Stress in Greek? Paper presented at the ud Conference on Stress and Accent, 29 November–1 December 2012, University of Delaware. Revithiadou, A., Lengeris, A., and Ioannou, D. 2013. In Search of the Default Stress in Greek: Evidence from Perception. Paper presented at the 10th Old World Conference in Phonology (ocp10), 16–19 January 2013, Istanbul: Boğaziçi University. Revithiadou, A., Nikolou, K., and Papadopoulou, D. 2015. Stress in the absence of morphological conditioning: An experimental investigation of stress in Greek acronyms. Journal of Greek Linguistics 15(2): 187-234. Ross, J. R. 1982. The sound of meaning. In The Linguistic Society of Korea (ed.) Linguistics in the morning calm, pp. 275–290. Seoul: Hanshin Publishing Co. Ryan, K. M. 2013. Against final indifference. Talk presented at M@90 Workshop on Stress and Meter, MIT, 20–21 September 2013. Ryan, K. M. 2014. Onsets contribute to syllable weight: Statistical evidence from stress and meter. Language 90(2): 309–341. Ryan, K. M. 2016. Phonological weight. Language and Linguistics Compass 10: 720– 733. Ryan, K. M. 2019. Prosodic end-weight reflects phrasal stress. Natural Language & Linguistic Theory 37(1): 315-356. Selkirk, E. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, E. 1995. Sentence prosody: Intonation, stress, and phrasing. In John A. Goldsmith (ed.) The handbook of phonological theory, pp. 550–569. Oxford: Blackwell Publishing.

60

Selkirk, E. 2011. The syntax-phonology interface. In J. Goldsmith, J. Riggle and A. C. L. Yu (eds.) The handbook of phonological theory 2nd edition, pp. 435-484. Oxford: Wiley-Blackwell. Setatos, M. 1974. Φωλνινγία ηεο Κνηλήο Νενειιεληθήο [Phonology of Standard Modern Greek]. Athens: Papazisis. Setatos, M. 1987. Από ηα Αξραία ζηα Νέα Διιεληθά: Φσλνινγηθέο αιιαγέο [From Ancient Greek to Modern Greek: Phonological changes]. Studies in Greek Linguistics 8. 187–193. Shih, S., and Zuraw, K. 2017. Phonological conditions on variable adjective and noun word order in Tagalog. Language 93: e317 - e352. Siyanova-Chanturia, A., Conklin, K., & Van Heuven, W.J.B. 2011. Seeing a phrase "time and again" matters: the role of phrasal frequency in the processing of multiword sequences. Journal of Experimental Psychology: Learning, Memory and Cognition 37(3): 776-784. Steriade, D. 1982. Greek Prosodies and the nature of Syllabification. PhD Dissertation, MIT. Swanson, D. C. 1958. English loanwords in modern Greek. Word 14(1): 26-46. Thun, Nils. 1963. Reduplicative words in English. Lund: Carl Bloms Boktryckeri. Topintzi, N., and Baltazani, M. 2012. The acoustics of high-vowel loss in a Northern Greek dialect and typological implications. In P. Hoole, L. Bombien, M. Pouplier, Ch. Mooshammer, and B. Kühnert (eds.), Consonant Clusters and Structural Complexity, pp. 373–402. Interface Explorations series, de Gruyter. Topintzi, N., and Kainada, E. 2012. Acronyms and the placement of default stress in Greek. In 10th International Conference on Greek Linguistics (ICGL10), pp. 472- 479. Komitini: Democritus University of Thrace. Topintzi, N., and Versace, S. 2015. A linguistic analysis of the modern Greek dekapentasyllavo meter. Journal of Greek Linguistics 15(2): 235-269. Touratzidis, L. and Ralli, A. 1992. A Computational Treatment of Stress in Greek Inflected Forms. Language and Speech 35(4): 435–453. van Oostendorp, M. 2012. Stress as a proclitic in Modern Greek. Lingua 122(11): 1165- 1181. Wasow, T. 1997. End-Weight from the Speaker's Perspective. J Psycholinguist Res. 26: 347-361. Wescott, R. W. 1970. Types of vowel alternations in English. Word 26: 309–343.

61

Wright, S., and Hay, J. (2002). Fred and Wilma: A phonological conspiracy. In S. Benor, M. Rose, D. Sharma, J. Sweetland, and Q. Zhang, Gender and linguistic practice, pp. 175-191. Stanford, CA: CSLI Publications. Wright, Saundra, Jennifer Hay, and Tessa Bent. 2005. Ladies first? Phonology, frequency, and the naming conspiracy. Linguistics 44: 531–561. Zvonik E and Cummins F. 2003. The effect of surrounding phrase lengths on pause duration. Proceedings of Eurospeech 2003, pp. 777–780. Geneva, Switzerland.

62

APPENDICES Appendix A: Materials

Experiment 1: Real word targets

1. Πξόθεηηαη γηα έλα λέν ζακπνπάλ κε εθρπιίζκαηα ηνπιίπαο θαη βηνιέηαο. It is about a new shampoo with extracts of tulip and violet.

2. Σν κπζηηθό γηα ηα ηέιεηα θνπινπξάθηα είλαη λα πξνζζέζεηο θαλέια θαη καζηίρα. The secret for the perfect cookies is to add cinnamon and mastic.

3. Ζ ζηνιή ηεο θηινμελνύκελεο νκάδαο ήηαλ θίηξηλε θαη πξάζηλε. The kit of the away team was yellow and green.

4. Ο ζεθ επίζεο πξνηείλεη κηα ρεηκσληάηηθε ζαιάηα κε αληίδηα θαη ξεπάληα. The chef also suggests a winter salad with chicories and radishes.

5. Σν θέηθ ηεο θπξά-Μαξίαο είρε γαξληηνύξεο από θαξύδη θαη ιεκόλη. Miss Maria‟s cake had garnishes of walnut and lemon.

6. Θα πεηαρηώ κέρξη ηελ ιατθή γηα λα αγνξάζσ θξεκκύδηα θαη θαζόιηα. I will pop by the farmer‟s market to buy some onions and beans.

7. Σα γιπθά γηα ην ζρνιηθό πάξηη δελ πξέπεη λα πεξηέρνπλ μεξνύο θαξπνύο, όπσο θηζηίθη θαη θνπληνύθη. The sweets for the school party are not to contain any dry nuts such as peanut and hazelnut.

8. Έπεζε ζηελ πνιπζξόλα κηζόο άλζξσπνο, ζαλ ην ζώκα ηνπ λα ήηαλ ηξύπην θαη θνύθην. He fell into the armchair half a person, as if his body were full of holes and hollow.

9. Πάλε κέρξη ην ζνππεξκάξθεη θαη πάξε θαζέξη θαη ζαιάκη. Go to the supermarket and get some cheese and ham.

10. Σα έρσ όια ζην θαξόηζη, αιιά κνπ ιείπνπλ αθόκα αιεύξη θαη αιάηη. I have everything in the cart, but I still need flour and salt.

11. Σν παηδηθό δσκάηην ήηαλ βακκέλν κπεδ θαη κωβ. The children‟s room was painted beige and purple.

12. Σηο ληνκάηεο ηηο πνπιάεη κόλν αλ είλαη θξέζθεο θαη ληόπηεο. He sells the tomatoes only if they are fresh and local.

13. Αλππνκνλώ γηα ην Πάζρα, κόλν θαη κόλν γηα λα θάλεη ε κακά ηζνπξέθηα θαη θνπινύξηα. I can‟t wait for Easter, only because mum will make Easter bread and cookies. 63

14. Γε γίλεηαη εθθιεζία ρσξίο ηέκπιν θαη ηξνύιν. There is no church without an altar and dome.

15. Γελ πξόθεηηαη λα θάσ πξάζηλα όπσο καξνύιηα θαη ζπαλάθηα. I am not going to eat greens such as lettuce and spinach.

16. Γηώξγν, έηνηκν ην ηξαπέδη! Όπσο έξρεζαη θέξε πηξνύληα θαη θνπηάιηα. George, the table is set! On your way, bring forks and spoons.

17. Ξέξεηο όηη ην θαινθαίξη ηξώσ πνιύ θαξπνύδη θαη πεπόλη. You know that in summer, I eat lots of watermelon and honeydew.

18. Σν Μελέιαν ηνλ μέξαλε ζην ρσξηό σο άληξα ληόκπξν θαη ληνύξν. Everybody in the village knew Menelaos as honest and harsh.

19. Έπηλε πνιύ ηζάη θαη ε απιή ηεο ήηαλ γεκάηε ζάκλνπο, θπξίσο ιεβάληεο θαη κνιόρεο. She drank lots of herbal tea and her garden was full of bushes, mostly lavender and mallow.

20. Γηα λα απνθύγεηε δαγθώκαηα θηδηώλ, θνξάηε θάιηζεο θαη κπόηεο. In order to avoid snake bites, wear socks and boots.

21. Γελ ήμεξα ηη θηιέην θνηόπνπιν πξνηηκάο θαη ζνπ έθαλα δύν εηδώλ, κπνύηη θαη ζηήζνο. I did not know what chicken fillet you prefer so I made you two kinds, thigh and breast.

22. Ο Νίθνο πήγε ην απηνθίλεην γηα ιάδηα θαη ζέξβηο. Nikos took the car for an oil (change) and service.

23. Γύν πξάγκαηα πξέπεη λα ελδηαθέξνπλ θάζε άληξα: θήκε θαη ηζρύο. Every man should care about two things: reputation and clout.

24. Όηη θαη αλ ηελ ξσηνύζα, απαληνύζε «΢ίγνπξα» θαη «Αζθαιώο». Whatever I asked her, she responded with “Surely” and “Certainly.”

25. Σα ΢αββαηνθύξηαθα πάκε ζηελ ιέζρε θαη παίδνπκε κπηιηάξδν θαη κπάληκηληνλ. At weekends, we go to the club and play billiard and badminton.

26. Γηα νξεθηηθό θάγακε θάηη παηάηεο γεκηζηέο κε αβγό θαη κπέηθνλ. For starters we had some potatoes, stuffed with egg and bacon.

27. ΢ήκεξα ην θαηάζηεκά καο ζαο πξνζθέξεη -20% ζε παιηό θαη κπνπθάλ. Today, our store offers -20% in coats and jackets.

64

28. Φνξνύζε έλα παιηνκνδίηηθν θόξεκα από ηνύιη θαη ζαηέλ. She wore an old-fashioned dress made of tulle and satin.

29. Γελ πξόθεηηαη λα ηνπο θέξεηο κε ην κέξνο ζνπ ρσξίο θάπνην δόιωκα θαη δέιεαξ. You are not going to bring him to your side without some kind of bait and lure.

30. Δπηπρώο ε γηαηξόο δε δηέθξηλε θαξθηληθά θύηηαξα ζε λεθξά θαη ήπαξ. Fortunately, the doctor did not any cancer cells in the kidneys and liver.

31. Ζ Λίδα ηξειαίλεηαη γηα ηαηλίεο κε δόκπη θαη βακπίξ. Lisa loves movies with zombies and vampires.

32. Όηαλ έθπγε δελ ζπκάκαη λα θνξνύζε θάπνην θόζκεκα θαη αμεζνπάξ. When she left, I don‟t remember her wearing jewelry and accessory.

33. Οη Δπξσπαίνη δελ πξνηηκνύλ θάπνηα ακεξηθάληθα αζιήκαηα, όπσο ξάγθκπη θαη κπέηδκπνι. Europeans do not like some American sports, such as rugby and baseball.

34. Ζ εζληθή ζθελή ζα έρεη δηάθνξεο παξαζηάζεηο θέηνο, αιιά ζίγνπξα όρη κπαιέην θαη κηνύδηθαι. The National Theater will have a variety of performances this year, but surely not ballet and musicals.

35. Σα δώα ήηαλ θξπκκέλα ζε ηξηγύξσ βξάρηα θαη ηνύλει. The animals were hidden in surrounding rocks and tunnels.

36. Απηνύ ηνπ είδνπο ε θνύζηα ήηαλ πηα παζέ θαη κπαλάι. That kind of skirt was outmoded and unoriginal.

37. Σελ Παξαζθεπή ην βξάδπ ζα κεηαδνζεί δσληαλά ν αγώλαο αλάκεζα ζε Μπελθίθα θαη Σόηελακ. This Friday night, there will be a live broadcast of the match between Benfica and Tottenham.

38. Κάζε δσκάηην ηνπ μελνδνρείνπ καο ζπκπεξηιακβάλεη ινπηξό θαη ρακάκ. Every room of our hotel includes a bathroom and a Turkish bath.

39. Γελ ζα ζε λνηάδεη ηη ζρνιηάδεη θάζε θπξά θαη καληάκ. You should not care about the comments of some old-lady and madam.

40. Γηα ηα Χξηζηνύγελλα, έθιεηζε λα πάεη Ειζίλθη θαη Άκζηεξληακ. For Christmas, he booked a trip to Helsinki and Amsterdam.

41. ΢ηνλ ειεύζεξό ηνπ ρξόλν αζρνιείηαη κε αζπλήζηζηα αζιήκαηα, όπσο ηδνύλην θαη θξίθεη. He spends his free time doing unusual sports such as judo and cricket.

65

42. Ο δεκνζηνγξάθνο πξνζπάζεζε λα δηαρσξίζεη ηνπο πνιίηεο ζε κάδα θαη ειίη. The reporter tried to divide citizens between the masses and the elite.

43. Οη γνλείο ηνπ Πέηξνπ δελ ηνλ αθήλνπλ θαζόινπ λα ρξεζηκνπνηεί θηλεηό θαη ίληεξλεη. Petros‟ parents do not allow him to use a cellphone and the internet.

44. Ο δσνινγηθόο θήπνο ηεο Βνζηώλεο θηινμελεί θαη δηάθνξα αξπαθηηθά, όπσο πνύκα θαη θνγηόη. The zoo in Boston also houses various vultures such as pumas and coyotes.

45. Οη θεηηλέο ηζάληεο ηνπ ρεηκώλα είλαη από πόλπ θαη ζνπέλη. This winter‟s handbags are made of pony leather and suede.

46. Ννκίδσ δελ ππάξρεη ρεηξόηεξνο ζπλδπαζκόο ρηελίζκαηνο από αγνξέ θαη πεξκαλάλη. I think that there is not a worse combination for a hairstyle than pixie cut and permed.

47. Ο Γηώξγνο μέξεη πνιιά γηα ό,ηη έρεη λα θάλεη κε ζηλεκά θαη Χόιηγνπλη. George knows everything about cinema and Hollywood.

48. ΢ήκεξα ζα κηιήζνπκε γηα ηηο απώιεηεο ησλ Γεξκαληθώλ δπλάκεσλ ζε δύν ζεκαληηθέο κάρεο: Ρίκηλη θαη ΢ηάιηλγθξαλη. Today we will talk about the casualties of the German forces in two significant battles: Rimini and Stalingrad.

49. Δδώ ζηελ ζηάλε θαηνηθνύλ αξληά θαη θαηζίθηα. In this shed there are lambs and goats.

50. Σν ηξαπεδνκάληειν ήηαλ θαξό καύξν θαη θόθθηλν. The table cloth was black and red tartan

51. Αλ ζέιεηε λα έρεηε βακκέλα καιιηά, ε πνιηηηθή ηεο εηαηξίαο επηηξέπεη κόλν μαλζά θαη θαζηαλά. If you wish to have dyed hair, our company‟s policy only allows blonde or brown.

52. Γελ έρσ μαλαθνύζεη ρξώκαηα πνπ λα ηα ιέλε πεηξόι θαη βεξακάλ. I have never heard of colors named teal and mint green.

53. Οη ήξσεο ηεο ηζηνξίαο είλαη δσάθηα ηεο εμνρήο, όπσο πάπηεο θαη ρειώλεο. The characters of the story are countryside animals, such as ducks and turtles.

54. Ζ ζπληαγή γηα καγεηξίηζα πνπ κνπ έδσζε ε γηαγηά κνπ ζέιεη πξάζν θαη ζέιηλν. The magiritsa recipe I was given to by grandma requires leek and celery.

66

55. Από ηελ αγνξά κπαραξηθώλ, ε Μαίξε δήηεζε λα ηεο πάξνπκε θάξπ θαη θύκηλν. From the spice market, Mary asked us to get curry and cumin.

56. Πώο θαηάθεξε λα πείζεη ηνλ Αιέμε γηα ην πάξηη απόςε είλαη ζαύκα θαη κπζηήξην. Her managing to convince Alexis for the party tonight is a miracle and a mystery.

57. Γελ κπνξεί λα απνξξίπηεη όιεο ηηο ηδέεο ρσξίο λα δώζεη ν ίδηνο θακία ιύζε θαη πξόηαζε. He cannot reject every idea without giving any solution and suggestion.

58. Ζ Μαξία ζπλεηζθέξεη ζε έλα κπινγθ, όπνπ δίλεη ζπκβνπιέο γηα κόδα θαη νκνξθηά. Maria contributes to a blog, where she gives tips on fashion and beauty.

59. Από ηηο ζπεζηαιηηέ ηνπ καγαδηνύ μαθνπζηή είλαη ε ζνύπα κε κύδηα θαη ζαιηγθάξηα. Among the specialties of the restaurant, the most famous is the soup with mussels and snails.

60. Σν επξώ είλαη έλα αζηαζέο λόκηζκα, γηαηί δελ έρεη έλα εληαίν ζύζηεκα ζεζκώλ θαη κεραληζκώλ. The euro is an unstable currency because it does not have a comprehensive system of institutions and mechanisms.

61. Σν λέν βηβιίν Αγγιηθώλ πξνζεγγίδεη δηαθνξεηηθά ηελ γιώζζα εζηηάδνληαο ζηελ ρξήζε θαη παξαγωγή. The new English book approaches language differently focusing on use and production.

62. Ωο κέιε ηνπ ζπιιόγνπ πξέπεη λα έρνπκε έληνλε δξάζε θαη ζπκκεηνρή. As members of the association we must have increased activity and participation.

63. Πξέπεη λα ηξώκε θαη όζπξηα, όπσο θαθή θαη θαζνιάδα. We must eat legumes like lentil soup and bean soup.

64. Χξεηάδεζαη βηηακίλεο, νπόηε λα βάιεηο ζηε δηαηξνθή ζνπ θξνύηα όπσο ξόδη θαη πνξηνθάιη. You need vitamins, so your diet must include fruit such as pomegranates and oranges.

65. Γελ ηνπ άξεζε ην θζηλόπσξν γηαηί ήηαλ κνπληό θαη κειαγρνιηθό. He did not like autumn because it was gloomy and dreary.

67

66. ΢ηελ ζηάλε, ν Παληειήο θάζε κέξα αξκέγεη γίδεο θαη πξνβαηίλεο. Every day in the shed, Panagiotis milks goats and sheep.

67. Ζ γηαγηά είρε πνιιέο γιάζηξεο κε ληάιηεο θαη λεξαγθνύιεο. Grandma had a lot of pots with dahlias and buttercups.

68. Ο ζπλεηαηξηζκόο ηνπ ρσξηνύ εκπνξεύεηαη δηάθνξα παξαδνζηαθά ηπξηά θαη δπκαξηθά. The cooperative of the village trades in various traditional cheeses and pasta.

Experiment 2: Non-word targets

1. Πάιη μέραζα λα πνηίζσ ηα ινπινύδηα γηαηί ηνηίγω θαη ηνηέγω. I forgot to water the plants again because I (to'tiɣo ce to'teɣo).

2. Ζ Μαξία ήμεξε πώο λα θάλεη πην λόζηηκν ην θνηόπνπιν, πξνζζέηνληαο θαηίθα θαη θαηέθα. Maria knew how to make the chicken tastier, by adding (ka'tifa ce ka'tefa).

3. Γελ ηνλ πείξαδε θαζόινπ πνπ ηνλ θνξόηδεπε θάζε κπίγνο θαη κπέγνο. He did not mind at all that he was made fun of by ('biɣos ce 'beɣos).

4. Ξέξεηο πσο αλ πξνρσξήζεηο παξαθάησ ζα ζπλαληήζεηο ηίβεο θαη ηάβεο. You know that if you move forward you will meet ('tives ce 'taves).

5. Ο παππνύο έπηλε ην νπδάθη ηνπ, θαη γηα κεδέ δήηεζε κπίλα θαη κπάλα. Grandpa was having his ouzo and asked a snack of ('bina ce 'bana).

6. Γελ κπνξεί λα βγεη από ην ζπίηη γηαηί πνθίβεη θαη πνθάβεη. He cannot get out of the house because (po'civi ce po'kavi).

7. Κάλε έλα δηάιεηκκα γηα θνπίγη θαη θνπόγη. Have a break for (ko'piʝi ce ko'poʝi).

8. Κνίηα ηνλ, πεξπαηάεη θαη λνκίδεη όηη είλαη θαλέλαο γθίβνο θαη γθόβνο. Look at him, he walks and thinks that he is some ('ɉivos ce 'govos).

9. ΢ε όπνηα άζθεζε δελ μέξσ ηη λα θάλσ, απιά κπίκω θαη κπόκω. Whenever I do not know what to do in an exercise, I just ('bimo ce 'bomo).

10. Γελ κπνξώ λα ηελ αθνύσ πηα, όιν ληίρεη θαη ληνύρεη. I cannot hear her anymore, she constantly ('diçi ce 'duçi).

11. Ζ Διέλε κπήθε ζην δσκάηην θαη ήηαλ ράιηα, ζθέην ληαπίκη θαη ληαπνύκη. Eleni got into the room and it was a mess, a real (da'pimi ce da'pumi).

68

12. Κάλαλε πνιύ παξέα, νη δπν ηνπο ήηαλ γθίκεο θαη γθνύκεο. They were very good friends, the two of them were ('ɉimis ce 'gumis).

13. Ζ αιπζίδα είρε ζθνπξηάζεη εληειώο, είρε γίλεη ζθέηε ηέζα θαη ηάζα. The chain was completely rusty, it was ('tesa ce 'tasa).

14. ΢ην ρσξηό παξαθύιαγαλ παληνύ νη θαθόβνπιεο ληνθέξεο θαη ληνθάξεο. In the village, lurking around there were devious (do'ceres ce do'kares).

15. Κνίηα πώο θπθινθνξνύλ ληπκέλνη έηζη, ζαλ πέδνο θαη πάδνο. Look at them walking around dressed like that, like ('peðos ce 'paðos).

16. Θα δηέιπε ηελ κνύρια ρξεζηκνπνηώληαο ηεθέζη θαη ηεθόζη. He would dissolve the mold using (te'cesi ce te'kosi).

17. Οη θίιεο ζνπ ζα ζε βνεζήζνπλ ζε ό,ηη θαη αλ πέγεηο θαη πόγεηο. Your girlfriends will help you in whatever you ('peʝis ce 'poʝis).

18. Μελ ηνλ εκπηζηεύεζαη, είλαη κπέλαο θαη κπόλαο. Do not trust him, he is ('benas ce 'bonas).

19. Ό,ηη ηδέα είρε ξίμεη ήηαλ κπέθα θαη κπνύθα. Whatever idea he had come up with was ('befa ce 'bufa).

20. Όπσο έηξερε ν Γηώξγνο, ηνπ έπεζαλ από ηελ ηζέπε ληηηέξηα θαη ληηηνύξηα. As Giorgos was running, he dropped (di'terʝa ce di'turʝa).

21. Πξνζπαζείο λα θάλεηο θάηη θαιό, αιιά ζην ηέινο πάληα γθέζεηο θαη γθνύζεηο. You are trying to do something good, but in the end you always ('ɉeζis kai 'guζis). 22. ΢πλόδεςαλ ηε λύθε ζηελ εθθιεζία κεηά θάρωλ θαη θόρωλ. They escorted the bride to the church with ('kaxon ce 'koxon).

23. Έηζη όπσο καγείξεςε ηε ζνύπα, είρε γίλεη ηαθάβα θαη ηαθόβα. In the way that he had cooked the soup, it had become (ta'kava ce ta'kova).

24. Δζύ μέξεηο από απηά, αθνύ όιε κέξα κπάθεηο θαη κπόθεηο. You know about these things since all day long you ('bacis ce 'bocis).

25. Ο Παλαγήο ήηαλ πνιύ θαιόο εξγάηεο, ζσζηόο θνηάδνο θαη θνηνύδνο. Panagis was a very good worker, a real (ko'taðos ce ko'tuðos).

26. Ξεθηλάσ λα πσ θάηη θαη ην μερλώ επεηδή ληάκω θαη ληνύκω. I keep forgetting what I want to say because I ('damo ce 'dumo).

27. Απηό δελ θάλεη γηα ηε δνπιεηά, γηαηί εθηόο από πνιύ κηθξό, είλαη πηθάδη θαη πηθνύδη. This is not good enough for our task, because apart from being very small, it is 69

(pi'kaði ce po'kuði).

28. Πεο κνπ ηη έγηλε αληί λα ηόθεηο θαη ηνύθεηο. Tell me what happened instead of ('tofis ce 'tufis).

29. Ζ κηθξή ΢νθία ήηαλ κηα γιύθα θνξώληαο κηθξά πνηόληα θαη πνηνύληα. Little Sofia was so sweet wearing little (po'toɲa ce po'tuɲa).

30. Γελ ζα ζνπ θάλνπλ ηε ράξε, αλ δελ ζπκθσλήζνπλ πξώηα ζαλ γθόβαο θαη γθνύβαο. They will not help you if they do not agree first like ('govas ce 'guvas).

31. Ζ ζεία από ην Κηιθίο κνπ έζηεηιε ηερί θαη θεβίο. My aunt from Kilkis has sent me (te'çi ce ce'vis).

32. Ζ δηαθνζκήηξηα κηινύζε γηα θάπνηα πεξίεξγα ρξώκαηα, όπσο γθηλό θαη ληηκόο. The decorator was talking about some strange colors such as (ɉi'no ce di'mos).

33. Σα δύν λέα είδε αξάρλεο νλνκάζηεθαλ από ηηο πεξηνρέο Γθέβα θαη Σέδαο. Two new kinds of spider were named after the regions of ('ɉeva ce 'teðas).

34. Απηέο νη ρώξεο δελ έρνπλ δνιάξηα, νύηε επξώ, αιιά θαηίξα θαη κπαθίξαο. These countries do not use dollars or euros, but rather (ka'tira ce ba'ciras).

35. Γηα ην θζηλόπσξν, γλσζηνί νίθνη κόδαο πξνηείλνπλ θνξέκαηα ζε γξακκέο δαξί θαη βαιίλ. For fall, well-known fashion houses suggest dresses in (ða'ri ce va'lin) fits.

36. Έθαηζε ινηπόλ θαη άθνπζε ηα παξαδνζηαθά ηνπο όξγαλα πηκπηξί θαη ηηγθηξίλ. So he sat down and listened to their traditional instruments, (pibi'ri ce tiɉi'rin).

37. Σα θνθηέηι καο έρνπλ άξσκα από ηα εμσηηθά θξνύηα γθόλν θαη κπόκνλ. Our cocktails smell of the exotic fruit ('gono ce 'bomon).

38. Ζ Διπίδα έθηηαμε ζηα παηδηά δύν γιπθά από ηε ΢θσηία: ηάξε θαη θάιελ. Elpida made two Scottish deserts for the children: ('tare ce 'kalen).

39. ΢εθώζεθαλ νη άληξεο λα ρνξέςνπλ ηνπο πνιεκηθνύο ρνξνύο θνδά θαη ηνγάξ. The men stood up to dance the war dances (ko'ða ce to'ɣar).

40. Σα πθάζκαηα πνπ είλαη ηδαληθά γηα παιηό έρνπλ ίλεο ληερό θαη γθεθόξ. The fabrics that are ideal for coats have fibers of (de'xo ce ɉe'for).

41. Ζ Λίδα θνξνύζε έλα ζηεθάλη από άλζε θέλν θαη ηέκνξ. Lisa wore a wreath with blossoms of ('ceno ce 'temor).

42. Οη δύν αζζέλεηεο πνπ ζέξηζαλ ηε ρώξα ηνλ πεξαζκέλν αηώλα νλνκάζηεθαλ ληίβη θαη κπίγηξ. 70

The two diseases that reaped through the country during the last century were named ('divi ce 'biʝir).

43. Από ηελ αγνξά κπαραξηθώλ είρε αγνξάζεη θέθα θαη πέκαδ. From the spice market he had bought ('cefa ce 'pemað).

44. Θα ήηαλ ε πξώηε θνξά πνπ ε Εσή ζα δνθίκαδε κπνκά θαη ληνλάδ. It would be the first time that Zoi would try (bo'ma ce do'nað).

45. Ζ πνδηά ζα έπξεπε λα ξαθηεί κε ξαθέο πάληη θαη θάκπηδ. The apron should have been sewed with ('padi ce 'kabið) seams.

46. Ζ ζνθνιαηέληα επηθάιπςε είρε εηνηκαζηεί κε κπνλεθί θαη γθαλεηίδ. The chocolate cover was prepared with (bone'ci ce gane'tið).

47. Σα ζθαζάξηα ήηαλ επηθίλδπλα γηαηί κπνξνύζαλ λα κεηαδώζνπλ ηνπο ηνύο γθίβν θαη κπίδνλη. The beetles were dangerous because they could spread the ('ɉivo ce 'biðod) viruses.

48. ΢ην ηξαπέδη ζηέθνληαλ ηα αγαπεκέλα ηεο θαγεηά από ηελ παηξίδα, θνπγά θαη πνπκάλη. On the table were her favorite foods from home, (ku'ɣa ce pu'mad).

49. Θα δνθίκαδε λα θάλεη ην άικα ρσξίο ληάθν θαη γθάρνλη. She would attempt the leap without ('dafo ce 'gaxod).

50. Ζ κηθξή κπνπηίθ βξηζθόηαλ ζηε γσλία ησλ νδώλ Σεθαβέ θαη Κεηαδέλη. The small boutique was in the corner of (teka've ce ceta'ðed).

51. Ο ρνξόο απαηηνύζε επηπιένλ εμάζθεζε ζηηο θηγνύξεο κπαβνξέ θαη ληαδνξέζ. The dance required extra practice of the moves (bavo're ce daðo'reζ).

52. Γηα ηε ζπγγξαθή ηεο εξγαζίαο ρξεζηκνπνηήζηε κόλν γξακκαηνζεηξέο θόρα θαη ηόραζ. For the composition of the paper you should only use the fonts ('koxa ce 'toxaζ).

53. Ο εθηππσηήο ρξεηάδεηαη θαηλνύξγηα γθνλί θαη κπνκίζ. The printer needs new (go'ni ce bo'miζ).

54. ΢ηελ εθζθαθή βξέζεθαλ λέα δείγκαηα θηξέκα θαη ηηλέξαζ. In the excavation, new specimens of (ci'rema ce ti'renaζ) were found.

55. Σν ξνιόη δνπιεύεη κε ειβεηηθνύο κεραληζκνύο πνβό θαη ηνδόη. The clock works with the Swiss mechanisms (po'vo ce to'ðot).

71

56. Ζ γάηα έθαλε εκβόιην γηα γθόλη θαη κπόκηη. The cad had the vaccine for ('goni ce 'bomit).

57. Σα παξαδνζηαθά δπκαξηθά δελ πξέπεη λα πεξηέρνπλ πνβά θαη γθνδάη. The traditional pasta should not contain (po'va ce go'ðat).

58. Ο θαθέο καο πξνέξρεηαη από εηδηθέο πνηθηιίεο θέγη θαη πέβηη. Our coffee comes from the specials varieties of ('ceʝi ce 'pevit).

59. Σν άξσκα ηνπ θξαζηνύ καο ζαο δξνζίδεη κε λόηεο πνλέ θαη ηνκέι. The scent of our wine will cool you down with notes of (po'ne ce to'mel).

60. Χξεηαδόηαλ κία θξέκα πξνζώπνπ ρσξίο ηαληίβα θαη θακπίδαι. She needed a face cream without (ta'diva ce ka'biðal).

61. Ζ άλνημε ζήκαηλε όηη ν αέξαο κνζρνβνινύζε ηνξέ θαη θνθέι. Spring meant that the air would smell of (to're ce ko'fel).

62. Γηα λα είλαη πην απζεληηθή ε ζπληαγή πξέπεη λα πξνζζέζεηο ηόζα θαη θόραι. In order for the recipe to be original, you must add ('toζa ce 'koxal).

63. Μαδί κε ην θξαζί, καο θέξαζαλ γθέξη θαη κπέιηκ. Along with the wine, they treated us with ('ɉeri ce 'belim).

64. Γηα ηνλ βήρα πνπ ηνλ βαζάληδε έπηλε ηζάη από κπόθη θαη γθόδηκ. For the cough he was suffering from he used to drink tea from ('bofi ce 'goðim).

65. Ζ αίζνπζα ηεο δεμίσζεο ήηαλ ζηνιηζκέλε κε ηνρό θαη γθνπόκ. The reception hall was decked with (to'xo ce go'pom).

66. Σν κνπζείν είρε έλα ππέξνρν έθζεκα γηα ηα επξήκαηα Παθέ θαη Σαβέκ. The museum had an outstanding exhibit of the discoveries of (pa'fe ce ta'vem).

67. Ξεθίλεζε γηα ην ηαμίδη ηνπ ρσξίο πέθα θαη θακέληα. He set off for his trip without ('peka ce ka'meda).

68. Ο Μάξηνο είλαη έκπηζηνο ππάιιεινο, θαλνληθόο γθάζαο θαη θαβάληαο. Marios is a trustworthy employee, a real ('gaζas ce ka'vadas).

69. Σνπ μεθίλεζε κηα αλνύζηα θνπβέληα πεξί θόρωλ θαη θαδόθωλ. He instigated a pointless discussion about ('koɣon ce ka'ðokon).

70. Καη από ηα αθνξνιόγεηα ζνπ έθεξα θόκπα θαη ηαβόθα. From the duty free shops I brought you ('foba ce ta'voka).

71. Μέζα ζηα εξγαιεία ηνπ είρε ξνύηη θαη θαληνύιη. In his toolbox he had ('ruti ce ka'duli).

72

72. Μέζα ζην δσκάηην δελ ππήξρε άληξαο πην δνύλνο θαη βαπνύξνο. Inside the room there was not a man more ('ðunos ce va'puros).

73. Έπξεπε λα δνπιεύσ ηώξα θαλνληθά, αιιά γθόκω θαη θαξόβω. I should be working now, but I ('gomo ce ka'rovo).

74. Ζ ζεία ζνπ ζα θέξεη ζαιάηα κε κάζα θαη κπνηάηα. Your aunt will bring salad with ('maζa ce bo'tata).

75. Γηα ην κπάλην ζα έπξεπε λα αγνξάζεηο πέκη θαη πεηέθη. For your bathroom you should have bought ('pemi ce pe'teci).

76. Σζηγαξίζηε ην θξεκκπδάθη πξνζζέηνληαο ζηγά-ζηγά ξίηε θαη καβίλε. Saute the onion while slowly adding ('riti ce ma'vini).

77. Έηζη όπσο βηαδόηαλε, ηνπ έπεζαλ θέθη θαη θηθνξέηη. As he was rushing, he dropped ('feci ce fiko'reti).

78. Θα βάισ λα πηνύκε ιίγε πόβα θαη ηνξηηόγθα. Let me pour us some ('pova ce tori'toga).

79. Άθνπ ηνλ, αληί λα θαζαξίδεη ηελ θνπδίλα, ληόθεη θαη κπνθαιώβεη. Listen, instead of cleaning the kitchen, he ('doci ce boka'lovi).

80. Σν παίδεη ζθιεξόο, όκσο ζηελ πξαγκαηηθόηεηα είλαη γθέκεο θαη ληνβακέηεο. He acts tough but in reality he is ('ɉemis ce dova'metis).

81. Ξεθίλεζε λα δσγξαθίδεη ρσξίο γθάξα θαη θνπηαλάια. He began painting without ('gara ce kuta'nala).

82. Μαδί κε ηνλ θαθέ, ζαο έβγαια κπόζα θαη κεθνππόηα. Along with the coffee, I have brought you ('bosa ce meku'pota).

83. Μαιώλαλε ζπλέρεηα ζαλ θνύζαο θαη θαξακάλαο. They fought constantly like ('fusas ce kara'manas).

84. Δίρε απνθαζίζεη λα ηειεηώζεη ηε δνπιεηά ηεο κε ή ρσξίο κνύβη θαη πηλαθνύξη. She decided to finish her work with or without ('muvi ce pina'kuri).

85. Σνλ ππνδέρζεθαλ ζην ρσξηό ππό ηνπο ήρνπο πάδωλ θαη θνξαηάκωλ. He was welcomed to the village under the sounds of ('pazon ce fora'tamon).

86. Θέιεη λα ζε βνεζήζεη κα κπόθεη θαη θηλαηόξεη. He wants to help you but he ('bofi ce cina'tori).

73

Appendix B: Results

1. Vowels Experiment 1 Observed Expected Item Greek IPA Gloss rates out order of 85 1 ηνπιίπαο θαη βηνιέηαο tu'lipas ce vʝo'letas tulip and violet 33 i-e 2 θαλέια θαη καζηίρα ka'nela ce ma'stixa cinnamon and mastic 42 3 πξάζηλε θαη θίηξηλε 'prasini ce 'citrini green and yellow 48 i-a 4 αληίδηα θαη ξεπάληα a'diðʝa ce re'paɲa chicories and radishes 47 5 θαξύδη θαη ιεκόλη ka'riðη ce le'moni walnut and lemon 50 i-o 6 θξεκκύδηα θαη θαζόιηα kre'miðʝa ce fa'soʎa onions and beans 41 7 θνπληνύθη θαη θπζηίθη fu'duci ce fi'stici hazelnut and peanut 46 i-u 8 ηξύπην θαη θνύθην 'tripço ce 'kufço full of holes and empty 30 9 θαζέξη θαη ζαιάκη ka'seri ce sa'lami cheese and ham 32 e-a 10 αιεύξη θαη αιάηη a'levri ce a'lati flour and salt 43 11 κπεδ θαη κσβ bez ce mov beige and purple 40 e-o 12 θξέζθεο θαη ληόπηεο 'fresces ce 'dopçes fresh and local 52 13 ηζνπξέθηα θαη θνπινύξηα tsu'reca ce ku'lurʝa Easter bread and cookies 45 e-u 14 ηέκπιν θαη ηξνύιν 'teblo ce 'trulo altar and dome 46 15 καξνύιη θαη ζπαλάθη ma'ruli ce spa'naci lettuce and spinach 36 a-u 16 πηξνύληα θαη θνπηάιηα pi'ruɲa ce ku'taʎa forks and spoons 44 17 θαξπνύδη θαη πεπόλη kar'puzi ce pe'poni watermelon and honeydew 61 u-o 18 ληόκπξν θαη ληνύξν 'dobro ce 'duro honest and harsh 33 19 κνιόρεο θαη ιεβάληεο mo'loçes ce le'vades mallow and levander 43 a-o 20 κπόηεο θαη θάιηζεο 'botes ce 'katlses boots and socks 59

Experiment 2 Observed Item Greek IPA Expected order rates out of 86 1 ηνηίγσ θαη ηνηέγσ to'tiɣo ce to'teɣo 45 2 θαηίθα θαη θαηέθα ka'tifa ce ka'tefa i-e 37 3 κπίγνο θαη κπέγνο 'biɣos ce 'beɣos 48 4 ηίβεο θαη ηάβεο 'tives ce 'taves 45 5 κπίλα θαη κπάλα 'bina ce 'bana i-a 47 6 πνθίβεη θαη πνθάβεη po'civi ce po'kavi 38 7 θνπίγη θαη θνπόγη ko'piʝi ce ko'poʝi 46 8 γθίβνο θαη γθόβνο 'ɉivos ce 'govos i-o 42 9 κπίκσ θαη κπόκσ bimo ce 'bomo 41 10 ληίρεη θαη ληνύρεη 'diçi ce 'duçi 52 11 ληαπίκη θαη ληαπνύκη da'pimi ce da'pumi i-u 41 12 γθίκεο θαη γθνύκεο 'ɉimis ce 'gumis 45 13 ηέζα θαη ηάζα 'tesa ce 'tasa 51 14 ληνθέξεο θαη ληνθάξεο do'ceres ce do'kares e-a 43 15 πέδνο θαη πάδνο 'peðos ce 'paðos 52

74

16 ηεθέζη θαη ηεθόζη te'cesi ce te'kosi 61 17 πέγεηο θαη πόγεηο 'peʝis ce 'poʝis e-o 40 18 κπέλαο θαη κπόλαο 'benas ce 'bonas 38 19 κπέθα θαη κπνύθα 'befa ce 'bufa 33 20 ληηηέξηα θαη ληηηνύξηα di'terʝa ce di'turʝa e-u 45 21 γθέζεηο θαη γθνύζεηο 'ɉeζis kai 'guζis 45 22 θάρσλ θαη θόρσλ 'kaxon ce 'koxon 38 23 ηαθάβα θαη ηαθόβα ta'kava ce ta'kova a-o 48 24 κπάθεηο θαη κπόθεηο 'bacis ce 'bocis 62 25 θνηάδνο θαη θνηνύδνο ko'taðos ce ko'tuðos 31 26 ληάκσ θαη ληνύκσ 'damo ce 'dumo a-u 52 27 πηθάδη θαη πηθνύδη pi'kaði ce po'kuði 43 28 ηόθεηο θαη ηνύθεηο 'tofis ce 'tufis 29 29 πνηόληα θαη πνηνύληα po'toɲa ce po'tuɲa u-o 40 30 γθόβαο θαη γθνύβαο 'govas ce 'guvas 43

2. Final Coda

Coda is Item Greek IPA Gloss Target second (out of 85) 21 κπνύηη θαη ζηήζνο 'buti ce 'stiζνs thigh and breast 35 22 ιάδηα θαη ζέξβηο laðʝa ce 'servis oil and service 38 s 23 θήκε θαη ηζρύο 'fimi ce i'sçis reputation and clout 24 24 ζίγνπξα θαη αζθαιώο 'siɣura ce asfa'los surely and certainly 40 25 κπηιηάξδν θαη κπάληκηληνλ bi'ʎarðo ce 'badmidon billiard and badminton 60 26 αβγό θαη κπέηθνλ a'vɣo ce 'beikon egg and bacon 68 n 27 παιηό θαη κπνπθάλ pal'to ce bu'fan coat and jacket 33 28 ηνύιη θαη ζαηέλ 'tuli ce sa'ten tulle and satin 29 29 δόισκα θαη δέιεαξ 'ðoloma ce 'ðelear bait and lure 36 30 λεθξά θαη ήπαξ ne'fra ce 'ipar kidneys and liver 41 r 31 δόκπη θαη βακπίξ 'zobi ce va'bir zombie and vampire 39 32 θόζκεκα θαη αμεζνπάξ 'kozmima ce aksesu'ar jewelry and accessory 49 33 ξάγθκπη θαη κπέηδκπνι 'ragbi ce 'beizbol rugby and basebol 35 34 κπαιέην θαη κηνύδηθαι ba'leto ce 'mjuzikal ballet and musical 48 l 35 βξάρηα θαη ηνύλει 'vraça ce 'tunel rocks and tunnels 54 36 παζέ θαη κπαλάι pa'se ce ba'nal outmoded and unoriginal 38 37 Μπελθίθα θαη Σόηελακ ben'fika ce 'totenam Benfica and Tottenham 42 38 ινπηξό θαη ρακάκ lu'tro ce xa'mam bathroom and Turkish bath 64 m 39 θπξά θαη καληάκ ci'ra ce ma'dam old lady and madam 41 40 Διζίλθη θαη Άκζηεξληακ el'sinki ce 'amsterdam Helsinki and Amsterdam 37 41 ηδνύλην θαη θξίθεη 'dzudo ce 'kricet judo and cricket 46 42 κάδα θαη ειίη 'maza ce e'lit masses and elite 51 t 43 θηλεηό θαη ίληεξλεη cini'to ce 'idernet cellphone and internet 68 44 πνύκα θαη θνγηόη 'puma ce ko'ʝot puma anf coyote 49 75

45 πόλπ θαη ζνπέλη 'poni ce su'ed pony leather and suede 35 46 αγνξέ θαη πεξκαλάλη aɣo're ce perma'nad pixie cut and permed 60 d 47 ζηλεκά θαη Χόιηγνπλη sine'ma ce 'xoliɣud cinema and Hollywood 61 48 Ρίκηλη θαη ΢ηάιληγθξαλη 'rimini ce 'stalingrad Rimini and Stalingrad 38

Experiment 2 Coda is second (out of Item Greek IPA Target 86) 31 ηερί θαη θεβίο te'çi ce ce'vis 39 32 γθηλό θαη ληηκόο ɉi'no ce di'mos 44 s 33 γθέβα θαη ηέδαο 'ɉeva ce 'teðas 49 34 θαηίξα θαη κπαθίξαο ka'tira ce ba'ciras 51 35 δαξί θαη βαιίλ ða'ri ce va'lin 43 36 πηκπηξί θαη ηηγθηξίλ pibi'ri ce tiɉi'rin 46 n 37 γθόλν θαη κπόκνλ 'gono ce 'bomon 48 38 ηάξε θαη θάιελ 'tare ce 'kalen 39 39 θνδά θαη ηνγάξ ko'ða ce to'ɣar 43 40 ληερό θαη γθεθόξ de'xo ce ɉe'for 47 r 41 θέλν θαη ηέκνξ 'ceno ce 'temor 46 42 ληίβη θαη κπίγηξ 'divi ce 'biʝir 42 43 θέθα θαη πέκαδ 'cefa ce 'pemað 50 44 κπνκά θαη ληνλάδ bo'ma ce do'nað 44 ð 45 πάληη θαη θάκπηδ 'padi ce 'kabið 44 46 κπνλεθί θαη γθαλεηίδ bone'ci ce gane'tið 43 47 γθίβν θαη κπίδνλη 'ɉivo ce 'biðod 48 48 θνπγά θαη πνπκάλη ku'ɣa ce pu'mad 40 d 49 ληάθν θαη γθάρνλη 'dafo ce 'gaxod 43 50 ηεθαβέ θαη θεηαδέλη teka've ce ceta'ðed 37 51 κπαβνξέ θαη ληαδνξέζ bavo're ce daðo'reζ 49 52 θόρα θαη ηόραζ 'koxa ce 'toxaζ 38 ζ 53 γθνλί θαη κπνκίζ go'ni ce bo'miζ 45 54 θηξέκα θαη ηηξέλαζ ci'rema ce ti'renaζ 48 55 πνβό θαη ηνδόη po'vo ce to'ðot 50 56 γθόλη θαη κπόκηη 'goni ce 'bomit 51 t 57 πνβά θαη γθνδάη po'va ce go'ðat 48 58 θέγη θαη πέβηη 'ceʝi ce 'pevit 43 59 πνλέ θαη ηνκέι po'ne ce to'mel 47 60 ηαληίβα θαη θακπίδαι ta'diva ce ka'biðal 45 l 61 ηνξέ θαη θνθέι to're ce ko'fel 41 62 ηόζα θαη θόραι 'toζa ce 'koxal 35 63 γθέξη θαη κπέιηκ 'ɉeri ce 'belim 47 64 κπόθη θαη γθόδηκ 'bofi ce 'goðim 53 m 65 ηνρό θαη γθνπόκ to'xo ce go'pom 39 66 παθέ θαη ηαβέκ pa'fe ce ta'vem 48

76

3. Syllable Count

Experiment 1 Short is Syllable Item Greek IPA Gloss first (out Pattern of 85) 49 αξληά θαη θαηζίθηα ar'ɲa ce ka'tsica lambs and goats 2 and 3 59 50 καύξν θαη θόθθηλν 'mavro ce 'kocino black and yellow 39 51 μαλζά θαη θαζηαλά ksan'ζa ce kasta'na blond and brown 47 52 πεηξόι θαη βεξακαλ pe'trol ce vera'man teal and mint green 57 53 πάπηεο θαη ρειώλεο 'papçes ce çe'lones ducks and turtles 53 54 πξάζν θαη ζέιηλν 'praso ce 'selino leek and celery 51 55 θάξπ θαη θύκηλν 'kari ce 'cimino curry and cumin 33 56 ζαύκα θαη κπζηήξην 'ζavma ce mi'stirio miracle and mystery 37 57 ιύζε θαη πξόηαζε 'lisi ce 'protasi solution and suggestion 22 58 κόδα θαη νκνξθηά 'moða ce omor'fça fashion and beauty 59 59 κύδηα θαη ζαιηγθάξηα 'miðʝa ce sali'garʝa mussels and snails 2 and 4 53 60 ζεζκώλ θαη κεραληζκώλ ζe'zmon ce mixani'zmon institutions and mechanisms 40 61 ρξήζε θαη παξαγσγή 'xrisi ce paraɣo'ɣi use and production 37 62 δξάζε θαη ζπκκεηνρή 'ðrasi ce simeto'çi activity and participation 24 63 θαθή θαη θαζνιάδα fa'ki ce faso'laða lentil soup and bean soup 49 64 ξόδη θαη πνξηνθάιη 'roði ce porto'kali pomegranate and orange 34 65 κνπληό θαη κειαγρνιηθό mu'do ce melaŋxoli'ko gloomy and dreary 64 66 γίδεο θαη πξνβαηίλεο 'ʝiðes ce prova'tines goats and sheep 61 67 ληάιηεο θαη λεξαγθνύιεο 'daʎes ce nera'gules dahlias and buttercups 55 68 ηπξηά θαη δπκαξηθά ti'rʝa ce zimari'ka cheeses and pasta 45

Experiment 2 Short is first Item Greek IPA Syllable Pattern (out of 86) 67 πέθα θαη θακέληα 'peka ce ka'meda 65 68 γθάζαο θαη θαβάληαο 'gaζas ce ka'vadas 59 69 θόγσλ θαη θαδόθσλ 'koɣon ce ka'ðokon 56 70 θόκπα θαη ηαβόθα 'foba ce ta'voka 54 71 ξνύηη θαη θαληνύιη 'ruti ce ka'duli 58 2 and 3 72 δνύλνο θαη βαπνύξνο 'ðunos ce va'puros 53 73 γθόκσ θαη θαξόβσ 'gomo ce ka'rovo 67 74 κάζα θαη κπνηάηα 'maζa ce bo'tata 59 75 πέκη θαη πεηέθη 'pemi ce pe'teci 54 76 ξίηε θαη καβίλε 'riti ce ma'vini 57 77 θέθη θαη θηθνξέηη 'feci ce fiko'reti 55 78 πόβα θαη ηνξηηόγθα 'pova ce tori'toga 49 79 ληόθεη θαη κπνθαιώβεη 'doci ce boka'lovi 58 80 γθέκεο θαη ληνβακέηεο 'ɉemis ce dova'metis 2 and 4 55 81 γθάξα θαη θνπηαλάια 'gara ce kuta'nala 48 82 κπόζα θαη κεθνππόηα 'bosa ce meku'pota 41 83 θνύζαο θαη θαξακάλαο 'fusas ce kara'manas 60 77

84 κνύβη θαη πηλαθνύξη 'muvi ce pina'kuri 52 85 πάδσλ θαη θνξαηάκσλ 'pazon ce fora'tamon 48 86 κπόθεη θαη θηλαηόξεη 'bofi ce cina'tori 56

78