The role of similarity in co-occurrence restrictions: Evidence from the Hebrew verbal system - Hadas Yeverechyahu&#39;s Thesis

Evan Cohen

In Semitic languages, homorganic consonants tend not to co-occur within the same stem (Greenberg 1950). Previous studies (e.g. McCarthy 1981, 1986; Frisch et al. 2004) suggested that these restrictions are due to similarity effects, that is, the greater the similarity between two (homorganic) consonants, the less likely they are to co-occur. The current study examines the restrictions in the Hebrew verbal system. I ask how similarity between consonants contributes to restrictions, and whether they are due to a universal constraint or influenced by language-specific factors. The study has three main parts. First, I applied Frisch et al.'s (2004) similarity model to the Hebrew consonant inventory. Second, I analyzed the Hebrew verbal lexicon, focusing on the co-occurrences of C1-C2 stem consonants in the verb classes kal (CaCaC) and pi'el (CiCeC). The analysis shows a highly significant correlation between the similarity scale and the lexicon, and also suggests that place of articulation has a major role in the restrictions (compared to other features). To strengthen and complement the lexical analysis, I conducted two psycholinguistic experiments: a lexical decision task and a word-likelihood judgment task, both examine the co-occurrence restrictions in the speakers' phonological system. The results of the judgment task were highly correlated with the similarity scale and with the lexical analysis. The experiments also highlight the role of place features in the restrictions. These findings suggest that there are similarity based co-occurrence restrictions on stem consonants C1-C2, both in the lexicon and in the speakers' phonological system. They also suggest that place features have a major role in the restrictions, such that consonants that share the major place feature are less likely to co-occur. However, the experiments cannot suggest whether the influence of similarity on the grammatical system is direct, or indirect through the lexical influences.

The Lester and Sally Entin Faculty of Humanities Department of Linguistics THE ROLE OF SIMILARITY IN CO-OCCURRENCE RESTRICTIONS: EVIDENCE FROM THE HEBREW VERBAL SYSTEM M.A. thesis submitted by Hadas Yeverechyahu Prepared under the guidance of: Prof. Outi Bat-El Dr. Evan-Gary Cohen December 2014 ACKNOWLEDGEMENTS Looking back on the five years I spent at the Linguistics department, as a BA and a graduate student, I feel grateful for the opportunity I have been given to study and explore the field I am passionate about the most. Many people have guided and accompanied me on this fascinating way, and I would like to thank them for that. First and Foremost, I would like to express my deep gratitude to my advisors, Outi Bat-El and Evan-Gary Cohen. It was in Outi's and Evan's courses during my BA studies that I discovered that my passion has a name - phonology. I would like to thank them for teaching me, guiding me, and always encouraging me to ask the right questions and seek for explanations open-mindedly. I appreciate their being always available for consultation, and their extremely helpful comments, from insights about the analysis to help in translating animals' names. I could not ask for better advisors. I was fortunate to be a student of Lior Laks in my first semester at the department, and ever since I benefited from his helpful advice and comments in personal correspondence, and at the TAU colloquium. I would also like to thank Orna Peleg, for her helpful comments on my research proposal, and for always showing interest in my study. I would like to thank other faculty members that I had the privilege to participate in their courses: Galit Adam, Mira Ariel, Irena Botwinik, Julie Fadlon, Julia Horbath, Aya Meltzer-Asscher, Nirit Kadmon, and Tali Siloni. I would also like to thank Moshe Florentin of the Hebrew Culture department, who taught me everything I know about the phonology and morphology of Biblical Hebrew. The lexical analysis was completed with the help of Shmuel Bolozky who sent me a digital copy of Evan-Shoshan dictionary. I would also like to thank the participants i in my experiments, and to Aya Vituri for the invaluable help with the statistical analysis and insights. I had the opportunity to present my study in several forums, and I would like to thank the audiences of the TAU colloquium, the Israeli Phonology Circle, and IGDAL 2 for their interesting comments. I would like to thank Tal Oded and Ruti Zusman, the former and current secretaries of the department (respectively), whose doors were always open. Thank you for all the help (and always in no-time) and for always opening my day with a smile. The department was my second home in the last five years, and I had the honor to share it with other excellent young linguists. I would like to thank my phonologist friends, Daniel Asherov, Si Berrebi, Yael Firer, Noa Karni, Enav Kedar, Stav Klein, Ezer Rasin, and Hadass Zaidenberg, for their helpful insights along the way. I would also like to thank Noa Brandel, Hila Davidovich, Noa Geller, Dolly Goldenberg, Omer Goldman, Anat Grubin, Netanel Haim, Ela Hillel, and Shiran Ofek for their comments on my experiments and for helping distributing them. Finally, special thanks to my dear friend, Tali Arad, for her wise comments, her advice, ideas and support. And to all of you, I would like to thank for the friendship and support along the way. I would also like to thank my non-linguistic friends, and especially to Naomi and Ohad Feldheim (and their adorable son Lahav), for hours of delightful talks about linguistics in general and on my thesis in particular, and for their interesting insights. Last but not least, I would like to thank my parents and my brother, for their great support and interest in my research. And finally, I would like to thank Chen Gafni. Thank you for always being there for me both professionally and personally, for your comments about any idea and thought, for reading every draft, and above all, thank you for your endless love and support. ii ABSTRACT In Semitic languages, homorganic consonants tend not to co-occur within the same stem (Greenberg 1950). Previous studies (e.g. McCarthy 1981, 1986; Frisch et al. 2004) suggested that these restrictions are due to similarity effects, that is, the greater the similarity between two (homorganic) consonants, the less likely they are to co-occur. The current study examines the restrictions in the Hebrew verbal system. I ask how similarity between consonants contributes to restrictions, and whether they are due to a universal constraint or influenced by language-specific factors. The study has three main parts. First, I applied Frisch et al.'s (2004) similarity model to the Hebrew consonant inventory. Second, I analyzed the Hebrew verbal lexicon, focusing on the co-occurrences of C1-C2 stem consonants in the verb classes kal (CaCaC) and pi'el (CiCeC). The analysis shows a highly significant correlation between the similarity scale and the lexicon, and also suggests that place of articulation has a major role in the restrictions (compared to other features). To strengthen and complement the lexical analysis, I conducted two psycholinguistic experiments: a lexical decision task and a word-likelihood judgment task, both examine the cooccurrence restrictions in the speakers' phonological system. The results of the judgment task were highly correlated with the similarity scale and with the lexical analysis. The experiments also highlight the role of place features in the restrictions. These findings suggest that there are similarity based co-occurrence restrictions on stem consonants C1-C2, both in the lexicon and in the speakers' phonological system. They also suggest that place features have a major role in the restrictions, such that consonants that share the major place feature are less likely to co-occur. However, the experiments cannot suggest whether the influence of similarity on the grammatical system is direct, or indirect through the lexical influences. iii TABLE OF CONTENTS 1. Introduction 1 2. Theoretical Background 3 2.1. Co-Occurrence Restrictions 3 2.2. Similarity 5 3. The Issue 8 3.1. Research Questions 8 3.2. Data 8 4. The Similarity Model 9 5. Hypotheses 13 6. The Lexical Analysis 15 6.1. Design 15 6.2. Results 17 6.2.1. Observed vs. Expected (O\E) 17 6.2.2. Observations 18 6.2.3. Correlation between the Lexicon and the Similarity Scale 22 6.2.4. Conclusions 23 7. The Experiments 24 7.1. The Lexical Decision Experiment 24 7.1.1. Participants 25 7.1.2. Stimuli 25 7.1.3. Procedure 26 7.1.4. Results 27 7.1.5. Discussion 28 7.2. Word-Likelihood Judgments Experiment 28 7.2.1. Participants 29 7.2.2. Stimuli 29 7.2.3. Procedure 32 7.2.4. Results 33 7.2.4.1. Pre-analysis: the Effect of C3 33 7.2.4.2. Observations 34 7.2.4.3. Comparisons to the Scales 37 8. General Discussion 39 9. Concluding Remarks 42 Appendix 45 iv CHAPTER 1: INTRODUCTION In Semitic languages, homorganic consonants (i.e. consonants that share place of articulation) tend not to co-occur within the same stem. For example, verbs as datam or kaɡam are not likely to be found in the lexicon of any Semitic language (Greenberg 1950). Previous studies (as McCarthy 1981, 1986; Frisch et al. 2004) attributed these restrictions to similarity effects, such that the greater the similarity between two (homorganic) consonants, the less likely they are to co-occur. The current study examines the co-occurrence restrictions in the Hebrew verbal system, focusing on the contribution of similarity between consonants to these restrictions. The study focuses on the co-occurrences of C1-C2 stem consonants in the verb classes kal (CaCaC) and pi'el (CiCeC), both in the verbal lexicon (lexical analysis) and in the phonological systems of the speakers (psycholinguistic experiments). Similarity between consonants was calculated based on Frisch et al.'s (2004) similarity model (originally proposed for Arabic), adjusted according to the Hebrew consonant inventory. Three main questions were asked in the study: a. What are the co-occurrence restrictions in the Hebrew verbal system? b. What is the role of similarity in the co-occurrence restrictions? c. Nature vs. Nurture: are co-occurrence restrictions caused by a universal constraint or influenced by language-specific lexical factors? The study has three main parts. Each part tested a different aspect of the cooccurrence restrictions, and then the correlations between the parts were examined. The parts of the study are as follows: a. Application of Frisch et al.'s (2004) model to the Hebrew consonant inventory (§4); b. Lexical analysis of co-occurrence restrictions in the Hebrew verbal lexicon (§6); 1 c. Two psycholinguistic experiments: a lexical decision task and a judgment task. Both experiments examine the role of co-occurrence restrictions in the speakers' phonological system (§7). The results show a highly significant correlation between the similarity scale and the lexical analysis, and also between the similarity scale and the results of the judgment experiment. A correlation was found between the lexical analysis and the judgment experiment results as well. These findings suggest that there are similarity based co-occurrence restrictions on stem consonants C1 and C2, both in the lexicon and in the speakers' phonological system. However, the experiments cannot suggest whether the influence of similarity on the grammatical system is direct, or indirect through the lexical influences. In addition, the results suggest that place of articulation has a major role in the restrictions (compared to other features), such that consonants that share the major place feature are less likely to co-occur. This finding strengthens previous claims regarding the important role of OCP-Place in co-occurrence restrictions in Semitic languages (McCarthy 1981, 1986; Frisch et al. 2004 among others). However, the highly significant correlation between the results and the similarity scale proposes that not only the major place feature affects co-occurrence restrictions; if so, we would expect to see no effect in non-homorganic pairs. The study is organized as follows: §2 provides a theoretical background for the study; §3 presents the main issue: the research questions and the data sources; §4 presents the accommodation of Frisch et al.'s model to Hebrew; §5 is dedicated to the different hypotheses of the study; §6 presents the lexical analysis; §7 presents the psycholinguistic experiments: §7.1 describes the lexical decision experiment and §7.2 describes the word-likelihood judgment experiment; §8 discusses the study's results; I conclude in §9. 2 CHAPTER 2: THEORETICAL BACKGROUND The study examines the correlation between co-occurrence restrictions and segment similarity. In this chapter, I provide the theoretical background for the study: §2.1 presents co-occurrence restrictions and the OCP, and §2.2 presents previous studies on similarity. 2.1 Co-Occurrence Restrictions In an extensive cross-linguistic research, Greenberg (1950) observed that in Semitic languages, there are no verbs in which the first two stem consonants are identical (e.g. didem), and more generally, that homorganic consonants tend not to co-occur within the same stem. McCarthy (1979, 1981, 1986) provides a theoretical account for this phenomenon based on the Obligatory Contour Principle (OCP; Leben 1973, Goldsmith 1976), which was originally formulated for tonal systems. McCarthy expanded the principle to root consonants in Semitic verbal systems (1979, 1981, 1986) and it was further broadened to segments in general, features, syllables, and even morphemes (see Yip 1998). A common definition of OCP, cited from McCarthy (1986a:208), appears in (1). (1) The Obligatory Contour Principle (OCP): At the melodic level, adjacent identical elements are prohibited. McCarthy (1986) suggests that stem consonants and vocalic patterns are independent morphological units, located on different tiers. Since stem consonants are adjacent on their tier, the OCP rules out any representation with adjacent identical element. Note that the second and the third stem consonants (C2 and C3) are allowed to be identical 3 (e.g. dimem 'to bleed', kilel 'to curse', miʃeʃ 'to grope').1 McCarthy suggests that in these verbs, the stem contains only two consonants, and the second consonant C2 spread into the empty C3 slot. This type of verb is beyond the scope of this study. Rose (2000) claims that these restrictions can be explained without referring to tiers. In her view, the OCP is not restricted to adjacent consonants but depends on proximity as well. Thus, identical consonants separated by vowels (i.e. CiVCi) also violate the OCP constraint, though to a lesser extent than C iCi given the higher proximity. Along this line, the restrictions on C1 and C2 will be greater than the restrictions on C1 and C3, since C1-C3 are farther away from each other. Greenberg (1950) indeed shows this tendency, as does Frisch et al. (2004). The current study examines only C1-C2, and leaves proximity for further research. Hebrew and Arabic supply evidence for these co-occurrence restrictions. Laks (2011) shows blocking due to OCP in Hebrew and Arabic, where some verbs fail to undergo valence-changing operations since such operation would lead to an OCP violation. For example, dike 'to make depressed' does not undergo the valence-changing operation to *hitdake 'to get depressed', although it is semantically possible. If such a derivation had occurred, it would have created an OCP violation (t-d). In addition, OCP restrictions have empirical support from psycholinguistic experiments: Frisch and Zawaydeh (2001) for Arabic, Berent and colleagues (Berent and Shimron 1997, Berent, Everett and Shimron 2001 among others) for Hebrew. Bat-El (2003) claims that these restrictions are not unique to Semitic languages, and that co-occurrence restrictions on stem consonants can be found in non-Semitic languages as well. In English, for example, there are no monosyllabic words of the form sCVC in which the same non-coronal consonant (i.e. labial or velar) appears in both 1 The verbs are presented in 3rd person singular past throughout. 4 sides of the vowel, for example *spep, *skik (Fudge 1969, Clements and Keyser 1983, Davis 1984). In Japanese, co-occurrences restrictions on homorganic consonants are found in Yamato (native-Japanese) stems. In addition, Japanese has blocking effects due to OCP violations. Consider, for example, the phenomenon of Rendaku - voicing of the first consonant of the second member in a compound. For historical reasons, h alternates with b, as in nui 'saw' + hari 'needle' → nui-bari 'sewing needle'. However, when the stem begins with h followed by m, Rendaku is blocked in order to avoid two near homorganic consonants, for example mai 'dance' + hime 'princess' → mai-hime 'dancing princess', and not *mai-bime (Kawahara et al. 2006). Note that when there is a nonlabial consonant between the h and the m, Rendaku does occur (e.g. ryoori-basami 'cooking scissors', naga-bakama 'long hakama'). This finding suggests that proximity plays a role as well. 2 McCarthy (1986) suggests that blocking due to an OCP violation is universal. Odden (1988) stipulates that blocking differs cross-linguistically, and language differ in the sets of relevant features for the principle. In Optimality Theory (Prince and Smolensky 1993) this is represented by different constraint rankings in different languages. 2.2 Similarity Different studies (Pierrehumbert 1993, Frisch et al. 2004, Mielke 2009 among others) have addressed the question of how segment similarity should be measured. The current study focuses on the phonological approach that is based on articulatory phonological features; other approaches, like those based on acoustic parameters (see, for example, Mielke 2009), are beyond the scope of this study. 2 See Yip (1988) for more examples of OCP as process-trigger or process-blocker. 5 Pierrehumbert (1993) calculates similarity between two segments by counting the number of feature values the segments share. Frisch et al. (2004) expand this model to a natural-classes-based model, in which similarity value is computed for each pair of segments by the number of natural classes they share. Thus, in Frisch et al.'s model, similarity is computed by dividing the number of shared natural classes of two segments by the sum of the shared and non-shared natural classes of the two segments. The formula appears in (2). (2) Frisch et al.'s (2004) similarity formula 𝑠𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦 = 𝑠ℎ𝑎𝑟𝑒𝑑 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑙𝑎𝑠𝑠𝑒𝑠 𝑠ℎ𝑎𝑟𝑒𝑑 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑙𝑎𝑠𝑠𝑒𝑠 + 𝑛𝑜𝑛 𝑠ℎ𝑎𝑟𝑒𝑑 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑙𝑎𝑠𝑠𝑒𝑠 By this procedure, a similarity scale for each language can be computed, based on the contrastive features and natural classes of the language. Frisch et al. (2004) tested the model on Arabic verb stem consonants, looking for OCP restrictions in the verbal paradigms. First, they showed that OCP restrictions do occur in the lexicon, where combinations of consonants with shared features are underrepresented systematically. Next, using the above formula, Frisch et al. constructed a similarity scale based on natural classes defined according to contrastive phonological features found in Arabic. 3 Then, the results of the lexical study were examined in light of the similarity scale. The study showed a strong correlation between them, namely the similarity scale, based on natural classes, successfully explaining the co-occurrence restrictions in the Arabic lexicon. The current study will examine this model in Hebrew, by a lexical study and psycholinguistic experiments. 3 [±consonantal], [±sonorant], [±continuant], [±acute], [±strident], [±nasal], [±lateral], [labial], [coronal], [dorsal], [pharyngeal], [radical], [±anterior], [±back], [±voice], [±spread glottis], and [±constricted glottis]. 6 Next, the question arises as to which features are relevant to similarity. Rose and Walker (2004) claim that [sonorant], [continuant] and place features are the most important in computing similarity. Kawahara (2007) suggests that manner features (such as palatalization, voicing, nasalization, and continuity) contribute to similarity more than place features. This is compatible with claims that manner features are perceptually more salient, and that speakers tend to rely on acoustic parameters while calculating similarity (see Mielke 2009). Kaisse (1988) claims that the OCP applies to feature groups and not just to single features, and so provides direct evidence for Feature Geometry, which argues for feature hierarchies (Clements 1985, Sagey 1986, Clements and Hume 1995, see also McCarthy 1988). Along this line, Padgett (1995:181) revised the definition of the OCP to take into account feature hierarchies (3): (3) The Obligatory Contour Principle (OCP): At the melodic level, adjacent identical elements F F are prohibited, iff all subsidiary features stipulated for F are also identical. Along the line of Frisch et al.'s (2004) study, the current study examines natural classes and does not test the influence of every feature individually. The only feature that is examined separately is place of articulation, following the importance of OCPPlace as suggested by previous studies (see §2.1). Further research is needed in order to test the influence of other features and the correlations with feature hierarchies. 7 CHAPTER 3: THE ISSUE 3.1 Research Questions The goal of this study is to examine the co-occurrence restrictions in the Hebrew verbal system, focusing on the contribution of similarity between consonants to these restrictions. Three main questions are addressed: a. What are the co-occurrence restrictions in the Hebrew verbal system? While previous studies (McCarthy 1981, 1986; Pierrehumbert 1993; Frisch et al. 2004) focused on OCP-Place, shared place as a necessary feature for the effect, the current study attributes equal weight to all features. b. What is the role of similarity in the co-occurrence restrictions? c. Nature vs. Nurture: are co-occurrence restrictions caused by a universal constraint or influenced by language-specific lexical factors? 3.2 Data The study examines the co-occurrence restrictions in two sources of data: a. The Hebrew verbal lexicon, focusing on C1-C2 stem consonants in verb classes kal (CaCaC) and pi'el (CiCeC). I use the list of verbs from the Even-Shoshan dictionary (edition 1970 with completions from 1983). 4 b. Two psycholinguistic experiments, a lexical decision task and a word-likelihood judgment task, to examine the role of similarity in the speakers' phonological system. All the data were analyzed with respect to Frisch et al.'s (2004) model. This is, inter alia, since Arabic and Hebrew are historically related (Schwarzwald 2002 among many others), and Frisch et al.'s model successfully explained the OCP effect in Arabic verbs. 4 Many thanks to Shmuel Bolozky for an electronic version of the verb list. 8 CHAPTER 4: THE SIMILARITY MODEL The first part of the study applies Frisch et al.'s (2004) similarity model to the Hebrew consonant inventory. As discussed in §2, the model computes a single similarity value (from 0 to 1) for each pair of consonants, and the computation is based on the natural classes to which the consonants belong. The classes are defined according to the language contrastive features. The Hebrew consonant inventory appears in (4), and the set of contrastive features I used appears in (5). (4) Hebrew consonants Bilabial Labio- Alveolar dental Plosive p f Palatal v Uvular t d s z ɡ k ʃ x5 ʦ Affricate Nasal m n Lateral l Approximant (5) Velar alveolar b Fricative Palato- ʁ6 j Set of contrastive features p b m f v t d s z ʦ ʃ n j k ɡ x ʁ [±sonorant] - - + - - - - - - - - + + + - - - + [LAB] √ √ √ √ √ l √ √ √ √ √ √ √ √ √ [COR] √ √ √ √ [strident] [±anterior] + + + + + - + + √ √ √ √ [DOR] [±continuant] - - [±voice] - + - + + - - + + - + - - + - - - + 5 + - + + - - + + - + - Bolozky and Kreitman (2007) consider the Hebrew dorsal fricative to be uvular. Nevertheless, its exact place of articulation has no consequences for the current study, since minor place features for the dorsals are not contrastive in Modern Hebrew. 6 The Hebrew rhotic is considered a uvular approximant with certain frication (Bolozky and Kreitman 2007), IPA: [ʁ̞]. Hereinafter it will be transcribed as ʁ. 9 I excluded borrowed consonants (ʒ, ʤ, ʧ and w) from the analysis due to their rare appearance in the verbal system, and the glottals (ʔ and h) due to their tendency to be omitted in Modern Hebrew.7 This was done mainly for the sake of comparison between the lexical analysis and the experiments' results. The feature system I used is based on binary (6a) and unary (6b) values: (6) Features: a. Binary: [±sonorant], [±continuant], [±voice], [±anterior] b. Unary: place features: [LAB] (labials), [COR] (coronals), [DOR] (dorsals); [strident] Two issues should be noted: First, [±voice] is relevant only for obstruents; it is not contrastive among sonorants, and it has been claimed that the voice feature of the sonorants is inherent in them and therefore differs from the voice feature of the obstruents (Rice 1993). Second, I refer to stridency as a unary feature, such that the value [-strident] is not a part of the system. The stridents in Hebrew show a common phonological behavior – they undergo metathesis in binyan hitpa'el (e.g. hit-sapeʁ → histapeʁ 'to have a haircut'). Therefore, [strident] is relevant for the Hebrew phonological system.8 The non-strident consonants, on the other hand, do not show any common phonological behavior in Hebrew, and [-strident] is also not necessary for minimal distinction between consonants in the system. For these reasons, I excluded the [-] value of this feature from the analysis. 9 The natural classes were defined based on this feature system, down to the level of singletons. 7 The question of whether the phonological system of Hebrew represents glottals is beyond the scope of this paper. 8 Note that [strident] is more of an acoustic rather than articulatory feature. Nonetheless, it is widely used (also in Frisch et al.'s model) and explains different phonological processes in different languages. 9 Frisch et al. (2004) used [-strident] only for non-strident coronal fricatives. Hebrew has no such consonants in its inventory. 10 Based on this phonological feature system, I computed the similarity value for each pair of consonants, using Frisch et al.'s (2004) formula. The formula was presented in (2), and is repeated in (7). The natural classes and the similarity values were calculated via a Microsoft Excel macro.10 (7) Frisch et al.'s (2004) similarity formula 𝑠𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦 = 𝑠ℎ𝑎𝑟𝑒𝑑 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑙𝑎𝑠𝑠𝑒𝑠 𝑠ℎ𝑎𝑟𝑒𝑑 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑙𝑎𝑠𝑠𝑒𝑠 + 𝑛𝑜𝑛 𝑠ℎ𝑎𝑟𝑒𝑑 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑙𝑎𝑠𝑠𝑒𝑠 For example, the similarity value for p and b is calculated as follows: they share 7 natural classes, and do not share 8 classes, namely, there are 8 natural classes in which only one of them is a member (see list in (8)). Therefore, the similarity value for the pair p,b is: (8) 7 7+8 = 0.467.11 Shared and non-shared natural classes for the pair p-b: a. Shared classes: [-son] = {p,b,f,v,t,d,s,z,ʦ,ʃ,k,ɡ,x}, [-son, LAB] = {p,b,f,v}, [son, LAB, -cont] = {p,b}, [-son, -cont] = {p,b,t,d,ʦ,k,ɡ}, [LAB] = {p,b,m,f,v}, [LAB, -cont] = {p,b,m}, [-cont] = {p,b,m,t,d,ʦ,n,k,ɡ}. b. Non-shared classes: i. p and not b: [-son, LAB, -cont, -voice] = {p}, [-son, LAB, -voice] = {p,f}, [-son, -cont, -voice] = {p,t,ʦ,k}, [-son, -voice] = {p,f,t,s,ʦ,ʃ,k,x}; ii. b and not p: [-son, LAB, -cont, +voice] = {b}, [-son, LAB, +voice] = {b,v}, [-son, -cont, +voice] = {b,d,ɡ}, [-son, +voice] = {b,v,d,z,ɡ}. Note that although there is only one feature that distinguishes these two segments, [±voice], they have eight non-shared natural classes. Since the calculation is 10 Many thanks to Chen Gafni for programming the Natural Classes Generator on Microsoft Excel platform. The full list of natural classes appears in Appendix A. 11 I treat similarity as symmetrical, and (a)symmetry is beyond the scope of this paper. See also §9. 11 based on natural classes and not on features directly, the distance between them in this model is more notable. Table (9) presents the similarity values for the consonants in Hebrew, and table (10) the most similar pairs on the scale. A full list appears in Appendix B. (9) p Similarity values b m f v t d s z ʦ ʃ n l j k ɡ x ʁ p 1.000 0.467 0.200 0.313 0.167 0.250 0.143 0.067 0.037 0.192 0.074 0.050 0.000 0.000 0.313 0.176 0.100 0.000 b 1.000 0.200 0.167 0.313 0.136 0.263 0.032 0.077 0.107 0.036 0.050 0.000 0.000 0.167 0.333 0.048 0.000 m 1.000 0.063 0.063 0.050 0.053 0.000 0.000 0.038 0.000 0.214 0.059 0.059 0.063 0.067 0.000 0.077 f 1.000 0.429 0.091 0.045 0.192 0.125 0.071 0.217 0.000 0.050 0.050 0.111 0.056 0.313 0.063 v 1.000 0.043 0.095 0.107 0.227 0.034 0.120 0.000 0.050 0.050 0.053 0.118 0.167 0.063 t 1.000 0.500 0.296 0.192 0.700 0.185 0.200 0.087 0.042 0.263 0.150 0.087 0.000 d 1.000 0.172 0.304 0.375 0.107 0.211 0.091 0.043 0.150 0.294 0.043 0.000 s 1.000 0.520 0.414 0.500 0.069 0.185 0.103 0.069 0.034 0.185 0.037 z 1.000 0.233 0.296 0.080 0.217 0.120 0.038 0.083 0.120 0.043 ʦ 1.000 0.226 0.154 0.069 0.033 0.200 0.115 0.069 0.000 ʃ 1.000 0.037 0.115 0.208 0.077 0.038 0.208 0.042 n 1.000 0.313 0.167 0.053 0.056 0.000 0.063 l 1.000 0.467 0.000 0.000 0.048 0.200 j 1.000 0.000 0.000 0.048 0.200 k 1.000 0.462 0.313 0.063 ɡ 1.000 0.176 0.067 x 1.000 0.200 ʁ 1.000 (10) Most similar pairs12 pair 12 similarity value 1 ʦ-t 0.7 2 s-z 0.52 3 t-d 0.5 3 s-ʃ 0.5 4 p-b 0.467 4 l-j 0.467 5 k-ɡ 0.462 6 f-v 0.429 7 ʦ-s 0.414 Excluding identical consonants, which have a similarity value of 1. 12 CHAPTER 5: HYPOTHESES After calculating the similarity scale, I examined the correlation between co-occurrence restrictions and the similarity value of the first two stem consonants of the verbs. The correlation was examined on two levels: the lexical level and the phonological level (in the word-likelihood judgment experiment). The lexical analysis and the wordlikelihood judgment experiment (as well as their correlation) may have different results, which would lead to different conclusions. a. Lexicon 1 - Experiment 1: In this scenario, the same similarity-based co-occurrence restrictions are found both in the lexicon and in the speakers' judgments. Such results may indicate that similarity plays a role in co-occurrence restrictions in Hebrew. However, it will not suggest whether the influence of similarity on the grammatical system is direct, or indirect through the lexical influences. b. Lexicon 0 - Experiment 0: In this scenario, similarity-based co-occurrence restrictions are not found in Hebrew at all. Based on previous studies on OCP in Arabic (Greenberg 1950; McCarthy 1981, 1986; Frisch et al. 2004), this is the least plausible scenario. c. Lexicon 1 - Experiment 0: In this scenario, co-occurrence restrictions are found in the lexicon but not in the speakers' judgments. Such results may indicate that OCP was active in previous stages of Hebrew (many verbs in the Modern Hebrew lexicon have origins in Biblical Hebrew or in Mishnaic Hebrew), but nowadays the constraint is no longer active. d. Lexicon 0 - Experiment 1: In this scenario, co-occurrence restrictions are not found in the lexicon but are found in the speakers' judgments. Such results may indicate that OCP is not active in the lexicon, but the speakers are sensitive to it, nevertheless. A 13 plausible explanation would be that the OCP is a universal principle, which the sensitivity to it does not come from the segmental distributions in a specific language. This hypothesis would be supported by studies such as Berent (2008), who found that speakers of Korean, which does not have clusters, are nonetheless sensitive to SSG (Sonority Sequencing Generalization) violations. 14 CHAPTER 6: THE LEXICAL ANALYSIS The second part of the study analyzes the Hebrew verbal lexicon, based on Frisch et al.'s (2004) model. Unlike Frisch et al.'s analysis of Arabic, the current study examines not only homorganic consonants, but also every other possible combination of consonants. The list of verbs is taken from Even-Shoshan dictionary (edition 1970 with completions from 1983), and I used Barkali (1964) for full paradigms. 6.1 Design The study focuses on two verb classes (binyanim): kal (CaCaC) and pi'el (CiCeC). I chose these classes since they show different behaviors throughout the paradigm: while in pi'el C1-C2 are separated by one vowel throughout the inflectional paradigm, in kal the future paradigm gives rise to adjacent C1-C2 (see Appendix C for sample paradigms). Thus, it is possible to examine whether this difference in distance has an impact on the results. The analysis was conducted from a synchronic point of view, with the aim of comparing its results with the psycholinguistic experiments results. Thus, I analyzed only regular verbs (shlemim, see Zadok 2012), in which all three-stem consonants appear synchronically throughout the paradigm. Therefore, I excluded from the analysis the glottals (ʔ, h), v (orthographic: va"v; historical: w, synchronic: v) and j. For example, the verb ʃama(ʕ) 'to hear' historically ended with a ʕ and traditionally is considered as part of the regular verbs. However, nowadays final ʕ is omitted, so the verb is in a template of CaCa. Along the lines of Zadok (2012), this verb is not part of the regular verbs, and therefore was omitted from the analysis. In addition, consonants that have undergone a historical change are considered by their synchronic status. Thus, historical 15 tʕ is considered as t; historical q as k; historical ħ as x. Overall, 779 verbs in kal and 678 verbs in pi'el were analyzed.13 The analysis takes into account paradigms (and not only stems), such that each verb appears in three forms drawn from the past, present and future paradigms (all forms are in 3rd, singular, masculine). In this way, alternations throughout the paradigm can be considered, including differences in the distance between C1 and C2 (e.g. ʃamaʁ 'he saved' – C1VC2, jiʃmoʁ 'he will save' C1C2), and the spirantization of b,p, and k to v,f, and x respectively (e.g. katav 'he wrote', jixtov 'he will write'), see Appendix C for sample paradigms. For example, consider the pairs d-ʃ and b-ʃ. Each pair has only one verb in pi'el: diʃen 'to fertilize' for d-ʃ and biʃel 'to cook' for b-ʃ. However, due to spirantization alternations, d-ʃ has three occurrences in the lexicon: diʃen-medaʃenjedaʃen 'to fertilize Past-Present-Future', while b-ʃ has only one: biʃel 'to cook Past'. The present and future forms, mevaʃel and jevaʃel respectively, contain v instead of b due to spirantization, and thus contribute to the pair v-ʃ. After selecting the relevant verbs, I counted how many forms there were for each C1-C2 pair. For example, consider the pair d and m: 24 forms in the tested lexicon begin with this pair: 9 forms in kal (6 for d-m and 3 for m-d), and 15 forms in pi'el (6 for d-m 13 A few comments are addressed: a. In binyan kal, future tense, an epenthetic vowel may be inserted after a synchronic x that historically originated in ħ, for example jaxaʃov~jaxʃov 'he will think'. However, synchronically, speakers tend not to epenthesize a vowel in these cases (i.e. stick to the standard form), and evidence for variation between the two forms appears even in the Bible (e.g. taħbol~taħavol 'you ms. will take as pledge', Exodus, 22;25, Deuteronomy, 24;17, respectively). Therefore, I included these forms in the analysis. b. Verbs in pi'el with C2 ʁ have (normatively, at least) a vocalic pattern of CeCeC (e.g. seʁek 'he combed'), and not the standard CiCeC, due to historical changes. Since it is plausible to assume that the different vocalic pattern does not influence the similarity between C1-C2, I included these forms in the analysis. c. I included in the analysis verbs in kal with C1 n, although in some of these verbs the n is deleted in the future form, for example nafal-jipol (and not *jinpol; p~f alternation due spirantization) 'he fell\will fall' (respectively). In these verbs, only past and present forms were taken into account. 16 and 9 for m-d). For example, the triplet madad-moded-imdod 'to measure Past-PresentFuture' represent three instances. Next, I compared the observed (O) results to the expected (E) ones (O/E), based on consonant frequencies, in order to examine what (if any) the co-occurrence restrictions on C1-C2 are. According to previous studies on OCP in Arabic (Greenberg 1950, Frisch et al. 2004 among others), there is a solid basis to assume that some restrictions will be shown in the Hebrew lexicon as well. After calculating the O/E ratio, I compared the results to the similarity model, in order to examine if a correlation can be found between co-occurrences and the similarity values. Two questions were asked: a. Observed vs. Expected (O/E): Are there any differences between the observed and the expected occurrences of each consonant pair in the lexicon? In other words, is the number of occurrences of each pair similar to what would be obtained if the lexicon were random? b. Correlation with the similarity scale: Is there any correlation between the occurrences in the lexicon and the similarity scale? 6.2 Results 6.2.1 Observed vs. Expected (O/E) First, in order to look for differences between the observed and expected cooccurrences, a chi-square test was conducted for each verb class (binyan) separately (a full list of the occurrences appears in Appendix D). In one calculation, the order of the consonants was taken into account (e.g. d-t was calculated separately from t-d) and in the other the order was not inserted as a factor (e.g. d-t and t-d were calculated together as one item). In addition, the tests took into account the frequency of each tested consonant in the corpus (a full list of the frequencies appears in Appendix E). Thus, the 17 expected results refer to what would be expected if the single consonants were combined to pairs of C1-C2 randomly. The results show highly significant differences between the observed and expected, in all the tested cases: binyan kal: with no consideration of order: χ2 = 970.24, p < 0.0001; including consideration of order: χ2 = 1069.47, p < 0.0001; binyan pi'el: with no consideration of order: χ2 = 767.8215, p < 0.0001; including consideration of order: χ2 = 912.03, p < 0.0001. These results show that there is a gap in the lexicon between the observed and expected consonant co-occurrences. Based on previous studies (see §2), it is plausible to assume that similarity is one of the factors that causes this gap, and the next sub-sections test this assumption. 6.2.2 Observations Next, a few interesting observations can be made by looking at the bottom of the occurrences list. Since a significant gap between observed and expected was found also in the list that do not take order and binyanim into account, the following sub-section deals with the combining list. Thirty pairs of consonants do not appear in the lexicon at all. Table (11) presents them and their similarity values on the similarity scale. 18 (11) No occurrences at the lexicon (in brackets: place on the list, out of 66)14 Pair Similarity Pair Similarity Pair Similarity b-b 1 z-z 1 d-z 0.304 (11) d-d 1 ʦ-t 0.7 (1) m-b 0.2 (23) f-f 1 s-z 0.52 (2) m-p 0.2 (23) ɡ-ɡ 1 t-d 0.5 (3) ʁ-l 0.2 (23) p-p 1 b-p 0.467 (4) b-f 0.167 (28) ʁ-ʁ 1 k-ɡ 0.462 (5) p-v 0.167 (28) s-s 1 f-v 0.429 (6) m-f 0.063 (51) ʃ-ʃ 1 ʦ-s 0.414 (7) m-v 0.063 (51) t-t 1 b-v 0.313 (10) ʦ-ʦ 1 n-l 0.313 (10) v-v 1 p-f 0.313 (10) A few observations can be made. First, it is notable that the most similar pairs indeed do not appear in the lexicon. Identical consonants tend not to co-occur (except for five pairs, which will be discussed in the next paragraph), and places 1-7 in the most similar pairs (except for one) do not appear in the lexicon as well. Note that the pair sʃ is also ranked third in the similarity scale with a similarity value of 0.5. It has nine occurrences in the lexicon, three verbs in three tenses each, from which only one verb is in use (but with low frequency) in Modern Hebrew (ʃisef 'to slit someone's throat').15 Five identical consonant pairs have more than zero occurrences in the tested lexicon, summarized in table (12). 14 15 Identical consonants have a similarity value of 1, and are not considered in the numbering. The other two verbs are ʃasaf 'to split (Middle Ages Hebrew)' and ʃasas 'to despoil (Biblical Hebrew)'. 19 (12) Identical consonant pairs with occurrences in the lexicon Pair Occurrences Different Verbs Details k-k 1 1 One form – normative: kikev 'to star', colloquial: kixev. l-l 3 1 lilev-melalev-jelalev 'to strengthen with a palm branch Past-Present-Future' – unused in Modern Hebrew. n-n 3 1 nines-menanes-jenanes 'to make smaller PastPresent-Future' – unused in modern Hebrew. m-m 6 2 mimen-memamen-jemamen 'to finance PastPresent-Future' and mimeʃ-memameʃ-jemameʃ 'to realize Past-Present-Future' – both are in used in Modern Hebrew. x-x 15 9 All of them contained historically one spirantized k (x) and ħ (in Modern Hebrew – x); therefore, the identical x-x do not appear throughout the paradigm. They are used very rarely or not at all in Modern Hebrew (for example xaxaʁ 'to lease'). 16 Overall, there are only two verbs with originally identical C 1-C2 that are in use in Modern Hebrew, both with m: mimen-memamen-jemamen 'he financed Past-PresentFuture' and mimeʃ-memameʃ-jemameʃ 'he realized Past-Present-Future'. The others are not part of the commonly used lexicon (at least some of them for semantic reasons; lilev is not used since it is not customary to use a palm branch as a strengthening device nowadays). The largest group of verbs with identical C1-C2 contains x in both positions. However, all these cases are a combination of a historical ħ and spirantized allophone The Hebrew orthography shows evidence to the historical change: ‫ ח‬stands for the historical ħ and ‫כ‬ stands for k, or for x which is a historical allophone of k. 16 20 of k. Hence, historical reasons are responsible to the identical C1-C2 in the synchronic lexicon; originally, C1 and C2 were different from each other in these verbs. Another interesting finding is the verb kikev 'he starred'. This verb is widely used in Modern Hebrew, but speakers pronounce it kixev. It is plausible to assume that this pronunciation is influenced by several factors: first, this verb is strongly related to the noun koxav 'a star', and in fact, the first four discussed forms are denominatives (also lilev-lulav 'palm branch', nines-nanas 'midget', mimen-mamon 'money'). Thus, under word-based morphology, the verb was derived from the word koxav itself, and the speakers want to keep maximum faithfulness between the noun and the verb (see Aronoff 1976 and Bat-El 1994). Second, changing the second k into x reduces the similarity between the two. The change is possible because it would not cause a change in orthography (the letter ‫ כ‬allows both consonants, k and x), and because x is a phoneme in Modern Hebrew and can appear in a non-spirant position (i.e. a position in which there are no conditions for spirantization). The results so far show a correlation between similarity and co-occurrence restrictions. Nonetheless, it can be noted that all the pairs that do not appear in the lexicon (except for ʁ-l, which will be discussed in the next paragraph), also share a major place of articulation. Therefore, the co-occurrence restriction can be connected to OCP-Place violation, as was suggested for Arabic and for Semitic languages in general (McCarthy 1979, 1981, 1986; Frisch et al. 2004 among others). Moreover, the list of zero occurrences also contains homorganic pairs that are rated low on the similarity scale, for example m-f and m-v (both 0.063, place 51 from 66), a finding that strengthens the claim regarding the major role of place features in the co-occurrence restrictions. 21 The pair ʁ-l acts like homorganic pairs that do not appear in the lexicon. Since the Hebrew rhotic is considered uvular and not coronal (Bolozky and Kreitman 2007), this finding is slightly surprising, and can be related to the special status of the Hebrew rhotic. Note that the pair ʁ-x, which share a major place of articulation of dorsals, has 72 occurrences in the lexicon. Both ʁ-x and ʁ-l have a similarity value of 0.2 in the similarity model, a fact that adds more to the puzzle.17 To conclude, the 30 pairs that have zero occurrences in the tested lexicon suggest that similarity plays a role in co-occurrence restrictions. Almost all the identical C1-C2 and the most similar pairs (according to the model) do not appear in the lexicon. However, it is notable that all the pairs (except for l-ʁ) that do not appear in the lexicon are homorganic. This suggests that place features have a special role in the cooccurrence restrictions. 6.2.3 Correlation between the Lexicon and the Similarity Scale Next, a Zero-Inflated Poisson Regression model was built in order to test statistically the correlation between the lexicon and the similarity scale. The similarity factor in the model is expected to be negative, since the more consonants are similar to each other (their similarity value is closer to 1), the less they are expected to co-occur in the lexicon. The model shows that the similarity factor is significantly negative in all cases: binyan kal: Estimate = -0.112, SD = 0.023, p < 0.0001; binyan pi'el: Estimate = -0.147, SD = 0.021, p < 0.0001; kal and pi'el together: Estimate = -0.092, SD = 0.014, p < 0.0001.18 17 There is evidence that the Biblical Hebrew rhotic is dorsal (it behaves like pharyngeals and glottals, e.g. cannot undergo gemmination and cause vowel lowering). However, some studies suggest that the Biblical rhotic has two variants – coronal and dorsal – and in some Mizrahi Jews communities the rhotic is pronounced as coronal (Blau 2010). 18 For simplicity of the statistical calculations, in the calculations for pi'el and for the sum of kal and pi'el only the second consonant and the similarity value were taken into account. 22 The similarity factor in these models indicates the correlation between the appearance in the lexicon and the similarity scale: the more two consonants are similar to each other, the less their chances to co-occur as C1-C2 in a Hebrew verb. 6.2.4 Conclusion The lexical analysis tested pairs of stem consonants C1-C2 in the Hebrew verbal lexicon, and shows a significant difference between observed and expected (O/E) cooccurrences. Assuming that languages are systematic, this result suggests that there are factors that shape the lexicon and impose restrictions. Statistical models suggest that similarity (based on Frisch et al.'s 2004 model) is one of these factors. The next chapter tests the role of similarity in co-occurrence restrictions from a synchronic point of view, focusing on the phonological system of Hebrew speakers. 23 CHAPTER 7: THE EXPERIMENTS In chapter 6, I described the lexical analysis, which looked for similarity based cooccurrence restrictions. The results show that the gaps in the lexicon are found in correlation to the similarity scale. However, a lexical analysis cannot supply a complete account for co-occurrence restrictions. First, this kind of analysis can only highlight what exists, and not ask directly about what is absent. Second, the tested lexicon represents an upper bound of the Hebrew speaker's vocabulary (not all speakers are familiar with all forms in the lexicon), thus in any case it cannot directly represent the phonological knowledge of a particular Hebrew speaker. Third, lexicons have lots of exceptions for example due to historical residue. In order to complete the picture, an experiment on nonce-verbs in Hebrew was conducted. In the lexical decision experiment the participants were asked to make a lexical decision about verbs and non-verbs, and in the word-likelihood judgment experiment the participants were asked to give word-likelihood ratings for nonce-verbs in Hebrew. 7.1 The Lexical Decision Experiment Following the observation that C1 and C2 obey the OCP-Place in Semitic languages (Greenberg 1950; McCarthy 1981, 1986), I conducted a psycholinguistic experiment that examined OCP-Place effects in the Hebrew verbal system (Yeverechyahu 2012). The results indicate that Hebrew speakers are sensitive to the OCP-Place in the verbal system, and that sharing place features (i.e. homorganic consonants) is a sufficient condition for the violation. 24 7.1.1 Participants 33 participants took part in the experiment (18 females, 15 males). All of them were native Hebrew speakers who were born in Israel, between ages 21 and 29 (mean age 25, SD=2.21). None of them had studied Linguistics academically. Two additional participants whose mean RTs were greater than the total participants' mean in more than two standard deviations were discarded from the analysis (Mtotal=1206ms, SD=325). 7.1.2 Stimuli The stimuli were 30 Hebrew verbs and 30 nonce-verbs in a legal Hebrew verb template. All items were in binyan pi'el, 3rd person, masculine, singular, in past tense, namely in the template CiCeC (e.g. ɡisem nonce-verb). In all verbs, C2 and C3, as well as C1 and C3, did not share place of articulation or manner of articulation, in order to focus on the OCP effect in C1 and C2 alone. In addition, the stimuli included only regular verbs (shlemim, see Zadok 2012 and §6.1) in which all three stem consonants appear synchronically throughout the paradigm, and nonce-verbs that look like regular verbs. The stimuli were selected with the aid of Barkali's (1964) Hebrew verbs dictionary. The 30 nonce-verbs were divided to five groups, as follows (a full stimuli list appears in Appendix F): a. Non-shared features: C1 and C2 were different in place, manner and voice (e.g. ɡisem);19 b. Place: C1 and C2 shared the place of articulation (coronals) and differed in manner and voice (e.g. disem); c. Place and Manner: C1 and C2 shared place and manner of articulation (coronal fricatives-stridents) and differed in voice (e.g. zisem); 19 Sonority was not taken into account. 25 d. Place and Voice: C1 and C2 shared place of articulation and voice (voiceless coronals) and differed in manner (e.g. tisem); e. Identical C1-C2: C1 and C2 are identical (e.g. sisem). In all the verbs that shared one or more features, the shared features were constant (coronal for place, stridency for manner, and voiceless for voice).20 Referring to Frisch et al.'s (2004) model, this was done in order to avoid differences in the similarity between different groups (i.e. two coronals may be less similar to each other than two labials, since there are more natural classes of coronals in the language than of labials). Since the study focuses on auditory similarity and not visual similarity, all the stimuli were presented auditorily. This was done in order to focus on the auditory channel, and avoid orthographic or visual influence. The stimuli were recorded by a 30year-old male native speaker of Hebrew. 7.1.3 Procedure The experiment was designed using the E-prime software (2.0). The participants sat in front of an Asus Eee mini-laptop equipped with earphones, and heard different stimuli. They were asked to determine whether each stimulus was an existing Hebrew verb. "Existing verb" responses were given by pressing 1 and "nonce-verb" responses by pressing 0, such that opposing responses were made using different fingers and hands. A short training block was passed at the beginning of the experiment to ensure that the participants understood the task. The training block contained ten stimuli (five existing verbs, five nonce-verbs), and the participants were given feedback (a smiley or a disappointed face icon) following each response. 20 Except for two nonce-verbs with identical segments, which were affricates and not fricatives (ʦ), in order to reach six different nonce-verbs. 26 The order of the items in the experiment and in the training was randomized across subjects. Accuracy and reaction times (hereinafter: RTs) measured from stimulus onset were collected.21 Each subject was tested individually, and each session lasted approximately five minutes. 7.1.4 Results The results suggest that the subjects are sensitive to the OCP-Place constraint. Accuracy for all nonce-verbs was extremely high (99.19% correct answers), errors were excluded from the RT analysis. RTs for the non-homorganic C1-C2 among the nonce-verbs were significantly greater than the RTs for the homorganic C1-C2 (t(34)=5.99, p<0.0001), as can be seen in Figure (13). Thus, the subjects needed more time to decide that nonhomorganic C1-C2 nonce-verbs were not part of their lexicon. This finding suggests that when C1 and C2 are homorganic, the gap in the lexicon is systematic and predicted by the OCP-Place constraint. The OCP-Place provides the subjects with a cue that these verbs are less likely to be Hebrew verbs. (13) Mean RT: homorganic\non-homorganic C1-C2 (ms) 21 An ANOVA test reveals that the sound samples were not statistically different in duration (F(5,24)=2.36, p=0.08). 27 However, the differences among the five stimulus groups which violated OCPPlace to different degrees were not significant (F(3,128)=0.61, p=0.61). It seems that sharing the place feature, namely homorganic C1 and C2, is a sufficient condition to determine quickly that the stimulus is a nonce-verb. This finding is compatible with McCarthy's (1979, 1981, 1986) claim that the shared place of C1 and C2 causes OCPPlace violation in Semitic stems. 7.1.5 Discussion The lexical decision experiment sheds light on the role of the OCP in the Hebrew verbal system, and opens the door for further research on the topic. Particularly, the experiment focuses on homorganic consonants (as in Frisch et al. 2004), and it raises the question of whether co-occurrence restrictions will also be observed among non-homorganic consonants that do share some features (voice or manner). For example, in the nonceverb dibem, C1 d and C2 b share voice ([+voice]) and manner (stops), but not place (coronal and labial, respectively). What will be the effect of the similarity between C 1 and C2 in this case? Second, the experiment focused on division to place, manner and voice, without looking into phonological features (as [±son], [±cont] etc.) and natural classes. Since co-occurrence restrictions may be phonological by nature (and not purely phonetic), it is interesting to address the issue from a more phonological point of view, which takes into account phonological features and natural classes. Frisch et al.'s (2004) model is based on such properties, and therefore will be suitable for analysis the results. 7.2 Word-Likelihood Judgment Experiment In light of the results of the lexical decision experiment, a word-likelihood judgment experiment was conducted. The aim of this experiment is to broaden the lexical decision experiment by considering any C1-C2 combination, and comparing the results to the similarity scale. 28 7.2.1 Participants 138 participants participated in the experiment (79 females, 59 males). All of them were Hebrew native speakers who were born in Israel, between ages 20 and 40 (mean age 27, SD=3.84). 14 of them were BA Linguistics students, but none of them had taken advanced courses in morphology or phonology. 7.2.2 Stimuli The stimuli in the experiment were 331 Hebrew nonce-verbs in binyan kal, 3sg past, in the template of CaCaC (e.g. dadam). All of them are non-existing verbs in Hebrew. However, they were put in a verb template in order to cue the participants to consider them as potential Hebrew verbs, so I expected that phonological factors (and not morphological) would affect the participants' judgments. In order to make the experiment in a reasonable length, I focused only on one verb class, binyan kal. Testing all the consonant combinations in both kal and pi'el would have made the experiment too long, and would have made it harder for the participants to be focused during the entire session. Since both kal and pi'el showed sensitivity to similarity effects in the lexical analysis, I decided to focus in the current study on kal alone. I preferred kal over pi'el since the spirantization in the kal's paradigm (see Appendix C) allows us to test more consonants, as there are no f or v in the base forms of binyan pi'el.22 As with the lexical research, the experiment focuses on the similarity between C1 and C2, while ignoring C3. The stimuli's stem consonants were selected in the following way: 22 Except for several nominatives, for example fiʃel 'to screw up' from the word faʃla 'mistake' (borrowed from Arabic). 29 C1 and C2: As in the lexical analysis, the stimuli included all Hebrew consonants in the shlemim paradigm (Zadok 2012), with the correct spirantization restrictions (see §6.1). Thus, the consonants b and p appeared as stops in C1 and as fricatives v and f in C2 (respectively). For example, stimuli contained nonce-verbs such as bafat, but not fabat. k and x could appear in both positions: in C1 (non-spirant position) k represents alternating k or non-alternation k (historical q), and x represents only non-alternating x (historical ħ). In C2 (spirant position), k represents only non-alternating k, and x represents alternating x or non-alternation x. Overall, 17 different consonants were examined, the same as in the similarity scale and the lexicon analysis. 23 In this way, the relations between the similarity scale, the lexical research, and the experiment results can be tested. In order to make the experiment shorter, the stimuli contained each consonant pair in only one order. For example, the pair k and l appeared as k-l and not as l-k. The order of each pair was chosen to be the one in which there are more verbs, as found in the lexical research. For example, there are 17 occurrences of k-l and 10 occurrences of l-k, and therefore the stimuli in the experiment contained k-l and not l-k. In many cases, the more frequent order in the lexicon was the order in which C1 is less sonorous than C2. This is an interesting point that may indicate the role of sonority in co-occurrence restrictions, and it opens a door for further research. Based on this observation, for pairs that have the same number of occurrences in the lexicon in each order (usually 0), I chose the order in which C1 was less sonorous than C2. C3: The role of C3 and its similarity to C1 and C2 was not tested in this study. Nevertheless, it is plausible that C3 influences co-occurrences as well, and therefore it 23 Except for j, which was considered in the similarity scale since it is a Hebrew consonant, but was eliminated from the lexical analysis and the experiment since it causes changes in the paradigm. 30 should be neutralized. In order to control for the effect of C 3, each C1-C2 pair was combined with three different C3s, such that C3 would be the least similar to C1-C2. Based on previous studies, the control on C3 was based on place of articulation (see McCarthy 1979, 1981, 1986; Frisch et al. 2004 among others for OCP-Place violations in Semitic languages). C3s were selected in the following way: a. If C1 and C2 did not share a major place of articulation (labial, coronal or dorsal), C3 was in the third place of articulation. Note that C3 could not be b or p since it is a spirant position. For example, C3s for the pair b (labial) and d (coronal) were dorsals. b. If C1 and C2 shared a major place of articulation, C3 was chosen in the following way: if C1 and C2 were labials, the chosen C3s were dorsals; if C1 and C2 were coronals, the chosen C3s were labials; if C1 and C2 were dorsals, the chosen C3s were coronals. c. When this strategy did not result in enough suitable candidates for C3 (see next paragraph for reasons for candidate elimination), I chose a different place of articulation, such that C2 and C3 would not be at the same place. No stimuli were part of the Hebrew verbal lexicon. Moreover, they did not share stem consonants with existing verbs in different classes (binyanim), for example nonceverbs such as baʁaʃ (shared stem consonants with hivʁiʃ 'to brush X') and bakaʃ (shared stem consonants with bikeʃ 'to request') were not included in the experiment. Stimuli that obeyed these conditions but formed existing words in Hebrew (e.g. paɡaz 'shell (projectile)') were also excluded. In cases in which three different options for C3 could not be found, one form was repeated in order to obtain three stimuli for each C1-C2 pair. Overall, the stimuli contained 147 tested consonant pairs C1-C2, forming 331 different nonce-verbs, and 31 441 stimuli including repetitions. In order to make the experiment shorter, each participant was exposed to a random list of 49 stimuli (in random order as well), in which each tested pair of segments was presented at most once. The nonce-verbs were inserted into frame sentences in the template of male proper name + verb + an animal. For example, xen baʦaɡ et ha-tanin 'Chen baʦaɡ-ed (nonce-verb) the crocodile'. Inserting the nonce-verbs into frames had two main reasons: first, the template of CaCaC is used both for verbs and nouns, for example paɡaz 'shell (projectile)', zamaʁ 'singer'. Combining the nonce-words in sentences ensures that the participants would refer to the nonce-word as a verb and not as a noun. Second, the sentences were intended to make the experiment more interesting than presenting verbs in isolation, and I hoped that it would make participants more attentive to the task. The fixed template was aimed at controlling for semantic effects on the judgments. The stimuli were recorded by a 26-year-old female native speaker of Hebrew (the author), a full stimuli list appears in Appendix G. 7.2.3 Procedure The experiment was conducted online via the Qualtrics website (www.qualtrics.com). Each participant heard a random list of 49 sentences, and was asked to rate the wordlikelihood of each nonce-verb. The ratings were on a scale of 1 to 7, which was defined as follows (here translated to English), based on Frisch and Zawaydeh (2001): (14) The rating scale 1 – No. The verb sounds terrible; it cannot be a valid verb in Hebrew. 3 – Not likely. The verb sounds strange; I doubt it can be a valid verb in Hebrew. 5 – Maybe. The verb sounds a bit strange, but it can possibly be a valid verb in Hebrew. 7 – Yes. The verb sounds good; it can be a valid verb in Hebrew. 2, 4 and 6 are found between these guidelines. 32 The instructions were shown after each question. Since I hypothesized that word-likelihood would be gradient rather than dichotomous, I chose to use a scale and not a yes\no decision in order to allow participants to express small differences in judgments (see Kawahara 2011 for a different approach). As in the lexical decision experiment, this experiment also focuses on auditory similarity and not visual similarity. Therefore, all the sentences were presented auditorily, in order to focus on the auditory channel, and avoid orthographic or visual influence. The experiment took approximately ten minutes. 7.2.4 Results The results were calculated as follows: first, a pre-analysis was done in order to check whether the effect of C3 was neutralized. Second, the correlation between the results and the previous parts of the study (the lexicon analysis and the similarity scale) was tested. 7.2.4.1 Pre-analysis: The Effect of C3 Since the study focuses on C1 and C2, C3 was carefully chosen in order to reduce its influence on the results. Hence, before analyzing the effects of C1 and C2, I checked whether C3 in each triplet of stimuli influenced the participants' judgments. For example, I checked whether the ratings for bafat, bafad and bafan were significantly different. In order to test this possibility, a Kruskal-Wallis test was run on the 119 cases in which different C3s were combined to the tested C1-C2 pairs. Out of the 119 cases, only 19 were found significant, namely there were 19 triplets in which combining different C3s affected the results (see Appendix H for a full list). Note that statistically, even if the assumption that C3 did not affect the results was true, we would expect an error of 5%, namely six cases in which C3 affected the results. 33 Many of the anomalous stimuli are nonce-verbs that are very similar to real Hebrew verb, usually different only in the voicing of one consonant. For example, the nonce-verb baʁas is similar to the Hebrew verb paʁas 'to spread/slice', the nonce-verb baʦax is similar to paʦax 'to start' and the nonce-verb padak is similar to badak 'to check'. This finding implies another type of similarity that affects judgments: Not only similarity between consonants is relevant to grammar, but also similarity on a higher level, between words. This finding is compatible with Frisch and Zawaydeh (2001) findings in Arabic. Next, I converted the average rating for each triplet into a single value. The 19 cases in which a significant difference was found inside the triplets were eliminated from the calculations. 7.2.4.2 Observations A few interesting observations can be made by looking at the bottom of the results list, namely on the items that were rated the lowest on the word-likelihood scale (mean ratings for all tested consonant pairs appear in Appendix I). Table (15) presents the consonant pairs that received the ten lowest ratings, their similarity value on the similarity scale and their number of occurrences in the lexicon. 34 (15) the ten lowest ranked consonant pairs (in brackets: ranking on the similarity list, out of 66)24 Pair Rating Similarity Occurrences 1 s-s 2.05 1 0 2 ʦ-s 2.21 0.414 (7) 0 3 k-ɡ 2.27 0.462 (5) 0 4 ʃ-ʃ 2.29 1 0 5 z-z 2.51 1 0 6 t-d 2.53 0.5 (3) 0 7 t-ʦ 2.60 0.7 (1) 6 (2 verbs) 8 b-v 2.65 0.313 (10) 0 8 ʦ-ʁ 2.65 0 (66) 51 9 s-z 2.66 0.52 (2) 0 9 k-k 2.66 1 1 (kikev) 10 ɡ-ɡ 2.67 1 0 A few observations can be made. All the items (except one) are rated among the ten highest similar pairs on the similarity scale, share the major place feature, and have zero, or almost zero, occurrences in the lexicon. The correlation between the parameters is salient. The only pair among the most similar tested pairs that does not appear among the ten lowest pairs in the experiment is ʃ-s (similarity value: 0.5), with a mean result of 3.15 (placed 22). The only pair that does not follow this generalization is ʦ-ʁ, which has a similarity value of 0 and 51 occurrences in the lexicon. An explanation for this exception can be made by examining the chosen C3s for this pair: the three items were ʦaʁaʃ, ʦaʁas and ʦaʁaz, all C3s are stridents as C1 ʦ. A key condition in this experiment was that all stimuli would be nonce-verbs, and since the pair ʦ-ʁ has a large number of occurrences in the lexicon, there are not many options left for nonce-verb stimuli. Along 24 Identical consonants had similarity value of 1, and were not considered in the ranking. 35 the lines of the scheme presented in §7.2.2, all the chosen C3s were stridents. Therefore, the low ratings for this pair could have been influenced by the similarity between C1 and C3, and not only by the relations between C1 and C2. Nevertheless, not all the most similar pairs nor pairs with zero occurrences in the lexicon appear at the top of the experiment results. Looking at the other results reveals that all the pairs with identical C1-C2 are rated among the lowest 20. For example, l-l, m-m and n-n are rated 11th, 12th and 13th respectively, and x-x is rated 19th.25 Note that these are exactly the identical pairs that do have occurrences in the lexicon (together with k-k), and x-x had the highest number of occurrences, 15, affected by historical reasons (see §6.2.2 for discussion). This finding can show the correlation between the lexicon and the experiment, but not to point to the source of the influence. It could be that the small-but-not-zero occurrences in the lexicon affect the speakers' phonological system, or that these identical pairs show (for some reasons) fewer restrictions, and therefore are more flexible both in the lexicon and in the speakers' judgments. Nonetheless, the differences between the pairs of identical C 1-C2 are small, and it can be concluded that all of them show co-occurrence restrictions in the speakers' judgments. Next, I examined the other pairs that had zero occurrences in the lexicon. Table (16) summarizes the comparison. 25 The other pairs with identical C1-C2 were eliminated from the results according to the pre-analysis of the effect of C3 (§7.2.4.1). 36 (16) Other pairs with zero occurrences in the lexicon (in brackets: the rank on the relevant scale)26 Pair Rating p-f 2.81 (16) b-f 2.93 (18) d-z 3.00 (20) b-m 3.02 (21) p-v 3.21 (24) n-l 3.42 (27) m-f 3.43 (28) m-v 3.55 (34) ʁ-l 3.70 (38) It can be seen that not all these pairs were rated low in the judgment task. There are pairs that appear among the second or third group of ten, but also pairs that rated 34 and 38. Note that m-f and m-v, that rated low on the similarity scale (0.063 (51)), do not show strong co-occurrence restrictions in the judgments task, as the pair ʁ-l, in which C1 and C2 do not share a place of articulation. Hence, in these cases the similarity factors are stronger than the occurrences in the lexicon. 7.2.4.3 Comparisons to the Scales Next, a statistical analysis was done in order to examine the correlation between the experiment results on the one hand, and the lexicon analysis and the similarity scale on the other hand. Comparison to the similarity scale: An ordered logistic regression model was built in order to test the influence of the similarity factor on the results. This factor is assumed to be negative, since the greater the similarity between two consonant is, the participants are assumed to give lower ratings to the nonce-verbs. Indeed, the similarity 26 Not including the pairs that were eliminated due to spirantization (b-p, f-v) or during the pre-analysis of the effect of C3. 37 factor is negative and its influence is significant (Estimate = -1.79, SD = 0.1, p < 0.0001). This result shows that the more similar the consonants are (their similarity value is closer to 1), the lower the ratings the participants will give to the wordlikelihood of the nonce-verb. Comparison to the lexical analysis: An ordered logistic regression model was built in order to test the influence of the frequency factor on the results. This factor was assumed to be positive, since the more frequent the consonant pair in the lexicon is, the participants are assumed to give higher ratings to the nonce-verb. Indeed, the frequency factor is positive and its influence is significant (Estimate = 0.04, SD = 0.004, p < 0.0001). This result shows that nonce-verb containing frequent pairs of consonants receive higher ratings of word-likelihood. These results show a strong correlation between the word-likelihood ratings and the similarity scale and with the lexical analysis. Since a correlation was found between the lexical analysis and the similarity scale as well, it is not surprising that the participants' ratings correlate with both of them. 38 CHAPTER 8: GENERAL DISCUSSION The study examines the co-occurrence restrictions in the Hebrew verbal system, focusing on the contribution of similarity between consonants to these restrictions. The phenomenon was tested on two levels: lexical analysis and psycholinguistic experiments. The results suggest that there are co-occurrence restrictions on stem consonants C1 and C2, both in the lexicon and in the speakers' phonological system. The similarity factor was found to be significant in all the tested cases. The similarity model used in this study is based on Frisch et al.'s (2004) model, originally proposed for Arabic. The model is built with a phonological orientation: it is based on phonological natural classes and phonological features. Note that the model itself is not unique for linguistic similarity and can be adapted to similarity in any domain; it is the phonological features that make the similarity model language oriented. Since a correlation was found between the co-occurrence restrictions and the similarity model, the study supports the idea that phonological features can constitute a proper base for similarity calculations. However, the fact that there are articulatorybased similarity effects does not mean that acoustic factors do not play a role as well (see Mielke 2009). It would be interesting to examine the interaction between acoustic factors and phonological factors, and this is a window for further studies in the field. Previous studies on Semitic languages (Greenberg 1950; McCarthy 1981, 1986; Frisch et al. 2004 among others) demonstrated the important role of OCP-Place in cooccurrence restrictions in those languages. The current study strengthens this claim; both the lexical analysis and the experiments suggest that consonants that share the major place feature are less likely to co-occur. However, the results and the correlation with the similarity scale show that not only place feature has a role in co-occurrence restrictions, and correlations were found also between the similarity scale and 39 occurrences or judgments of pairs that do not share place of articulation. It is likely that the major place feature has a great weight in similarity, inter alia due to its high position in the feature geometry hierarchy (Clements 1985, Sagey 1986, Clements and Hume 1995, and see also Kaisse 1988 and Padgett 1995). Frisch et al.'s model does not suggest which phonological features are more important for similarity, and the current findings open a window for further research in the field. The results of the lexical analysis and the experiment suggest a correlation between the co-occurrence restrictions and the similarity scale. These correlations are statistical, and do not entail causal influence of similarity on the restrictions. The lexical analysis shows correlation to the similarity scale. Nonetheless, it is plausible to assume that other factors have influenced the lexicon as well; some of them are historical influences that do not have transparent evidence nowadays. For example, the 15 occurrences of x-x in the lexicon: historically, the origin of x is double, one is a result of historical ħ (which is assumed to have been pronounced farther backward than the synchronic x) and the other is a spirantized k. Indeed, all the 15 cases of cooccurrences are results of the historical reasons. The naïve Hebrew speakers may not know this detail, but the frequency of x in their lexicon and the cases of verbs with x in C1 and C2 are likely to affect their phonological system. Thus, the correlation between the lexical analysis and the similarity scale may suggest that similarity is one of the factors that influence the lexicon. However, it is not the only one. The speakers' word-likelihood judgments in the experiment were correlated both with the similarity scale and the lexical analysis results. It is not surprising that the experiment's results are correlated with both of them, because there is also a correlation between the similarity scale and the lexical analysis results. Therefore, the experiment 40 cannot suggest whether the influence of similarity is direct, or indirect through the lexical influences on the grammatical system (figure 17). (17) lexicon  similarity → grammatical system or: similarity → lexicon → grammatical system 41 CHAPTER 9: CONCLUDING REMARKS The study shows that there are co-occurrence restrictions both in the Hebrew lexicon and in the grammatical system of the speakers. These restrictions have a strong correlation with similarity between consonants, where the tendency is to avoid similar, close consonants. The results suggest that similarity affects the speakers' wordlikelihood judgments, but they cannot tell whether the effect is direct or indirect through the lexicon. The study opens the door to further research on similarity effects in Hebrew and in other languages. First, the current study examines the issue of similarity from a phonological point of view, and successfully shows the relevance of phonological properties to the subject. Future studies should examine the influence of each phonological feature individually, and examine which features are more important to similarity. Under the observed importance of the place of articulation, and previous studies on OCP-Place in Semitic languages (McCarthy 1979, 1981, 1986; Frisch et al. 2004 among others), there is a reason to assume that major place features have a large role in similarity. Second, the study took into account phonological features that are mostly based on articulatory factors, but not acoustic parameters per-se. However, it has been claimed that acoustic factors play a significant role in similarity as well (Kawahara 2007, Mielke 2009). Models that combine acoustic parameters (exclusively or with articulatory parameters) and test their predictions regarding similarity in Hebrew should be taken into consideration in further studies. Third, the lexical analysis does not show statistical differences between orders of consonants in the pairs (e.g. k-d vs. d-k), but a closer look reveals large differences in some pairs (e.g. k-d: 34 occurrences, d-k: 18 occurrences). It would be interesting to 42 examine possible causes to asymmetry in similarity pairs. Previous studies (Johnson 2012 for example) suggest that similarity is not symmetric, such that it is not necessary that k is similar to d in the same degree that d is similar to k. Frisch et al.'s (2004) model calculates similarity based on shared and non-shared natural classes, and thus cannot take symmetry into account. However, it is also plausible to assume that other factors cause these differences, such sonority or place of articulation. Furthermore, the current study does not show differences between the two tested verb classes (binyanim) kal and pi'el: They both showed the same tendency to avoid similar consonant pairs in the lexicon. However, there is a solid basis to assume that there are differences in co-occurrence restrictions between them. For one, the proximity between C1 and C2 is different between the templates throughout the paradigms: in pi'el they are separated with a vowel through the different tenses and persons, while in kal there are forms in which they are adjacent (e.g. future tense: jixtov 'he will write'). Since previous studies suggest that proximity plays a role in co-occurrence restrictions (Rose 2000 among others), and under the assumption that speakers have access to the paradigms and not only to the base forms, differences in restrictions are expected. Since the lexicon did not show significant differences, examination of this issue from a psycholinguistic point of view in needed. In addition, the current study focuses only on some verb classes and does not deal with nouns at all. It would be interesting to compare co-occurrence restrictions between nouns and verbs, and between two types of Hebrew nouns: Semitic nouns that have similar structure to verbs, with stems and templates (e.g. ʃmiʁa 'saving', template: CCiCa) and mono-morphemic, non-Semitic nouns (e.g. ʃulxan 'table'). It will also be interesting to expand the study to other verb classes (binyanim), and look for differences among them. 43 Finally, the study shows a correlation between the lexical analysis and the speakers' judgments. It would be interesting to further examine the influence of the lexicon on one's grammatical knowledge, and investigate the tension between universal constraints and language specific grammar. Since the current study found correlation between the similarity scale and the lexicon analysis, it cannot determine the source of the influence; a language in which there is no correlation between the lexicon and the universal constraints could shed more light on the influence of the lexicon. 44 APPENDIX Appendix A: Natural classes of Hebrew consonants j ʁ f v t d f v d s z ʦ ʃ s z ʦ ʃ [-son, COR, +strident, +ant] s z ʦ [-son, COR, +strident, +ant, +cont] s z [-son, COR, +strident, +ant, +cont, +voice] z [-son, COR, +strident, +ant, +cont, -voice] s [-son, COR, +strident, +ant, -cont] ʦ [-son, COR, +strident, +ant, -cont, -voice] ʦ [-son, COR, +strident, +ant, -voice] s [-son, COR, +strident, -ant] ʃ [-son, COR, +strident, -ant, +cont] ʃ [-son, COR, +strident, -ant, +cont, -voice] ʃ [-son, COR, +strident, +cont] s z [-son, COR, +strident, +cont, -voice] s ʃ [-son, COR, +strident, -voice] s ʦ [+son] m n l [+son, LAB] m [+son, LAB, -cont] m [+son, COR] n l j [+son, COR, +ant] n l [+son, COR, +ant, +cont] l [+son, COR, +ant, -cont] n [+son, COR, -ant] j [+son, COR, -ant, +cont] j [+son, COR, +cont] l [+son, COR, -cont] n [+son, DOR] ʁ [+son, DOR, +cont] ʁ [+son, +cont] l j [+son, -cont] m n [-son] p b [-son, LAB] p b [-son, LAB, +cont] f v [-son, LAB, +cont, +voice] v [-son, LAB, +cont, -voice] f [-son, LAB, -cont] p [-son, LAB, -cont, +voice] b [-son, LAB, -cont, -voice] p [-son, LAB, +voice] b v [-son, LAB, -voice] p f [-son, COR] t [-son, COR, +strident] j ʁ b ʦ 45 ʃ ʃ s z ʦ ʃ k ɡ x [-son, COR, +ant] t d s [-son, COR, +ant, -cont] t d ʦ [-son, COR, +ant, -cont, +voice] d [-son, COR, +ant, -cont, -voice] t ʦ [-son, COR, +ant, +voice] d z [-son, COR, +ant, -voice] t s ʦ [-son, COR, -voice] t s ʦ [-son, DOR] k ɡ x [-son, DOR, +cont] x [-son, DOR, +cont, -voice] x [-son, DOR, -cont] k [-son, DOR, -cont, +voice] ɡ [-son, DOR, -cont, -voice] k [-son, DOR, -voice] k x [-son, +cont] f v [-son, +cont, +voice] v z [-son, +cont, -voice] f [-son, -cont] z ʦ ʃ ɡ ʃ x ʦ k ɡ ʃ k x ʦ ʃ n l j ʦ n l ʃ l j x ʁ d ʦ n k ɡ s z s ʃ x p b t d [-son, -cont, +voice] b d ɡ [-son, -cont, -voice] p t ʦ k [-son, +voice] b v d z ɡ [-son, -voice] p f t s ʦ [LAB] p b m f v [LAB, -cont] p b m [COR] t d s z [COR, +ant] t d s z [COR, +ant, +cont] s z l [COR, +ant, -cont] t d ʦ n [COR, -ant] ʃ j [COR, -ant, +cont] ʃ j [COR, +cont] s z ʃ l [DOR] k ɡ x ʁ [DOR, +cont] x ʁ [+cont] f v s z [-cont] p b m t 46 j Appendix B: The similarity scale (based on Frisch et al.'s 2004 model) bl 0.000 bx 0.048 tl 0.087 kʦ 0.200 tʦ 0.700 bʁ 0.000 xl 0.048 xt 0.087 nt 0.200 bb 1.000 bj 0.000 xj 0.048 dl 0.091 pm 0.200 dd 1.000 dʁ 0.000 bn 0.050 tf 0.091 ʁl 0.200 ff 1.000 ɡl 0.000 fj 0.050 dv 0.095 ʁj 0.200 ɡɡ 1.000 ɡj 0.000 lf 0.050 px 0.100 xʁ 0.200 kk 1.000 kl 0.000 lv 0.050 sj 0.103 ʃx 0.208 ll 1.000 kj 0.000 pn 0.050 sv 0.107 ʃj 0.208 mm 1.000 mʃ 0.000 tm 0.050 bʦ 0.107 nd 0.211 nn 1.000 mz 0.000 vj 0.050 ʃd 0.107 nm 0.214 pp 1.000 nf 0.000 dm 0.053 kf 0.111 ʃf 0.217 ʁʁ 1.000 nv 0.000 nk 0.053 ɡʦ 0.115 zl 0.217 ss 1.000 pl 0.000 kv 0.053 ʃl 0.115 ʃʦ 0.226 ʃʃ 1.000 pʁ 0.000 ɡf 0.056 ɡv 0.118 zv 0.227 tt 1.000 pj 0.000 nɡ 0.056 ʃv 0.120 ʦz 0.233 ʦʦ 1.000 sm 0.000 ml 0.059 xz 0.120 pt 0.250 vv 1.000 tʁ 0.000 mj 0.059 zj 0.120 kt 0.263 xx 1.000 ʦʁ 0.000 kʁ 0.063 zf 0.125 bd 0.263 jj 1.000 xm 0.000 mf 0.063 bt 0.136 ɡd 0.294 zz 1.000 xn 0.000 mv 0.063 pd 0.143 st 0.296 bs 0.032 ʁf 0.063 kd 0.150 ʃz 0.296 ʦj 0.033 ʁn 0.063 tɡ 0.150 dz 0.304 sɡ 0.034 km 0.063 ʦn 0.154 bv 0.313 ʦv 0.034 ʁv 0.063 bf 0.167 nl 0.313 bʃ 0.036 ɡm 0.067 bk 0.167 pf 0.313 nʃ 0.037 ɡʁ 0.067 nj 0.167 pk 0.313 pz 0.037 ps 0.067 pv 0.167 xf 0.313 sʁ 0.037 sn 0.069 xv 0.167 xk 0.313 ʃɡ 0.038 ʦl 0.069 sd 0.172 bɡ 0.333 ʦm 0.038 ʦx 0.069 ɡx 0.176 ʦd 0.375 zk 0.038 ks 0.069 pɡ 0.176 ʦs 0.414 ʃʁ 0.042 ʦf 0.071 ʃt 0.185 fv 0.429 tj 0.042 pʃ 0.074 sl 0.185 kɡ 0.462 tv 0.043 bz 0.077 sx 0.185 bp 0.467 dx 0.043 mʁ 0.077 pʦ 0.192 lj 0.467 dj 0.043 kʃ 0.077 sf 0.192 ʃs 0.500 zʁ 0.043 nz 0.080 tz 0.192 td 0.500 df 0.045 ɡz 0.083 bm 0.200 sz 0.520 47 Appendix C: Sample derivations kal (3rd ms. sg.) pi'el (3rd ms. sg.) Past Present Future Past Present Future Regular Verbs ʃamaʁ ʃomeʁ jiʃmoʁ 'save' ʃimeʁ meʃameʁ jeʃameʁ 'preserve' C1: spirantization katav kotev jixtov 'write' kitev mexatev jexatev 'subscribe' C2: spirantization safaʁ sofeʁ jispoʁ 'count' sipeʁ mesapeʁ jesapeʁ 'tell' Appendix D: Lexical analysis results kal pi'el kal and pi'el bb 0 bb 0 sum 0 bb 0 bb 0 sum 0 bb 0 bb 0 sum 0 bd 8 db 4 sum 12 bd 4 db 9 sum 13 bd 12 db 13 sum 25 bf 0 fb 0 sum 0 bf 0 fb 0 sum 0 bf 0 fb 0 sum 0 bɡ 4 ɡb 4 sum 8 bɡ 0 ɡb 18 sum 18 bɡ 4 ɡb 22 sum 26 bk 2 kb 4 sum 6 bk 3 kb 21 sum 24 bk 5 kb 25 sum 30 bl 12 lb 4 sum 16 bl 4 lb 18 sum 22 bl 16 lb 22 sum 38 bm 0 mb 0 sum 0 bm 0 mb 0 sum 0 bm 0 mb 0 sum 0 bn 0 nb 3 sum 3 bn 0 nb 9 sum 9 bn 0 nb 12 sum 12 bp 0 pb 0 sum 0 bp 0 pb 0 sum 0 bp 0 pb 0 sum 0 bʁ 16 ʁb 5 sum 21 bʁ 6 ʁb 15 sum 21 bʁ 22 ʁb 20 sum 42 bs 6 sb 5 sum 11 bs 5 sb 15 sum 20 bs 11 sb 20 sum 31 bʃ 2 ʃb 8 sum 10 bʃ 1 ʃb 15 sum 16 bʃ 3 ʃb 23 sum 26 bt 6 tb 2 sum 8 bt 5 tb 9 sum 14 bt 11 tb 11 sum 22 bʦ 6 ʦb 3 sum 9 bʦ 2 ʦb 6 sum 8 bʦ 8 ʦb 9 sum 17 bv 0 vb 0 sum 0 bv 0 vb 0 sum 0 bv 0 vb 0 sum 0 bx 8 xb 11 sum 19 bx 5 xb 36 sum 41 bx 13 xb 47 sum 60 bz 6 zb 3 sum 9 bz 0 zb 9 sum 9 bz 6 zb 12 sum 18 dd 0 dd 0 sum 0 dd 0 dd 0 sum 0 dd 0 dd 0 sum 0 df 6 fd 0 sum 6 df 0 fd 2 sum 2 df 6 fd 2 sum 8 dɡ 6 ɡd 15 sum 21 dɡ 12 ɡd 15 sum 27 dɡ 18 ɡd 30 sum 48 dk 6 kd 19 sum 25 dk 12 kd 15 sum 27 dk 18 kd 34 sum 52 dl 15 ld 0 sum 15 dl 9 ld 0 sum 9 dl 24 ld 0 sum 24 dm 6 md 3 sum 9 dm 6 md 9 sum 15 dm 12 md 12 sum 24 dn 0 nd 10 sum 10 dn 3 nd 9 sum 12 dn 3 nd 19 sum 22 dp 3 pd 0 sum 3 dp 6 pd 1 sum 7 dp 9 pd 1 sum 10 dʁ 12 ʁd 9 sum 21 dʁ 6 ʁd 6 sum 12 dʁ 18 ʁd 15 sum 33 ds 0 sd 9 sum 9 ds 3 sd 9 sum 12 ds 3 sd 18 sum 21 dʃ 3 ʃd 9 sum 12 dʃ 3 ʃd 9 sum 12 dʃ 6 ʃd 18 sum 24 48 kal pi'el kal and pi'el dt 0 td 0 sum 0 dt 0 td 0 sum 0 dt 0 td 0 sum 0 dʦ 0 ʦd 3 sum 3 dʦ 0 ʦd 6 sum 6 dʦ 0 ʦd 9 sum 9 dv 8 vd 4 sum 12 dv 0 vd 8 sum 8 dv 8 vd 12 sum 20 dx 15 xd 11 sum 26 dx 9 xd 18 sum 27 dx 24 xd 29 sum 53 dz 0 zd 0 sum 0 dz 0 zd 0 sum 0 dz 0 zd 0 sum 0 ff 0 ff 0 sum 0 ff 0 ff 0 sum 0 ff 0 ff 0 sum 0 fɡ 4 ɡf 0 sum 4 fɡ 6 ɡf 0 sum 6 fɡ 10 ɡf 0 sum 10 fk 7 kf 24 sum 31 fk 10 kf 0 sum 10 fk 17 kf 24 sum 41 fl 4 lf 6 sum 10 fl 14 lf 0 sum 14 fl 18 lf 6 sum 24 fm 0 mf 0 sum 0 fm 0 mf 0 sum 0 fm 0 mf 0 sum 0 fn 1 nf 6 sum 7 fn 6 nf 0 sum 6 fn 7 nf 6 sum 13 fp 0 pf 0 sum 0 fp 0 pf 0 sum 0 fp 0 pf 0 sum 0 fʁ 11 ʁf 12 sum 23 fʁ 4 ʁf 0 sum 4 fʁ 15 ʁf 12 sum 27 fs 6 sf 16 sum 22 fs 8 sf 0 sum 8 fs 14 sf 16 sum 30 fʃ 4 ʃf 12 sum 16 fʃ 6 ʃf 0 sum 6 fʃ 10 ʃf 12 sum 22 ft 8 tf 22 sum 30 ft 12 tf 0 sum 12 ft 20 tf 22 sum 42 fʦ 3 ʦf 8 sum 11 fʦ 4 ʦf 0 sum 4 fʦ 7 ʦf 8 sum 15 fv 0 vf 0 sum 0 fv 0 vf 0 sum 0 fv 0 vf 0 sum 0 fx 6 xf 14 sum 20 fx 12 xf 0 sum 12 fx 18 xf 14 sum 32 fz 4 zf 2 sum 6 fz 6 zf 0 sum 6 fz 10 zf 2 sum 12 ɡɡ 0 ɡɡ 0 sum 0 ɡɡ 0 ɡɡ 0 sum 0 ɡɡ 0 ɡɡ 0 sum 0 ɡk 0 kɡ 0 sum 0 ɡk 0 kɡ 0 sum 0 ɡk 0 kɡ 0 sum 0 ɡl 18 lɡ 6 sum 24 ɡl 9 lɡ 0 sum 9 ɡl 27 lɡ 6 sum 33 ɡm 15 mɡ 6 sum 21 ɡm 15 mɡ 9 sum 24 ɡm 30 mɡ 15 sum 45 ɡn 12 nɡ 13 sum 25 ɡn 6 nɡ 18 sum 24 ɡn 18 nɡ 31 sum 49 ɡp 0 pɡ 8 sum 8 ɡp 12 pɡ 3 sum 15 ɡp 12 pɡ 11 sum 23 ɡʁ 27 ʁɡ 18 sum 45 ɡʁ 15 ʁɡ 18 sum 33 ɡʁ 42 ʁɡ 36 sum 78 ɡs 3 sɡ 9 sum 12 ɡs 0 sɡ 15 sum 15 ɡs 3 sɡ 24 sum 27 ɡʃ 9 ʃɡ 12 sum 21 ɡʃ 9 ʃɡ 15 sum 24 ɡʃ 18 ʃɡ 27 sum 45 ɡt 0 tɡ 0 sum 0 ɡt 3 tɡ 6 sum 9 ɡt 3 tɡ 6 sum 9 ɡʦ 0 ʦɡ 0 sum 0 ɡʦ 3 ʦɡ 0 sum 3 ɡʦ 3 ʦɡ 0 sum 3 ɡv 8 vɡ 2 sum 10 ɡv 0 vɡ 0 sum 0 ɡv 8 vɡ 2 sum 10 ɡx 9 xɡ 6 sum 15 ɡx 6 xɡ 3 sum 9 ɡx 15 xɡ 9 sum 24 ɡz 12 zɡ 0 sum 12 ɡz 6 zɡ 3 sum 9 ɡz 18 zɡ 3 sum 21 kk 0 kk 0 sum 0 kk 1 kk 0 sum 1 kk 1 kk 0 sum 1 kl 17 lk 10 sum 27 kl 14 lk 12 sum 26 kl 31 lk 22 sum 53 km 26 mk 3 sum 29 km 12 mk 15 sum 27 km 38 mk 18 sum 56 kn 10 nk 21 sum 31 kn 9 nk 33 sum 42 kn 19 nk 54 sum 73 kp 6 pk 12 sum 18 kp 12 pk 5 sum 17 kp 18 pk 17 sum 35 49 kal pi'el kal and pi'el kʁ 34 ʁk 20 sum 54 kʁ 0 ʁk 33 sum 33 kʁ 34 ʁk 53 sum 87 ks 16 sk 16 sum 32 ks 6 sk 33 sum 39 ks 22 sk 49 sum 71 kʃ 25 ʃk 24 sum 49 kʃ 14 ʃk 33 sum 47 kʃ 39 ʃk 57 sum 96 kt 21 tk 10 sum 31 kt 26 tk 21 sum 47 kt 47 tk 31 sum 78 kʦ 12 ʦk 0 sum 12 kʦ 9 ʦk 0 sum 9 kʦ 21 ʦk 0 sum 21 kv 16 vk 1 sum 17 kv 0 vk 6 sum 6 kv 16 vk 7 sum 23 kx 8 xk 13 sum 21 kx 3 xk 14 sum 17 kx 11 xk 27 sum 38 kz 2 zk 13 sum 15 kz 4 zk 15 sum 19 kz 6 zk 28 sum 34 ll 0 ll 0 sum 0 ll 3 ll 0 sum 3 ll 3 ll 0 sum 3 lm 3 ml 18 sum 21 lm 6 ml 6 sum 12 lm 9 ml 24 sum 33 ln 0 nl 0 sum 0 ln 0 nl 0 sum 0 ln 0 nl 0 sum 0 lp 3 pl 8 sum 11 lp 6 pl 7 sum 13 lp 9 pl 15 sum 24 lʁ 0 ʁl 0 sum 0 lʁ 0 ʁl 0 sum 0 lʁ 0 ʁl 0 sum 0 ls 0 sl 12 sum 12 ls 0 sl 21 sum 21 ls 0 sl 33 sum 33 lʃ 3 ʃl 24 sum 27 lʃ 3 ʃl 21 sum 24 lʃ 6 ʃl 45 sum 51 lt 9 tl 12 sum 21 lt 9 tl 18 sum 27 lt 18 tl 30 sum 48 lʦ 3 ʦl 18 sum 21 lʦ 0 ʦl 15 sum 15 lʦ 3 ʦl 33 sum 36 lv 8 vl 6 sum 14 lv 0 vl 8 sum 8 lv 8 vl 14 sum 22 lx 17 xl 28 sum 45 lx 6 xl 25 sum 31 lx 23 xl 53 sum 76 lz 0 zl 12 sum 12 lz 0 zl 6 sum 6 lz 0 zl 18 sum 18 mm 0 mm 0 sum 0 mm 6 mm 0 sum 6 mm 6 mm 0 sum 6 mn 0 nm 0 sum 0 mn 3 nm 15 sum 18 mn 3 nm 15 sum 18 mp 0 pm 0 sum 0 mp 0 pm 0 sum 0 mp 0 pm 0 sum 0 mʁ 18 ʁm 12 sum 30 mʁ 12 ʁm 6 sum 18 mʁ 30 ʁm 18 sum 48 ms 12 sm 12 sum 24 ms 9 sm 18 sum 27 ms 21 sm 30 sum 51 mʃ 12 ʃm 9 sum 21 mʃ 9 ʃm 18 sum 27 mʃ 21 ʃm 27 sum 48 mt 9 tm 21 sum 30 mt 18 tm 21 sum 39 mt 27 tm 42 sum 69 mʦ 9 ʦm 15 sum 24 mʦ 6 ʦm 18 sum 24 mʦ 15 ʦm 33 sum 48 mv 0 vm 0 sum 0 mv 0 vm 0 sum 0 mv 0 vm 0 sum 0 mx 18 xm 28 sum 46 mx 6 xm 24 sum 30 mx 24 xm 52 sum 76 mz 6 zm 6 sum 12 mz 6 zm 9 sum 15 mz 12 zm 15 sum 27 nn 0 nn 0 sum 0 nn 3 nn 0 sum 3 nn 3 nn 0 sum 3 np 1 pn 2 sum 3 np 15 pn 3 sum 18 np 16 pn 5 sum 21 nʁ 0 ʁn 3 sum 3 nʁ 0 ʁn 3 sum 3 nʁ 0 ʁn 6 sum 6 ns 7 sn 9 sum 16 ns 9 sn 15 sum 24 ns 16 sn 24 sum 40 nʃ 14 ʃn 9 sum 23 nʃ 21 ʃn 15 sum 36 nʃ 35 ʃn 24 sum 59 nt 18 tn 6 sum 24 nt 24 tn 6 sum 30 nt 42 tn 12 sum 54 nʦ 8 ʦn 12 sum 20 nʦ 6 ʦn 6 sum 12 nʦ 14 ʦn 18 sum 32 nv 8 vn 0 sum 8 nv 0 vn 0 sum 0 nv 8 vn 0 sum 8 50 kal pi'el kal and pi'el nx 19 xn 23 sum 42 nx 18 xn 21 sum 39 nx 37 xn 44 sum 81 nz 7 zn 6 sum 13 nz 6 zn 9 sum 15 nz 13 zn 15 sum 28 pp 0 pp 0 sum 0 pp 0 pp 0 sum 0 pp 0 pp 0 sum 0 pʁ 22 ʁp 6 sum 28 pʁ 2 ʁp 15 sum 17 pʁ 24 ʁp 21 sum 45 ps 12 sp 8 sum 20 ps 4 sp 9 sum 13 ps 16 sp 17 sum 33 pʃ 8 ʃp 6 sum 14 pʃ 3 ʃp 9 sum 12 pʃ 11 ʃp 15 sum 26 pt 16 tp 11 sum 27 pt 6 tp 15 sum 21 pt 22 tp 26 sum 48 pʦ 6 ʦp 4 sum 10 pʦ 2 ʦp 9 sum 11 pʦ 8 ʦp 13 sum 21 pv 0 vp 0 sum 0 pv 0 vp 0 sum 0 pv 0 vp 0 sum 0 px 14 xp 13 sum 27 px 6 xp 24 sum 30 px 20 xp 37 sum 57 pz 8 zp 1 sum 9 pz 3 zp 6 sum 9 pz 11 zp 7 sum 18 ʁʁ 0 ʁʁ 0 sum 0 ʁʁ 0 ʁʁ 0 sum 0 ʁʁ 0 ʁʁ 0 sum 0 ʁs 6 sʁ 39 sum 45 ʁs 9 sʁ 3 sum 12 ʁs 15 sʁ 42 sum 57 ʁʃ 9 ʃʁ 9 sum 18 ʁʃ 15 ʃʁ 3 sum 18 ʁʃ 24 ʃʁ 12 sum 36 ʁt 24 tʁ 24 sum 48 ʁt 21 tʁ 18 sum 39 ʁt 45 tʁ 42 sum 87 ʁʦ 12 ʦʁ 21 sum 33 ʁʦ 9 ʦʁ 9 sum 18 ʁʦ 21 ʦʁ 30 sum 51 ʁv 10 vʁ 8 sum 18 ʁv 0 vʁ 12 sum 12 ʁv 10 vʁ 20 sum 30 ʁx 28 xʁ 44 sum 72 ʁx 15 xʁ 21 sum 36 ʁx 43 xʁ 65 sum 108 ʁz 6 zʁ 15 sum 21 ʁz 0 zʁ 15 sum 15 ʁz 6 zʁ 30 sum 36 ss 0 ss 0 sum 0 ss 0 ss 0 sum 0 ss 0 ss 0 sum 0 sʃ 0 ʃs 6 sum 6 sʃ 0 ʃs 3 sum 3 sʃ 0 ʃs 9 sum 9 st 18 ts 3 sum 21 st 15 ts 3 sum 18 st 33 ts 6 sum 39 sʦ 0 ʦs 0 sum 0 sʦ 0 ʦs 0 sum 0 sʦ 0 ʦs 0 sum 0 sv 10 vs 3 sum 13 sv 0 vs 10 sum 10 sv 10 vs 13 sum 23 sx 38 xs 26 sum 64 sx 9 xs 21 sum 30 sx 47 xs 47 sum 94 sz 0 zs 0 sum 0 sz 0 zs 0 sum 0 sz 0 zs 0 sum 0 ʃʃ 0 ʃʃ 0 sum 0 ʃʃ 0 ʃʃ 0 sum 0 ʃʃ 0 ʃʃ 0 sum 0 ʃt 21 tʃ 6 sum 27 ʃt 15 tʃ 0 sum 15 ʃt 36 tʃ 6 sum 42 ʃʦ 3 ʦʃ 0 sum 3 ʃʦ 0 ʦʃ 0 sum 0 ʃʦ 3 ʦʃ 0 sum 3 ʃv 16 vʃ 1 sum 17 ʃv 0 vʃ 2 sum 2 ʃv 16 vʃ 3 sum 19 ʃx 42 xʃ 23 sum 65 ʃx 18 xʃ 19 sum 37 ʃx 60 xʃ 42 sum 102 ʃz 6 zʃ 0 sum 6 ʃz 6 zʃ 0 sum 6 ʃz 12 zʃ 0 sum 12 tt 0 tt 0 sum 0 tt 0 tt 0 sum 0 tt 0 tt 0 sum 0 tʦ 0 ʦt 0 sum 0 tʦ 0 ʦt 6 sum 6 tʦ 0 ʦt 6 sum 6 tv 4 vt 3 sum 7 tv 0 vt 10 sum 10 tv 4 vt 13 sum 17 tx 26 xt 36 sum 62 tx 9 xt 40 sum 49 tx 35 xt 76 sum 111 tz 3 zt 0 sum 3 tz 3 zt 3 sum 6 tz 6 zt 3 sum 9 ʦʦ 0 ʦʦ 0 sum 0 ʦʦ 0 ʦʦ 0 sum 0 ʦʦ 0 ʦʦ 0 sum 0 ʦv 6 vʦ 3 sum 9 ʦv 0 vʦ 4 sum 4 ʦv 6 vʦ 7 sum 13 51 kal pi'el kal and pi'el ʦx 12 xʦ 12 sum 24 ʦx 6 xʦ 12 sum 18 ʦx 18 xʦ 24 sum 42 ʦz 0 zʦ 0 sum 0 ʦz 0 zʦ 0 sum 0 ʦz 0 zʦ 0 sum 0 vv 0 vv 0 sum 0 vv 0 vv 0 sum 0 vv 0 vv 0 sum 0 vx 4 xv 14 sum 18 vx 10 xv 0 sum 10 vx 14 xv 14 sum 28 vz 3 zv 6 sum 9 vz 0 zv 0 sum 0 vz 3 zv 6 sum 9 xx 9 xx 0 sum 9 xx 6 xx 0 sum 6 xx 15 xx 0 sum 15 xz 13 zx 5 sum 18 xz 14 zx 0 sum 14 xz 27 zx 5 sum 32 zz 0 zz 0 sum 0 zz 0 zz 0 sum 0 zz 0 zz 0 sum 0 52 Appendix E: Frequencies in the lexicon kal: b C1 76 3.29% C2 57 2.44% sum 133 2.86% d C1 84 3.64% C2 102 4.36% sum 186 4.01% f C1 58 2.51% C2 128 5.48% sum 186 4.01% ɡ C1 132 5.72% C2 96 4.11% sum 228 4.91% k C1 240 10.40% C2 162 6.93% sum 402 8.66% l C1 72 3.12% C2 204 8.73% sum 276 5.94% m C1 114 4.94% C2 153 6.55% sum 267 5.75% n C1 135 5.85% C2 93 3.98% sum 228 4.91% p C1 116 5.03% C2 64 2.74% sum 180 3.88% ʁ C1 180 7.80% C2 300 12.84% sum 480 10.34% s C1 201 8.71% C2 108 4.62% sum 309 6.65% ʃ C1 216 9.36% C2 123 5.26% sum 339 7.30% t C1 150 6.50% C2 195 8.34% sum 345 7.43% ʦ C1 102 4.42% C2 78 3.34% sum 180 3.88% v C1 38 1.65% C2 114 4.88% sum 152 3.27% x C1 324 14.04% C2 282 12.07% sum 606 13.05% z C1 69 2.99% C2 78 3.34% sum 147 3.17% b C1 35 1.72% C2 180 8.85% sum 215 5.29% d C1 78 3.83% C2 117 5.75% sum 195 4.79% f C1 90 4.42% C2 0 0.00% sum 90 2.21% ɡ C1 117 5.75% C2 102 5.01% sum 219 5.38% k C1 146 7.18% C2 240 11.80% sum 386 9.49% l C1 63 3.10% C2 177 8.70% sum 240 5.90% m C1 114 5.60% C2 180 8.85% sum 294 7.23% n C1 186 9.14% C2 108 5.31% sum 294 7.23% p C1 45 2.21% C2 141 6.93% sum 186 4.57% ʁ C1 165 8.11% C2 126 6.19% sum 291 7.15% s C1 162 7.96% C2 90 4.42% sum 252 6.19% ʃ C1 180 8.85% C2 105 5.16% sum 285 7.01% t C1 129 6.34% C2 213 10.47% sum 342 8.41% ʦ C1 81 3.98% C2 57 2.80% sum 138 3.39% v C1 70 3.44% C2 0 0.00% sum 70 1.72% x C1 298 14.65% C2 144 7.08% sum 442 10.87% z C1 75 3.69% C2 54 2.65% sum 129 3.17% pi'el: 53 kal and pi'el: b C1 111 2.56% C2 237 5.42% sum 348 3.99% d C1 162 3.73% C2 219 5.01% sum 381 4.37% f C1 148 3.41% C2 128 2.93% sum 276 3.17% ɡ C1 249 5.74% C2 198 4.53% sum 447 5.13% k C1 386 8.89% C2 402 9.20% sum 788 9.04% l C1 135 3.11% C2 381 8.72% sum 516 5.92% m C1 228 5.25% C2 333 7.62% sum 561 6.44% n C1 321 7.39% C2 201 4.60% sum 522 5.99% p C1 161 3.71% C2 205 4.69% sum 366 4.20% ʁ C1 345 7.95% C2 426 9.75% sum 771 8.85% s C1 363 8.36% C2 198 4.53% sum 561 6.44% ʃ C1 396 9.12% C2 228 5.22% sum 624 7.16% t C1 279 6.43% C2 408 9.33% sum 687 7.89% ʦ C1 183 4.22% C2 135 3.09% sum 318 3.65% v C1 108 2.49% C2 114 2.61% sum 222 2.55% x C1 622 14.33% C2 426 9.75% sum 1048 12.03% z C1 144 3.32% C2 132 3.02% sum 276 3.17% 54 Appendix F: Lexical decision experiment: stimuli Nonce-Verbs 0 shared features 1 shared feature 2 shared features 2 shared features Identical C1-C2 (place) (place and manner) (place and voice) ɡisem disem ʃizek tiʦem ʦiʦeɡ ɡiʦem liseɡ sizek tiʃem ʃiʃeɡ likem liʦem ziseɡ ʦiʃem ʦiʦem miʃeɡ ʦizek zisem tisem ʃiʃem ʃimek zitem ziʃeɡ siʦeɡ sisem zikem ziʦeɡ ziʃem ʃiʦeɡ siseɡ Real Verbs bikeʃ kibel kipeʦ pinek ʃilem biʃel kibes kiʃef sibex ʃilev dileɡ kidem litef sikem ʃitef ɡibeʃ kilef mizeɡ silek ʃitek ɡilem kimet piked sipek tinef kibed kipel pileɡ ʃikem ʦilem Trial Session Real Verbs Nonce-Verbs dibeʁ bideɡ limed diɡev sipeʁ ɡidev sixek piɡet xibek tipeɡ 55 Appendix G: Word-likelihood judgment experiment: stimuli All stimuli are in the form of: proper name + nonce-verb + the animal Each participant got 49 random sentences, with a random verb from the bold triplet. Yaron bafad\bafan\bafat et ha-ʃablul (snail) Uri paxan\paxal\paxaʃ et ha-snai (squirrel) Ron bakaz\bakat\bakas et ha-dolfin (dolphin) Dror pafak\pafaɡ\pafaʁ et ha-atalef (bat) Yonatan baɡaz\baɡat\baɡal et ha-ʃoʁ (bull) Dan pamaɡ\pamak\pamax et ha-kaʁiʃ (shark) Dror baxat\baxaʦ\baxas et ha-dvoʁa (bee) Chen pavaɡ\pavaʁ\pavax et ha-taiʃ (male goat) Tomer baʃak\baʃaɡ\baʃax et ha-tolaat (worm) Ori pazak\pazax\pazaɡ et ha-janʃuf (owl) Itamar bamax\bamaʁ\bamaɡ et ha-tiɡʁis (tiger) Oz panak\panaʁ\panax et ha-zvuv (fly) Elad banak\banax\banaʁ et ha-ɡdi (young goat) Nitsan padak\padaʁ\padaɡ et ha-saknai (pelican) Uri bavak\bavax\bavaɡ et ha-ʤiʁafa (giraffe) Dor palaʁ\palaʁ\palak et ha-letaa (lizard) Elad baʁal\baʁan\baʁas et ha-kof (monkey) Omer pasaɡ\pasaɡ\pasaʁ et ha-ʦfaʁdea (frog) Yossi balaʁ\balaɡ\balax et ha-tinʃemet (barn owl) Doron paʃak\paʃaɡ\paʃak et ha-tuki (parrot) Idan basak\basax\basaɡ et ha-axbaʁ (mouse) Netanel paʦaɡ\paʦaɡ\paʦak et ha-akʁav (scorpion) Chen baʦaɡ\baʦax\baʦaɡ et ha-tanin (crocodile) Michael paʁan\paʁal\paʁan et ha-kof (monkey) Eyal bazaɡ\bazax\bazax et ha-paʁoʃ (flea) Yair pataɡ\pataɡ\pataɡ et ha-tolaat (worm) Nadav badaɡ\badaɡ\badaɡ et ha-oʁev (crow) Tomer ʁaval\ʁavan\ʁavat et ha-aʁje (lion) Yoav bataɡ\bataɡ\bataɡ et ha-ʦipoʁ (bird) Eran ʁafal\ʁafaz\ʁafan et ha-baʁvaz (duck) Yuval dafax\dafaʁ\dafaʁ et ha-jona (pigeon) Nir ʁalaf\ʁalav\ʁalam et ha-ʃual (fox) Gal damaʁ\damak\damaɡ et ha-dov (bear) Shay ʁanaf\ʁanam\ʁanav et ha-eɡel (calf) Doron dadav\dadaf\dadam et ha-pinɡwin (penguin) Saar ʁaʁam\ʁaʁaf\ʁaʁav et ha-xipuʃit (ladybug) Michael dazam\dazav\dazaf et ha-ʦav (turtle) Adam saɡaʃ\saɡaʦ\saɡat et ha-dvoʁa (bee) Ran davaɡ\davax\davaɡ et ha-ʦav (turtle) Yotam saxaz\saxaʃ\saxaʦ et ha-kipod (hedgehog) Nadav daxam\daxav\daxav et ha-paʁpaʁ (butterfly) Aviv sazam\sazav\sazaf et ha-kivsa (sheep) Ram dalam\dalam\dalav et ha-taʁneɡol (rooster) Boaz sadam\sadaf\sadav et ha-nameʁ (leopard) Amit daʁaf\daʁaf\daʁav et ha-zeev (wolf) Guy sasam\sasav\sasaf et ha-xatul (cat) Israel ɡafad\ɡafaz\ɡafal et ha-livjatan (whale) Matan safaz\safaʦ\safaz et ha-tuki (parrot) Netanel ɡamas\ɡamat\ɡamaʦ et ha-meduza (jellyfish) Itamar savaɡ\savak\savak et ha-ʃoʁ (bull) Erez ɡaɡaf\ɡaɡam\ɡaɡav et ha-dinozauʁ (dinosaur) Yoav sanav\sanam\sanav et ha-nemala (ant) Barak ɡaʦav\ɡaʦaf\ɡaʦam et ha-axbeʁoʃ (rat) Ran sataf\satav\sataf et ha-ʃafan (rabbit) Tal ɡavad\ɡavaʦ\ɡavaz et ha-nemala (ant) Yuval samaɡ\samaɡ\samaɡ et ha-naxaʃ (snake) Alon ɡazav\ɡazav\ɡazaf et ha-akaviʃ (spider) Or saʁam\saʁam\saʁam et ha-avaz (goose) Gil ɡaʁan\ɡaʁaʦ\ɡaʁan et ha-dov (bear) Oded salav\salav\salav et ha-ʃimpanza (chimp) Roee ɡaxaf\ɡaxaf\ɡaxav et ha-xipuʃit (ladybug) Oren ʃaɡaf\ʃaɡav\ʃaɡam et ha-ɡdi (young goat) Oren ɡalak\ɡalaʁ\ɡalaʁ et ha-janʃuf (owl) Omri ʃaʃaf\ʃaʃam\ʃaʃav et ha-letaa (lizard) Amir ɡadav\ɡadav\ɡadav et ha-akʁav (scorpion) Danny ʃalaʁ\ʃalaɡ\ʃalaʁ et ha-ez (nanny goat) Shachar kaɡam\kaɡaf\kaɡav et ha-boeʃ (polecat) Amit ʃafaɡ\ʃafak\ʃafak et ha-hipopotam (hippo) Ariel kakaf\kakav\kakam et ha-kelev (dog) Omri ʃaʁaf\ʃaʁaf\ʃaʁam et ha-saknai (pelican) 56 Ori kamad\kamad\kamaz et ha-daɡ (fish) Nir ʃadam\ʃadav\ʃadam et ha-sus (horse) Gal kadav\kadaf\kadaf et ha-pinɡwin (penguin) Eyal ʃasam\ʃasav\ʃasav et ha-aʁnav (hare) Matan kataɡ\katax\katax et ha-nameʁ (leopard) Yakir ʃaʦam\ʃaʦav\ʃaʦam et ha-jatuʃ (mosquito) Saar kalaɡ\kalaʁ\kalaɡ et ha-pil (elephant) Itay ʃaxam\ʃaxam\ʃaxam et ha-kaʁnaf (rhino) Moti kavaz\kavaz\kavat et ha-hipopotam (hippo) Dan ʃavaɡ\ʃavaɡ\ʃavaɡ et ha-kaʁnaf (rhino) Nimrod kafas\kafas\kafas et ha-paʁpaʁ (butterfly) Noam ʃatav\ʃatav\ʃatav et ha-ɡamal (camel) Moti kaʁaf\kaʁaf\kaʁaf et ha-xamoʁ (donkey) Dor ʃazav\ʃazav\ʃazav et ha-xaziʁ (pig) Yaniv kasav\kasav\kasav et ha-zebʁa (zebra) Tal taɡaf\taɡam\taɡav et ha-ez (nanny goat) Daniel kaʦam\kaʦam\kaʦam et ha-boeʃ (polecat) Ido tavak\tavaʁ\tavaɡ et ha-naxaʃ (snake) Danny kaʃam\kaʃam\kaʃam et ha-zvuv (fly) Israel tatav\tatam\tataf et ha-zebʁa (zebra) Assaf lavax\lavak\lavaɡ et ha-baʁvaz (duck) Ofir taʦav\taʦam\taʦaf et ha-ʤiʁafa (giraffe) Idan lafaʁ\lafaɡ\lafak et ha-xamoʁ (donkey) Yair tafak\tafak\tafaɡ et ha-dolfin (dolphin) Aviad lalam\lalaf\lalav et ha-xatul (cat) Ohad tamak\tamaɡ\tamaɡ et ha-kelev (dog) Erez mafax\mafaɡ\mafaʁ et ha-ʦipoʁ (bird) Tom tadav\tadav\tadaf et ha-ʦaʁʦaʁ (cricket) Ofer mavaʁ\mavaɡ\mavak et ha-axbeʁoʃ (rat) Yoni tazav\tazaf\tazaf et ha-kivsa (sheep) Or maʁas\maʁaʦ\maʁaz et ha-akaviʃ (spider) Itay taʁav\taʁav\taʁav et ha-aʁnav (hare) Alon mamax\mamaɡ\mamaʁ et ha-tavas (peacock) Boaz talav\talav\talav et ha-tavas (peacock) Amir mazax\mazak\mazak et ha-snai (squirrel) Ron ʦaʁaʃ\ʦaʁas\ʦaʁaz et ha-paʁoʃ (flea) Daniel maʃaʁ\maʃaʁ\maʃaɡ et ha-ʦvi (deer) Eran ʦafax\ʦafak\ʦafaɡ et ha-livjatan (whale) Tsachi malaʁ\malaʁ\malaʁ et ha-paʁa (cow) Barak ʦaxam\ʦaxaf\ʦaxav et ha-kipod (hedgehog) Yakir nadak\nadaɡ\nadax et ha-oʁev (crow) Assaf ʦasaf\ʦasam\ʦasav et ha-avaz (goose) Ben nafak\nafaʁ\nafaɡ et ha-ʃual (fox) Adam ʦaʦam\ʦaʦaf\ʦaʦav et ha-sus (horse) Noam nalav\nalam\nalaf et ha-jona (pigeon) Yossi ʦazav\ʦazam\ʦazaf et ha-baʁbuʁ (swan) Oded nanam\nanaf\nanav et ha-baʁbuʁ (swan) Nitsan ʦalaɡ\ʦalaɡ\ʦalaʁ et ha-kaʁiʃ (shark) Ohad nakas\nakas\nakaʦ et ha-zeev (wolf) Yotam ʦavak\ʦavak\ʦavaɡ et ha-ʦaʁʦaʁ (cricket) Tom navaɡ\navaɡ\navak et ha-paʁa (cow) Evyatar ʦadav\ʦadam\ʦadam et ha-oɡeʁ (hamster) Ariel nazav\nazam\nazav et ha-tanin (crocodile) Nimrod ʦamaɡ\ʦamaɡ\ʦamaɡ et ha-ʃimpanza (chimp) Ido naʃaɡ\naʃaɡ\naʃaɡ et ha-kenɡeʁu (kangaroo) Oz ʦanav\ʦanav\ʦanav et ha-dinozauʁ (dinosaur) Shay naɡam\naɡam\naɡam et ha-ʦfaʁdea (frog) Ram xalaɡ\xalaʁ\xalaʁ et ha-jatuʃ (mosquito) Guy natam\natam\natam et ha-taiʃ (male goat) Ofir xatak\xatak\xataɡ et ha-xasida (stork) Yaron namaʁ\namaʁ\namaʁ et ha-axbaʁ (mouse) Aviv xazav\xazaf\xazaf et ha-pil (elephant) Ofer pakaz\pakat\pakaʃ et ha-ʃablul (snail) Roee xaxav\xaxaf\xaxav et ha-kenɡeʁu (kangaroo) Yaniv paɡan\paɡas\paɡaʦ et ha-ɡamal (camel) Gil xaʁan\xaʁal\xaʁan et ha-eɡel (calf) Evyatar zafak\zafax\zafaʁ et ha-ʃafan (rabbit) Avi xafal\xafal\xafad et ha-daɡ (fish) Tsachi zazav\zazaf\zazam et ha-xaziʁ (pig) Omer xakav\xakav\xakaf et ha-xasida (stork) Lior zakav\zakam\zakam et ha-oɡeʁ (hamster) Shachar xavan\xavaz\xavan et ha-meduza (jellyfish) Yoni zavaʁ\zavaʁ\zavak et ha-aʁje (lion) Lior xanaʁ\xanaʁ\xanaʁ et ha-ʦvi (deer) Yonatan zalav\zalav\zalam et ha-atalef (bat) Ben xamaz\xamaz\xamaz et ha-taʁneɡol (rooster) Aviad zaʁaf\zaʁaf\zaʁaf et ha-tiɡʁis (tiger) 57 Appendix H: Word-likelihood judgment experiment: The triplets with significant difference in ratings bʁ bʦ dd df dʁ dx pd pk pl pm pʃ Average SD baʁal 3.43 1.79 baʁan 4.25 2.02 baʁas 5.39 baʦaɡ P-value Average SD paʦaɡ 3.80 1.32 paʦaɡ 3.07 1.59 1.75 paʦak 4.56 1.03 3.45 1.69 ʁaʁaf 3.29 1.82 baʦaɡ 2.36 1.22 ʁaʁam 2.07 1.44 baʦax 4.07 1.73 ʁaʁav 1.31 0.70 dadaf 3.08 2.22 ʃafaɡ 3.43 1.83 dadam 3.41 2.00 ʃafak 4.41 1.46 dadav 1.82 1.29 ʃafak 4.94 1.92 dafaʁ 4.69 1.35 saɡaʃ 3.25 2.14 dafaʁ 4.83 1.79 saɡat 3.71 1.69 dafax 3.43 1.79 saɡaʦ 2.25 1.57 daʁaf 5.18 1.88 ʃalaɡ 5.06 1.61 daʁaf 5.00 1.53 ʃalaʁ 3.78 1.59 daʁav 3.86 1.75 ʃalaʁ 4.06 2.10 daxam 3.14 1.79 savaɡ 5.14 1.41 daxav 5.06 1.85 savak 3.58 1.73 daxav 4.94 1.30 savak 3.67 1.50 padaɡ 3.12 1.36 tataf 3.18 1.78 padak 5.12 1.80 tatam 3.73 1.75 padaʁ 3.93 1.58 tatav 2.25 1.81 pakaʃ 4.50 1.70 ʦaʦaf 2.89 1.75 pakat 4.67 1.67 ʦaʦam 2.21 1.63 pakaz 3.00 1.81 ʦaʦav 1.70 1.49 palak 5.23 1.36 palaʁ 4.15 1.68 palaʁ 4.07 1.58 pamaɡ 1.92 0.64 pamak 3.50 1.15 pamax 4.24 1.92 paʃaɡ 3.27 1.94 paʃak 4.43 0.94 paʃak 4.83 1.72 pʦ 0.015 ʁʁ 0.032 ʃf 0.037 sɡ 0.015 ʃl 0.024 0.002 sv 0.004 tt ʦʦ 0.022 0.038 0.000 0.011 58 P-value 0.012 0.001 0.029 0.047 0.043 0.005 0.017 0.020 Appendix I: Word-likelihood judgment experiment: Results Average Rating Average Rating Average Rating Average Rating ss 2.05 ɡf 3.49 ɡd 4.05 nt 4.63 ʦs 2.21 sx 3.50 bn 4.06 ʃv 4.64 kɡ 2.27 pɡ 3.52 dl 4.08 ɡʁ 4.64 ʃʃ 2.29 sm 3.55 bd 4.09 nɡ 4.64 zz 2.51 mv 3.55 ʦx 4.09 ʦf 4.64 td 2.53 bs 3.56 ps 4.10 xv 4.64 tʦ 2.60 zk 3.60 ɡv 4.13 ʃʁ 4.65 bv 2.65 xt 3.68 xk 4.16 mʃ 4.73 ʦʁ 2.65 ʁl 3.70 ʦn 4.20 kd 4.74 kk 2.66 nʃ 3.70 st 4.20 xm 4.74 sz 2.66 xl 3.74 ʦv 4.21 ɡx 4.79 ɡɡ 2.67 pn 3.76 ɡz 4.22 zʁ 4.79 ll 2.68 ɡl 3.77 tɡ 4.23 nz 4.83 mm 2.69 bʃ 3.79 nv 4.24 bx 4.84 bt 2.69 zf 3.79 kt 4.25 bl 4.87 nn 2.71 dm 3.80 xz 4.27 kf 5.00 ʦz 2.73 ks 3.80 kʦ 4.28 kʁ 5.10 tz 2.73 zv 3.83 ʃɡ 4.28 xf 5.22 sf 2.80 ʁf 3.84 ʃd 4.29 px 5.39 pf 2.81 lf 3.85 zl 4.30 ʃx 5.41 tm 2.91 sd 3.85 mʁ 4.30 bf 2.93 ml 3.86 kʃ 4.31 ʃʦ 2.96 kl 3.87 tf 4.31 xx 2.96 nd 3.88 sʁ 4.33 dz 3.00 ʃt 3.90 bz 4.33 bm 3.02 pt 3.91 ɡm 4.34 ʃs 3.15 ʁv 3.93 nm 4.38 pz 3.16 kv 3.93 nk 4.39 pv 3.21 km 3.94 nf 4.39 ɡʦ 3.27 dv 3.96 mz 4.42 ʁn 3.31 tl 3.96 ʃz 4.46 bk 3.42 xn 3.98 tv 4.48 nl 3.42 lv 3.98 xʁ 4.48 mf 3.43 sn 4.02 sl 4.49 ʦd 3.44 ʦl 4.02 pʁ 4.52 ʦm 3.45 bɡ 4.04 tʁ 4.60 59 REFERENCES Aronoff, Mark. 1976. Word formation in Generative Grammar. Cambridge: MIT Press. Barkali, S. 1964. Luach hapeʼalim hashalem (18th ed.). Jerusalem: Rubin Mass. [in Hebrew] Bat-El, Outi. 1994. Stem modification and cluster transfer in Modern Hebrew. Natural Language and Linguistic Theory 12: 572-596. Bat-El, Outi. 2003a. Semitic verb structure within a universal perspective. In Joseph Shimron (ed.) Language Processing and Acquisition in Languages of Semitic, Root-based, Morphology. pp. 29-59. Amsterdam: John Benjamins. Berent, Iris, Everett, Daniel. L., and Joseph Shimron. 2001. Do phonological representations specify variables? Evidence from the Obligatory Contour Principle. Cognitive Psychology, 42(1), 1-60. Berent, Iris, Lennertz, Tracy, Jun, Jongho, Moreno, Miguel, A., and Paul Smolensky. 2008. Language universals in human brains. Proceedings of the National Academy of Sciences, 105, 5321-5325. Berent, Iris, and Joseph Shimron. 1997. The representation of Hebrew words: Evidence from the obligatory contour principle. Cognition, 64(1), 39-72. Blau, Joshua. 2010. The Phonology and Morphology of Biblical Hebrew. The Academy of Hebrew Language. [In Hebrew] Bolozky, Shmuel and Rina Kreitman. 2007. Uvulars in Israeli Hebrew – their phonetic and phonological status. The National Association of Professors of Hebrew's International Conference on Hebrew Language, Literature and Culture. Sydney, Australia July 2-4. Clements, George N. 1985. The Geometry of Phonological Features. Phonology Yearbook 2:225-252. 60 Clements, George N. and Elizabeth Hume. 1995. The Internal Organization of Speech Sounds. In John Goldsmith, ed., Handbook of Phonological Theory. Oxford: Basil Blackwell, Oxford, pp. 245-306 Clements, George N. and Samuel J. Keyser. 1983. CV Phonology. MIT Press, Cambridge, MA. Davis. Stuart M. 1984. Some Implications of Onset-Coda Constraints for Syllable Phonology. Chicago Linguistic Society 20: 46-51 Even-Shoshan, Avraham. 1970. The New Hebrew Dictionary. Jerusalem: Kiryat Sefer. Even-Shoshan, Avraham. 1983. The Condensed Hebrew Dictionary. Jerusalem: Kiryat Sefer Frisch, Stefan. A., Pierrehumbert, Janet. B., and Michael B Broe. 2004. Similarity avoidance and the OCP. Natural Language & Linguistic Theory, 22(1), 179228. Frisch, Stefan A. and Bushra Zawaydeh. 2001. The Psychological Reality of OCP-Place in Arabic, Language 77, 91–106. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5: 253-287. Goldsmith, John. A. 1976. Autosegmental phonology (Vol. 159). Bloomington: Indiana University Linguistics Club.Yip 1998 Greenberg, Joseph. 1950. The patterning of morphemes in Semitic. Word, 4, 196–208. Johnson, Keith. 2012. Acoustic and Auditory Phonetics 3rd Edition. Wiley-Blackwell, West Sussex. Pages 115-123 Kaisse, Ellen. 1988. Modern Greek continuant dissimilation and the OCP. Ms, University of Washington Kawahara, Shigeto. 2007. Half rhymes in Japanese rap lyrics and knowledge of similarity. Journal of East Asian Linguistics 16:113-144. 61 Kawahara, Shigeto. 2011. Modes of phonological judgment. Ms, Rutgers University. Available as ROA-1140 from the Rutgers Optimality Archive. Kawahara, Shigeto, Hajime Ono, and Kiyoshi Sudo. 2006. Consonant co-occurrence restrictions in Yamato Japanese. Japanese/Korean Linguistics 14:27-38. Laks, Lior. 2011. Morpho-phonological and Morpho-thematic Relations in Hebrew and Arabic Verb Formation. PhD Dissertation, Tel-Aviv University. Leben, William Ronald. 1973. Suprasegmental phonology (Doctoral dissertation, Massachusetts Institute of Technology). McCarthy, John. 1979. Formal problems in Semitic morphology and phonology. Dissertation MIT, Cambridge, MA. McCarthy, John. 1981. A prosodic theory of nonconcatenative morphology.Linguistic inquiry, 12, 373-418. McCarthy, John. 1986. OCP effects: Gemination and antigemination. Linguistic inquiry, 17(2), 207-263. McCarthy, John. 1988. Feature Geometry and Dependency: A Review, Phonetica 43, 84–108. Mielke, Jeff. 2012. A phonetically-based metric of sound similarity. Lingua 122:145163. Odden, David. 1988. Anti antigemination and the OCP. Linguistic inquiry, 19(3), 451476. Padgett, Jaye. 1995. Stricture in feature geometry. SCLI Publications. Pierrehumbert, Janet B. 1993. Dissimilarity in the Arabic Verbal Roots, in Proceedings of the North East Linguistics Society 23, pp. 367–381. 62 Prince, Alan, and Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Technical report RuCCSTR-2. Rutgers Center for Cognitive Science. Rice, Keren. D. 1993. A reexamination of the feature [sonorant]: the status of sonorant obstruents. Language Rose, Sharon. 2000. Rethinking geminates, long-distance geminates, and the OCP. Linguistic Inquiry, 31(1), 85-122. Rose, Sharon, and Rachel Walker. 2004. A typology of consonant agreement as correspondence. Language, 80(3), 475-531 Sagey, Elizabeth C. 1986. The Representation of Features and Relations in Non-linear Phonology. Ph.D. dissertation, MIT. Schwarzwald, Ora. 2002. Studies in Hebrew Morphology. The Open University of Israel [in Hebrew] Yeverechyahu, Hadas. 2012. OCP in Hebrew Verbs - an Empirical View. A poster presented at The Second International Graduate Student Conference on Diverse Approaches to Linguistics (IGDAL 2), Tel-Aviv University. Yip, Moira. 1988. The Obligatory Contour Principle and phonological rules: A loss of identity. Linguistic inquiry, 19(1), 65-100. Zadok, Gila. 2012. On the Instability of Hebrew Weak Verbs: Similarity, Variation, and Change. Ph.D. dissertation, Tel-Aviv University 63 ‫תקציר‬ ‫בשפות שמיות‪ ,‬עיצורים הומורגניים (עיצורים החולקים מקום חיתוך) נוטים שלא להופיע באותו‬ ‫גזע (‪ .)Greenberg 1950‬מחקרים קודמים (כדוגמת ‪)Frisch et al. 2004 ;McCarthy 1981, 1986‬‬ ‫מציעים כי הגבלות אלו נובעות מדמיון בין עיצורים‪ ,‬כך שככל ששני עיצורים (הומורגניים) דומים‬ ‫יותר זה לזה‪ ,‬כך הסבירות שיופיעו באותו גזע נמוכה יותר‪ .‬המחקר הנוכחי בוחן את ההגבלות על‬ ‫מופעי עיצורים במערכת הפועל בעברית‪ .‬אני שואלת כיצד דמיון בין עיצורים תורם להגבלות‪ ,‬והאם‬ ‫ההגבלות נובעות מאילוצים אוניברסליים או שהן מושפעות מגורמים תלויי שפה‪.‬‬ ‫למחקר שלושה חלקים עיקריים‪ .‬ראשית‪ ,‬התאמתי את מודל הדמיון של ‪Frisch et al.‬‬ ‫)‪ (2004‬למצאי העיצורים בעברית‪ .‬שנית‪ ,‬ניתחתי את לקסיקון מערכת הפועל בעברית‪ ,‬תוך‬ ‫התמקדות בעיצורים הראשון והשני של הגזע (‪ )C1-C2‬בבניינים קל ופיעל‪ .‬הניתוח הראה מתאם‬ ‫ברמת מובהקות גבוהה בין מודל הדמיון והלקסיקון‪ ,‬וכן מציע כי למקום החיתוך של העיצורים יש‬ ‫השפעה רבה‪ ,‬בהשוו אה לתכוניות אחרות‪ ,‬בקביעת ההגבלות‪ .‬על מנת לחזק ולהרחיב את הניתוח‬ ‫הלקסיקלי‪ ,‬ערכתי שני ניסויים פסיכובלשניים‪ :‬מטלת זיהוי לקסיקלי ומטלת שיפוטי סבירות עבור‬ ‫לא‪ -‬מילים‪ ,‬שניהם בוחנים את ההגבלות במערכת הפונולוגית של הדוברים‪ .‬ניתוח שיפוטי הסבירות‬ ‫הראה מתאם ברמת מובהקות גבוהה בין שיפוטי הדוברים ומודל הדמיון מחד‪ ,‬ובין שיפוטי‬ ‫הדוברים ומחקר הלקסיקון מאידך‪ .‬כמו כן‪ ,‬הניסויים גם כן מדגישים את תפקיד מקום החיתוך של‬ ‫העיצורים בהגבלות על מופעים‪.‬‬ ‫ממצאים אלו מציעים כי ישנן הגבלות מבוססות דמיון על שני העיצורים הראשונים בגזע‬ ‫הפעלים ב עברית‪ ,‬גם בלקסיקון וגם במערכת הפונולוגית של הדוברים‪ .‬כמו כן‪ ,‬הם מציעים כי‬ ‫למקום החיתוך של העיצורים השפעה רבה בקביעת ההגבלות‪ ,‬כך ששני עיצורים החולקים מקום‬ ‫חיתוך ראשי נוטים שלא להופיע באותו גזע‪ .‬אף על פי כן‪ ,‬הניסויים אינם יכולים להציע האם השפעת‬ ‫הדמיון על המע רכת הפונולוגית היא ישירה‪ ,‬או עקיפה דרך השפעות לקסיקליות‪.‬‬ ‫הפקולטה למדעי הרוח ע"ש לסטר וסאלי אנטין‬ ‫החוג לבלשנות‬ ‫תפקיד הדמיון בהגבלות על רצפי עיצורים‪:‬‬ ‫עדות ממערכת הפועל בעברית‬ ‫חיבור זה הוגש כעבודת גמר לקראת התואר‬ ‫"מוסמך אוניברסיטה" באוניברסיטת ת"א‬ ‫על ידי‬ ‫הדס יברכיהו‬ ‫העבודה הוכנה בהדרכת‪:‬‬ ‫פרופ' אותי בת‪-‬אל‬ ‫ד"ר אוון‪-‬גרי כהן‬ ‫דצמבר ‪4102‬‬

RELATED PAPERS

RELATED TOPICS

Log In

The role of similarity in co-occurrence restrictions: Evidence from the Hebrew verbal system - Hadas Yeverechyahu's Thesis

The role of similarity in co-occurrence restrictions: Evidence from the Hebrew verbal system - Hadas Yeverechyahu's Thesis

Related Papers

RELATED PAPERS

RELATED TOPICS