How do Hebrew speakers ‘trick or treat’? An experimental study of verb formation from Hebrew CCVC nouns Roy Becker-Kristal, UCLA Linguistics roybecker@humnet. ucla. edu
Introduction • This experimental study tests how Hebrew speakers coin novel denominative verbs from known CCVC nouns. • Beyond a mere description of a language-specific state of affairs, this study puts to test various morphophonological theories. • Particular attention is given to the experimental methodology, since comparable studies are rare in the morpho-phonological literature.
Background Hebrew verb conjugations (binyanim) • A small close set of CV template paradigms. Any Hebrew verb must pertain to one such template paradigm or binyan. • There are five binyanim, three of which, namely katal, hiktil and kitel can have any argument structure (niktal and hitkatel are limited to intransitives). katal and hiktil are skeletally rigid, while kitel is more ‘plastic’. • The (sub-)binyanim are named after their Past-3 rd-Sg. Msc template (e. g. katal=/C 1 a. ’C 2 a. C 3/)*. * The binyanim are more commonly named using the consonants p/f-’-l, e. g. pa’al, pi’el, hif’il etc. The use of k-t-l, which is common in the noun template system, is more transparent, and hence used here.
Background The binyanim: katal • • The default binyan in Ancient Hebrew. Huge lexical count. Rather non-productive in Modern Hebrew. Highly diverse base template paradigm (six-to-nine very different templates), e. g. /lis. ’tom/ to seal: /sa. ’tam/ Past 3 Sg. Msc /sat. ’mu/ Past 3 Pl /so. ’tem/ Pres. Sg. Msc /sot. ’mim/ Pres. Pl. Msc /jis. ’tom/ Fut 3 Sg. Msc /jis. te. ’mu/ Fut 3 Pl (Prescriptively also: /stam. ’tem/ Past 2 Pl. Msc, /sit. ’mu/ Imp 2 Pl)
Background The binyanim: hiktil • Strongly associated with causation, but can accommodate any argument structure. • Large lexical count. • Productive mostly in secondary derivation (of causatives from existing non-causative verbs). • Very stable base template paradigm (only two-three rather similar templates). e. g. /le. hak. ’lit/ to record: /hik. ’lit/ Past 3 Sg. Msc /hik. ’lat. nu/ Past 1 Pl /mak. ’lit/ Pres. Sg. Msc /mak. li. ’tim/ Pres. Pl. Msc /jak. ’lit/ Fut 3 Sg. Msc /jak. ’li. tu/ Fut 3 Pl (Prescriptively also: /hak. ’let/ Imp 2 Sg)
Background The binyanim: kitel (canonical C 1 VC 2 VC 3 form) • • The default binyan in Modern Hebrew. Huge lexical count. Highly productive. Moderately stable base template paradigm (five rather similar templates), e. g. /le. ka. ’bel/ to receive: /ki. ’bel/ Past 3 Sg. Msc /kib. ’lu/ Past 3 Pl /ki. ’bal. nu/ Past 1 Pl /me. ka. ’bel/ Pres. Sg. Msc /me. kab. ’lim/ Pres. Pl. Msc /je. ka. ’bel/ Fut 3 Sg. Msc /je. kab. ’lu/ Fut 3 Pl
Background The binyanim: kitel (non-canonical C 0 VC 0 forms) • kitel can accommodate clusters in its consonantal slots: CCC-initial: /st ip. ’tez/ he stripteased CCCC-medial: /t ins. ’k eb/ he transcribed* CC-final: /sig. ’ment/ he segmented* • Moderate lexical count, but highly productive, especially when accommodating loanwords. • Stable base template paradigm (three templates: C 0 i. C 0 e. C 0, C 0 i. C 0 a. C 0, C 0 a. C 0 e. C 0). • CVCCVC kitel verbs (‘kirtel’) are frequent. CCVCCVC kitel verbs (‘kristel’) are also attested. Peculiarly, CCVCVC kitel verbs (‘kritel’) are extremely rare. *Speech-technology jargon. Linguistic circles use different terms.
Background The binyanim: kitel sub-binyanim • The plasticity of kitel is exploited to accommodate bases containing affixes, creating several sub-binyanim. • These sub-binyanim host many denominative verbs from affix-/reduplication-derived CVCCVC nouns. • Prefixed sub-binyanim: iktel, miktel, tiktel, iktel • Suffixed sub-binyanim: kitlen • Reduplication sub-binyanim: kitlel, ktolel. • All these sub-binyanim have moderate-to-low lexical count, but kitlel is extremely productive and iktel has also been becoming fashionable.
Background Nouns and denominative verbs in Hebrew • Nouns (and adjectives) enjoy much greater structural flexibility in Hebrew (and other Semitic languages). There are many more CV-templates, and templates are not at all obligatory. • Therefore, the derivation of a denominative verb involves phonological mapping of free structures onto rigid frames, with inevitable faithfulness violations. Examples: excessive truncation: /kon. fi. gu. ’ at s. ja/ configuration → /kin. ’feg/ he configured full reduplication: /’zap/ zap → /zip. ’zep/ he zapped
Background Denominative verbs in Hebrew (Cont. ) • For some structures it is in fact impossible to derive a denominative verb, e. g. : CV: /’t si/ fleet → ? ? ? CVCVCC: /ko. ’mand/ command (in linguistics) → ? ? ? • Yet, in most cases, not only is it possible to derive a denominative verb, but native Hebrew speakers also have clear intuitions how to derive it. Hypothetical examples: CVCCVC: /’bun. ke / bunker → /bin. ’ke / he bunkered CVCC: /’ ift/ shift → / if. ’tet/ he pressed the ‘shift’ key • The overwhelming majority of newly coined denominative verbs in Hebrew are in either the ‘canonical’ (C 1 i. ’C 2 e. C 3) or in the C 1 VC 2. ’C 3 VC 4 variants of kitel.
Background CCVC nouns and denominative verbs • CCVC nouns are tri-consonantal, therefore all possible binyanim are relevant for verb derivation. • The template paradigms of the most productive subbinyanim (canoincal kitel and kitlel) are the least structurally faithful: /C 1 C 2 VC 3/ rendition of the base never occurs in their paradigm, whereas it does occur in katal, hiktil and various less productive variants of kitel such as iktel or ktolel. • Thus there is a conflict between the morpho-lexical requirement to use the default operation and the phonological requirement to maximize source-derivative faithfulness.
Background Informal systematic survey of CCVC nouns and denominative verbs in the Hebrew lexicon: • 300 -350 CCVC nouns (and adjectives). • 39 unambiguous cases of denominative verbs derived from CCVC nouns: 10 in katal, 19 in hiktil, six in kitel, three in kitlel, four in other kitel variants. • Although these numbers are not reliable for quantitative analysis, the morpho-lexical vs. phonological conflict is apparent: The otherwise preferred strategies kitel and kitlel fare much worse that the template-faithful hiktil.
Background Lexical survey (cont. ): • Apparent ‘island of reliability’ in CCi. C: out of 12 CCi. Csourced verbs, nine (75%) are in hiktil. • Balanced variation between strategies elsewhere. • Huge variation in CCo. C-sourced verbs: 10 verbs, five different strategies: katal: e. g. /xa. ’ ap/ he napped, from /’x op/ nap hiktil: e. g. /his. ’mil/ he turned left, from /’smol/ left kitel: e. g. /bi. ’lef/ he bluffed, from /’blof/ bluff kitlel: /bil. ’geg/ he blogged, from /’blog/ blog ktolel: / no. ’ e / he scrounged, from /’ no / scrounge
Theoretical Approaches The ‘Semitologist’ Approach*: • In Semitic languages, lexical entries include the consonantal root as a component in their representation. • Derivation of the phonological contents of a novel word requires the extraction of a consonantal root, the choice of a template, and the combination of the two together. • Since derivation goes through consonantal root extraction, vocalic contents and prosodic structure of the source of the derivation should play only a marginal role. • Verb derivation strategies should not differ significantly between CVCVC-sourced and CCVC-sourced nouns, nor between CCVC nouns containing different vowels. *Berent et al. (1997, 2001, 2002) clearly manifest this approach, but might not necessarily subscribe to the current line of argumentation.
Theoretical Approaches The ‘Universalist Phonological’ Approach*: • Semitic languages adhere to the same phonological constraints and representations as other languages. • Lexical entries do not contain the consonantal root as an independent representation component. • Derivation of the phonological contents of a novel word is achieved by direct source-derivative mapping, while adhering to faithfulness and markedness constraints. • Direct mapping and source faithfulness imply that, in the derivation of denominative verbs, the skeletal structure and vowel quality of the noun source should play a significant role in determining the derivation strategy. *See e. g. Bat-El (1994 a, 1994 b).
Theoretical Approaches The ‘Lexicalist’ Approach*: • Morphological strategies are determined by the speaker’s knowledge of paradigm distribution in the existing lexicon. • When choosing a morphological strategy, the speaker applies arbitrary statistical knowledge on phonological structures, rather than direct phonological mapping. • The choice of derivation strategies of novel CCVCsourced denominative verbs should match the distribution of strategies among the existing 39 CCVC-sourced verbs. • Hypothetical strategies absent from the Hebrew lexicon should be avoided even if they are phonologically optimal. *This inflection-based approach, advocated in e. g. Albright(2002), Albright&Hayes(2002), might not extend to derivational morphology.
The Experiments on novel word derivation • No standard format (unlike inflectional ‘wug’-tests). • The morphological task is somewhat unnatural (most people do not invent words on a daily basis). • The task tends to be open-ended, and is seldom confined to a small n-ary choice. • Many times no derivational strategy is appropriate – a problem for a strict forced-choice task. • High risk of over-creativity on one hand, and strategy drift on the other.
The Experiment - Guidelines How to do an experiment on derivation? • Ensure that the subjects are in the right state of mind! • Obviate the task by using real word stimuli (not ‘wugs’) and straightforward meanings for the derivatives! • Keep the task simple! Cover most relevant strategies and have the subjects choose, rather than derive themselves. • Don’t force the subjects to choose – leave a default alternative with no derivation available! • Instruct subjects to prefer naturalness over creativity! • Keep the experiment short, but don’t forget to use fillers! • If the experiment is phonological, do not rely solely on written form!
The Experiment - Subjects • To date, 24 subjects have participated (a few others were recruited but dropped out in the middle). • Native Hebrew speakers, aged 22 -36, educated (undergrads to post-docs). • Homogeneous and representative of General Israeli Hebrew (mostly second generation native speakers). • Living in Israel at present or have recently relocated from Israel to Los-Angeles. • No thorough background in theoretical linguistics. • Neither academic nor professional experience in applying prescriptivist norms.
The Experiment - Stimuli • 63 nouns, ranging from core vocabulary, through slang and loanwords to English words used in code-mixing. • 48 CCVC stimuli, comprising of 16 CCi. C stimuli and 8 stimuli for each of CCe. C, CCa. C, CCo. C and CCu. C. • 15 control/filler stimuli: 5 stimuli for each of CV. ’CVC, VC. ’CVC and CCV. ’CVC, all perfectly balanced for vowel qualities in both stressed and unstressed syllables. • Stimuli never had existing semantically and morphophonologically related verbs (denominative or not). • Stimuli were phonologically compatible with all of the available strategies (not yielding phonotactic violations). • Different stimuli never shared the same consonantal root.
The Experiment - Task • Given a noun stimulus, a related predicative meaning, and a sentence illustrating the use of the predicate, the subject had to choose the most natural denominative verb among eight derivational options (in Past-3 rd-Sg-Msc). • For CCVC stimuli, these options included the seven (sub-) binyanim katal, hiktil, kitel, kitlel, iktel, ktolel (lexical count: 1), ktilel (fictive, lexical count: 0). • For the control conditions there were also seven options, but these varied (since, for some stimuli, certain triconsonantal options had to be replaced by dummies). • The eighth option was always a paraphrastic rendition of the predicate with no derivation. The subjects were instructed to choose this default option only as last resort.
The Experiment - Procedure • For each experiment session, the subject receives by email a Power. Point presentation, an answer sheet and detailed installation instructions. • The subject is instructed to ‘condergo’ the session while sitting alone in front of the computer. • The presentation begins with a multi-media introduction to denominative verb derivation, illustrating both real verbs and (obvious) intuitions about some hypothetical verbs. • The introduction states explicitly that the experiment tests how Hebrew speakers derive verbs from nouns. However, the phonological aspects studied are not disclosed. • The subject is explicitly instructed to prefer naturalness to creativity.
The Experiment – Presentation • The 63 stimuli were divided to three sessions of 21 stimuli in each (16 CCVC and 5 of the other conditions). • The order of the 5 non-CCVC stimuli was randomized. • The order of the 16 CCVC stimuli was randomized, and they were divided to four blocks of four stimuli. These four blocks were interlaced between the non-CCVC stimuli. • For each stimulus, the order of the seven non-default derivational strategies was randomized. • Graphemic as well as carefully-controlled, naturallysounding, recorded auditory renditions were created for both the stimuli and all the derivational options. • Each stimulus was presented, with its derivational options, on a separate slide in a Power. Point presentation.
The Experiment – Example Following is an example of a stimulus presentation slide (English glosses in curled brackets are later addition): . 2 [shvil] ( סלולה {latinization} {/ vil/ path (a narrow and/or unpaved road)}. " משמעות הפועל: "להוליך בשביל {Verb meaning: “to lead through a path”. }. דוגמת שימוש: המדריך ____ את החניכים {Usage example: The guide ____ the boy-scouts. } 7. שיבל 4. שבל 1. שבילל {/ i. vel/ (kitel)} {/ a. val/ (katal)} {/ vi. lel/ (ktilel)} 5. שבולל 2. השביל {/ vo. lel/ (ktolel)} {/hi. vil/ (hiktil)} 9. הוליך בשביל 6. שיבלל 3. אישבל {led through the path} {/ iv. lel/ (kitlel)} {/ i. vel/ ( iktel)} האזינו לכל אפשרויות הגזירה וסמנו בדף את זו שנראית לכם כמתאימה ביותר {Listen to all derivations and mark the most appropriate on the questionaire} לא ו/או צרה )דרך
The Experiment – Special Focus Special focus on hiktil in CCVC stimuli: • The number of CCi. C stimuli is twice that of other vowels. • Two binary non-phonological parameters are controlled: a) transitivity: whether or not the predicate takes a direct object (strict causativity was avoided, because of the risk of a ceiling effect). b) homophony: whether or not the Hebrew lexicon already has a hiktil verb with these three consonants. • Each combination of {vowel, ±trans, ±hom} is represented by two stimuli (four stimuli for CCi. C). • Homophony in other binyanim: For the 48 stimuli there were 13 potential katal homophones, 11 kitel, 5 kitlel and 2 potential iktel homophones.
CV. ’CVC Results • Hegemony of kitel (and kitlel) as expected. • Somewhat surprising viability of katal, maybe due to /CV. ’CVC/ skeletal faithfulness. • Negligible use of hiktil (and iktel). • Results are in line with all three theoretical approaches.
CV. ’CVC Results • Hegemony of iktel. • Mostly in line with the phonological approach. • The ‘Semitologist’ claim that the nouns contain an underlying / / cannot be ruled out. • iktel is also in line with the lexical distribution approach, but ktilel is not (yet it’s marginal).
CCV. ’CVC Results • CCV. ’CVC stimuli do not constitute a problem for derivation. • Equally co-dominated by 4 consonant kitel variants CVC. ’CVC (skeletally unfaithful, common in the lexicon) and CCV. ’CVC (skeletally faithful but nonexistent in the lexicon). • Reduplication is superfluous and appears only marginally.
CCVC Results • Pluralism of strategies. • Marginal use of katal. • Substantial increase in the popularity of hiktil, ktilel and ktolel, mostly at the expense of kitel. • Both the phonological and lexicalist approaches can explain the popularity of hiktil, but ktilel and ktolel can be explained only by the phonological approach.
CCe. C Results Relative to CV. ’CVC: • CCe. C is probably the ‘default’ CCVC source. • High popularity of hiktil and ktilel (/i/-faithful). • Slightly higher popularity of iktel (/e/-faithful, but root-prefixed). Might be non-significant. • Significant decrease in popularity of kitel, but remains viable. • Stability of kitlel.
CCa. C Results Relative to CCe. C: • Higher popularity of katal, tightly related to the /a/melody. Surprising given the unfaithful skeletal structure (katal is CCVCfaithful in future tense, where the vowel is /o/), and the non-productive status of katal. • Decrease in popularity of hiktil and ktilel. • Like CCe. C otherwise.
CCi. C Results Relative to CCe. C: • hiktil rules (/i/-faithful, Island of Reliability). • hiktil dominance explains decrease in katal, kitel, kitlel, ktolel and iktel. • Stability of ktilel (also /i/faithful). Unclear whether stability is genuine or an artifact of counter-balance between /i/-faithfulness (positive) and dominance of hiktil (negative).
CCo. C Results Relative to CCe. C: • ktolel (/o/-faithful) rules. • ktolel dominance explains the decrease in kitel, kitlel, iktel and hiktil. • Significant drop for ktilel, perhaps also for hiktil. • Probably higher popularity of katal, explainable by the future tense (/Ciktol/) or by lexical statistics. Might be non-significant.
CCUC Results Relative to CCe. C: • Similar to CCe. C, except for the ‘mirror image’ in ktolel vs. ktilel. • Back-rounded faithfulness explains ktolel popularity, while height difference underlies its more limited span than in CCo. C.
Transitivity Results • Transitivity facilitates derivation: 96% vs. 89% non-default responses. • Transitivity significantly increases hiktil (and also kitel) responses. • This increase is probably responsible for the decrease in katal, kitlel and ktilel responses (and not vice-versa).
Homophony Results • Results are based on responses for 24 (vs. 24) stimuli for hiktil, 13 (vs. 35) for katal and 11 (vs. 37) for kitel. • Quite predictably, potential homophony impedes derivation.
Analysis - Stability • ktolel, hiktil, ktilel and katal are very sensitive to the source vowel. • Conversely, ktilel, kitel and iktel are stable across different vowels. • This suggests (at least) two different derivation mechanisms. Notice in particular the difference between iktel and the four sensitive strategies.
Analysis – hiktil vs. ktilel • Phonologically, hiktilel h/k and ktilel are CC{eaou}C 71/507 63/507 1. 13 ![+trans (14. 0%) (12. 4%) equivalent (/CCi. C/ -hom] faithful). Why is there CC{eaou}C 61/184 17/184 3. 59 hiktil bias? [+trans (33. 2%) (9. 2%) a) morpho-syntactic and –hom] lexical knowledge. CCi. C 113/276 67/276 1. 69 ![+trans (40. 9%) (24. 3%) b) hiktil as island of –hom] reliability for CCi. C 54/92 8/92 6. 75* • When both factors [+trans (58. 7%) (8. 7%) are combined hiktil is –hom] as popular as kitel * 1. 13 x 3. 59 x 1. 69=6. 85≈6. 75 for CV. ’CVC nouns!!! Where is the catch …? ? ? ; -)
Analysis – ktilel vs. ktolel • For non-low vowel CCVC stimuli, ktilel and ktolel mirror each other: Strong bias towards the fully faithful vowel in CCi. C and CCo. C, and slightly weaker bias towards the backness&rounding faithful vowel in CCe. C and CCu. C. CCi. C ktilel ktolel 20. 4% 1. 9% CCe. C ktilel ktolel 21% 4. 5% i/(i+o) 91. 4% ktilel 4. 8% i/(i+o) 82. 2% ktilel 5. 1% CCa. C ktilel ktolel i/(i+o) 15. 2% 6. 4% 70. 3% CCu. C ktolel 19. 8% CCo. C ktolel 37. 3% i/(i+o) 19. 5% i/(i+o) 12. 0%
Analysis – hiktil vs. ktolel • This mirror imaging is only partially replicated for hiktil vs. ktolel in non-low vowel ![+trans –hom] CCVC stimuli. The CCi. C~CCo. C asymmetry is predicted by the /i/-hiktil island of reliability, but CCe. C~CCu. C asymmetry is unexplained. CCi. C hiktil ktolel 40. 9% 2. 2% CCe. C hiktil ktolel 20. 9% 6. 2% i/(i+o) 95. 0% i/(i+o) 77. 1% CCu. C hiktil ktolel 12. 3% 22. 1% CCo. C hiktil ktolel 9. 2% 39. 2% CCa. C hiktil ktolel i/(i+o) 13. 5% 7. 1% 65. 4% i/(i+o) 35. 7% i/(i+o) 19. 0%
Tentative Conclusions • The experiment elicited derivations successfully: only 8% of no-derivation answers, equally distributed between all conditions; only marginal over-creativity; the results are always interpretable by at least one theoretical approach. • The results for CV. ’CVC and VC. ’CVC stimuli are predictable by all approaches. As such they are not too informative, but reinforce the experimental design. • The popularity of kritel for CCV. ’CVC nouns, of katal for CCa. C, of ktolel for CCo. C and CCu. C, and of ktilel for CCi. C and CCe. C (and CCa. C) can only be explained by the universalist phonological approach: In choosing these strategies the speakers neither extract a consonantal root nor consult the existing lexicon.
![Tentative Conclusions • The phonological approach also explains hiktil~ktilel equivalence for CC{eaou}C ![+trans –hom]](https://present5.com/presentation/b7ff1d2b5f3c5dbaac83d60c2390631a/image-42.jpg "Tentative Conclusions • The phonological approach also explains hiktil~ktilel equivalence for CC{eaou}C ![+trans –hom]")
Tentative Conclusions • The phonological approach also explains hiktil~ktilel equivalence for CC{eaou}C ![+trans –hom] stimuli and the ktolel~ktilel mirror image for CC{ieou}C stimuli. • The differences between the phonologically equivalent hiktil and ktilel elsewhere are best explained by morphosyntactic and lexical knowledge, and demonstrate the (secondary) role of lexical statistics in derivation. • The stability of kitel, kitlel and iktel across all CCVC conditions is best explained by the root-extraction and/or the lexical distribution approaches. The account by the phonological approach, relying on /CV…VC/ as the optimal base (single-consonant edges) requires more ‘handwaving’, in particular for iktel.
Selected References Albright, A. (2002). ‘Islands of reliability for regular morphology: evidence from Italian. ’ Language 78, 684 -709. Albright, A. and Hayes, B. (2002). ‘Modeling English past tense intuitions with minimal generalizations’. In Maxwell, M. (ed): Proceedings of the 6 th Meeting of the ACL Special Interest Group in Computational Phonology. Philadelphia, July 2002. ACL. Bat-El, O. (1994 a). ‘System modification and cluster transfer in Modern Hebrew. ’ Natural Language and Linguistic Theory 12, 571 -593. Bat-El, O. (1994 b). ‘Resolving prosodic mismatch in Modern Hebrew verb formation. ’ In van der Hulst and van de Weijer (eds): Leiden in Last: HIL Phonology Papers I. The Hague: Holland Academic Graphics. 25 -40. Berent, I. , and Shimron, J. (1997). ‘The representation of Hebrew words: Evidence from the Obligatory Contour Principle. ’ Cognition 64, 39 -72. Berent, I. , Everett, D. L. , and Shimron, J. (2001). ‘Do phonological representations specify variables? Evidence from the Obligatory Contour Principle. ’ Cognitive Psychology 42(1), 1 -60. Berent, I. , Marcus, G. , Shimron, J. and Gafos, A. (2002). ‘The scope of linguistic generalizations: evidence from Hebrew word formantion. ’ Cognition 83, 113 -139.
Скачать презентацию How do Hebrew speakers trick or treat An
b7ff1d2b5f3c5dbaac83d60c2390631a.ppt