1 / 23

The influence of syllabic constituency on learning CV-affix vs. VC-affix associations: Constituency is more than depende

Vsevolod Kapatsinski Indiana University vkapatsi@indiana.edu. The influence of syllabic constituency on learning CV-affix vs. VC-affix associations: Constituency is more than dependency. Linguistic Society of America Annual Meeting January 3-6, 2008. Chicago, IL.

jabari
Télécharger la présentation

The influence of syllabic constituency on learning CV-affix vs. VC-affix associations: Constituency is more than depende

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Vsevolod Kapatsinski Indiana University vkapatsi@indiana.edu The influence of syllabic constituency on learning CV-affix vs. VC-affix associations:Constituency is more than dependency Linguistic Society of America Annual Meeting January 3-6, 2008. Chicago, IL

  2. Syllabic constituency in English • English syllables tend to have an onset-rime structure, rather than body-coda structure • Treiman and Danis (1988), Lee (2006): • If English speakers are presented with a list of syllables to memorize, they tend to make errors in which the onset of one syllable is combined with the rime of another syllable, rather than the body of one syllable being recombined with the coda from a different syllable. On1 On2 Cd1 Cd2 N2 Cd2 N1 Cd1 b   d  g b  gd   On1 N1 On2 N2 b   d  g d   b  g

  3. Tree-structure Rm Dependency On N Cd On N Cd st a nd st a nd Main question How should we represent syllabic constituency? On1 On2 N2 Cd2 N1 Cd1 b   d  g

  4. Rm On N Cd On N Cd st a nd st a nd Main question What is constituency? Tree-structural view (e.g., Fudge 1987) Constituency is unithood. A constituent is allocated a node in the hierarchy. Dependency-based view (e.g., Vennemann 1988) Constituency is connection strength. Connections between parts of a constituent are stronger than connections that cross constituent boundaries.

  5. What does it mean to say that the rime is allocated a node? • Under both tree-structural and dependency-based accounts, linguistic structures are represented with a network • A network is a set of nodes connected by links

  6. What does it mean to say that the rime is allocated a node? • In a network, links must connect nodes.  A node is something that can be linked to / associated with something else *

  7. What does it mean to say that the rime is allocated a node? • Thus, to say that the rime is allocated a node while the body is not means to say that the rime can be associated with something else while the body cannot (or at least cannot as easily) • I.e., the rime is more associable than the body *

  8. What does it mean to say that the rime is allocated a node? • So, the tree-structural account predicts that learning rime-affix associations should be easier than learning body-affix associations • For instance, learning that Cg stems take the affix /mn/ while C stems take /num/ should be easier than learning that gC stems take /mn/ while C stems take /num/. • The dependency-based account does not make this prediction.

  9. An analytical issue • How do we know that a particular association is an association of the whole and not an association of the parts? • If we find that the rime is more easy to associate with an affix than the body is, we need to make sure that this is not because the coda is more associable than the onset before we argue for a tree-structural model.

  10. Solution: XOR learning • The parts and the whole should not be associated with the same affix. • For instance, if // is paired with /num/ than /C/ would be paired with /mn/, as would /V/. • In general, Parts of the stem Affixes A X B A B C D C D Y ¬B ¬A ¬D ¬C N Cd On N

  11. Prefix Suffix mn-C C-mn • C  mn • Cg  num • C  mn • gC  num Prefix Suffix mn-C C-mn XOR learning: making parts and wholes compete Stage I Stage II • CC  num • CC  mn • CV  num • CVg  mn • CC  noom • CC  mn • VC  num • gVC  mn Subjects exposed to rime-affix pairings Subjects exposed to body-affix pairings 17 native English speakers in each condition (68 total)

  12. Methodological Details • During training, minimal pairs are presented to aid acquisition • ma – mn; mi – mn • ma – mn; mag – num • During test, min/noom replaced with noise • Amplitude contour, length average of mn/num • Tests involve generalization to novel syllables • During feedback, the entire phrase is pronounced when the subject makes the guess

  13. Prediction • Tree-structural • Rime exists, body does not • Rime more associable Accuracy on rimes should be better than on bodies Even if syllables containing familiar rimes or bodies are novel and Even if accuracy on codas and onsets does not differ //-/mn/ //-/mn/ /g/-/num/ /g/-/num/ and /mn/-// /mn/-// /num/-/g/ /num/-/g/ //-/num/ /g/-/mn/ /num/-// /mn/-/g/

  14. Rimes are more associable than bodies Complete generalization: Accuracy with novel syllables containing familiar rimes (72%) is not lower than accuracy with familiar syllables containing familiar rimes (69%). Subjects are not just memorizing syllables. No evidence for an adjacency effect: rime-prefix associations learned as well as rime-suffix ones. /mn/-// /mn/-// //-/mn/ //-/mn/

  15. Even though codas are as associable as onsets and subjects assigned to rime and body conditions are equally good learners The lack of a difference between onset and coda also suggests that the rime/body difference is not due to the fact that the rime occurs after the body.

  16. Summary • Rimes are more associable than bodies • This is not due to • Adjacency • Differences in associability between onsets and codas • Serial order • Differences in learning ability between subjects

  17. Theoretical implications • Rimes are more associable than the bodies • This is predicted by the tree-structural account of constituency but not by the dependency-based account • In general, the tree-structural account proposes that constituents are processing units: the chunks that are likely to be parsed out of the signal (the rime is more likely to be parsed out of the signal than the body) • This claim needs to be verified separately at different levels in the linguistic hierarchy. • One way in which this can be done is using XOR learning and related configural learning paradigms (e.g., biconditional discrimination, Saavedra 1975) whose essential feature is that the associations of the whole are not associations of the parts.

  18. An open question • What makes something likely to be extracted from the signal? • Overall statistics of the language • Overall, nucleus-coda co-occurrences are much less random than onset-nucleus co-occurrences in English while the opposite is true in Korean (Kessler and Treiman 1997, Lee 2006). • Accordingly, English speakers tend to recombine onsets and rimes in serial recall while Korean speakers tend to recombine bodies and codas (Lee 2006, Yoon and Derwing 2001) • even when statistics of co-occurrence within the syllables presented for memorization do not favor either the body or the rime (Lee 2006). Tree-structural model: • Rimes extracted in English • Bodies extracted in Korean

  19. An open question • What makes something likely to be extracted from the signal? • Between-part co-occurrence (Lee 2006) • Lee (2006): English speakers can show a tendency to recombine bodies and codas if syllables in the to-be-memorized list are carefully selected so that the nucleus co-occurs with the coda more often than it co-occurs with the onset • Potential lexicality(e.g., Cutler et al. 2001) • In an English syllable with a lax vowel, the body cannot be a free-standing word while the rime can. Thus, the rime is a possible word while the body is not  Not all rimes are equally likely to be extracted from the signal

  20. An open question • What makes something likely to be extracted from the signal? • Degree of coarticulation(e.g., Goldinger and Azuma 2003) • Usefulness in the task (e.g., Nosofsky et al. 1994) • A node representing a linguistic chunk may be parsed out when needed (when it has properties that are not inherited from its parts).  Not all instances of the same rime or body are equally likely to be extracted from the signal

  21. Conclusion • The results of the present study provide support for the view that syllabic constituents are more likely to be parsed out of the signal than non-constituents, the idea inherent in tree-structural representations of constituency. • There are multiple factors that can make something likely to be parsed out of the signal • Language-specific, syllable-specific, token-specific factors • Open question: do all factors that contribute to constituency contribute to parseability? • XOR learning, and configural learning more generally, presents a new way to assess the nature of constituent structure in various domains.

  22. Acknowledgments • NIH for funding • Nicholas Altieri for pronouncing the stimuli • Tessa Bent, Adam Buchwald and Rob Felty for helpful comments • Luis Hernandez for programming help

  23. References Bybee, J. L. 2002. Sequentiality as the basis of constituent structure. In T. Givon and B. F. Malle, eds. The Evolution of language out of pre-language. Benjamins. Cutler, A., McQueen, J. M., Norris, D., & Somejuan, A. 2001. The roll of the silly ball. In E. Dupoux (ed.), Language, Brain, and Cognitive Development: Essays in Honor of Jacques Mehler (pp.181-94). Cambridge, MA: MIT Press. Fudge, E. 1987. Branching structure within the syllable. Journal of Linguistics, 23, 359-377. Goldinger, S. D., and T. Azuma. 2003. Puzzle-solving science: The quixotic quest for units in speech perception. Journal of Phonetics, 31, 305-20. Healy, A.F. (1994). Letter detection: A window to unitization and other cognitive processes. Psychonomic Bulletin & Review, 1, 333-344. Kessler, B., and R. Treiman. 1997. Syllable structure and the distribution of phonemes in English syllables. Journal of Memory and Language 37: 295-311. Lee, Y. 2006. Sub-syllabic constituency in Korean and English, Unpublished Doctoral Dissertation, Northwestern University. Nosofsky, R. M., T. J. Palmeri, and S. C. McKinley. 1994. Rule-plus-exception model of classi Saavedra, M. A. 1975. Pavlovian conditioning in the rabbit. Learning and Motivation, 6, 314-26. Treiman, R., & Danis, C. 1988. Short-term memory errors for spoken syllables are affected by the linguistic structure of the syllables. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 145-52. Vennemann, T. 1988. The rule dependence of syllable structure. In On language: Rhetorica, Phonologica, Syntactica: A Festschrift for Robert P. Stockwell from his Friends and Colleagues, edited by C. Duncan-Rose and T. Vennemann, 257-83. London: Routledge. Yoon, Y. B., and B. L. Derwing. 2001. A language without a rhyme: Syllable structure experiments in Korean. Canadian Journal of Linguistics 46: 187-237.

More Related