Experimental evidence for product- oriented generalizations (or not) Vsevolod Kapatsinski Indiana University Dept. of Linguistics Cognitive Science Program.

Slides:



Advertisements
Similar presentations
The influence of domain priors on intervention strategy Neil Bramley.
Advertisements

Development of a German- English Translator Felix Zhang.
1 Rule reliability and productivity Velar palatalization in Russian and artificial grammar Vsevolod Kapatsinski Indiana University
Tone perception and production by Cantonese-speaking and English- speaking L2 learners of Mandarin Chinese Yen-Chen Hao Indiana University.
Contrastive Analysis, Error Analysis, Interlanguage
ANALYZING MORE GENERAL SITUATIONS UNIT 3. Unit Overview  In the first unit we explored tests of significance, confidence intervals, generalization, and.
18 and 24-month-olds use syntactic knowledge of functional categories for determining meaning and reference Yarden Kedar Marianella Casasola Barbara Lust.
Infant sensitivity to distributional information can affect phonetic discrimination Jessica Maye, Janet F. Werker, LouAnn Gerken A brief article from Cognition.
ASPECTS OF LINGUISTIC COMPETENCE 5 SEPT 11, 2013 – DAY 7 Brain & Language LING NSCI Harry Howard Tulane University.
Statistical NLP: Lecture 3
Do Children Pick and Choose? An Examination of Phonological Selection and Avoidance in Early Lexical Acquisition. Richard G. Schwartz and Laurence B. Leonard.
The Linguistics of SLA.
LIN 540G Second Language Acquistion
For more information, please write to: * This research was partially supported by the Israel Science Foundation (Grant No. 4806)
A Study of Speech Perception: Julie Langevin Communication Sciences and Disorders Faculty Mentor: Timothy Bryant The Psychological Reality of the Obligatory.
Rules and analogy in Russian loanword adaptation and novel verb formation Vsevolod Kapatsinski Indiana University Dept. of Linguistics & Cognitive Science.
PSY 369: Psycholinguistics Language Acquisition: Morphology.
Casenhiser and Goldberg (2005) Ability to learn to pair novel constructional meaning with novel form Known nouns and nonsense verb arranged in non- English.
Lecture 6 Outline: Tue, Sept 23 Review chapter 2.2 –Confidence Intervals Chapter 2.3 –Case Study –Two sample t-test –Confidence Intervals Testing.
Providence University College of Management Calculating and Reporting Wu-Lin Chen Department of Computer Science and Information Management.
Experimental evidence for product- oriented and source-oriented generalizations Vsevolod Kapatsinski Indiana University Dept. of Linguistics Cognitive.
Introduction Regular system: for every input, the grammar produces only one output Ways to achieve regularity Minimize competition between generalizations.
Language processing What are the components of language, and how do we process them?
Psych 156A/ Ling 150: Acquisition of Language II Lecture 13 Learning Biases.
Quantitative and Qualitative Approaches Dr. William M. Bauer
Chapter 12 Inferential Statistics Gay, Mills, and Airasian
323 Morphology The Structure of Words 1.1 What is Morphology? Morphology is the internal structure of words. V: walk, walk+s, walk+ed, walk+ing N: dog,
Modeling Language Acquisition with Neural Networks A preliminary research plan Steve R. Howell.
English Lexicology Morphological Structure of English Words Week 3: Mar. 10, 2009 Instructor: Liu Hongyong.
McEnery, T., Xiao, R. and Y.Tono Corpus-based language studies. Routledge. Unit A 2. Representativeness, balance and sampling (pp13-21)
Lemmatization Tagging LELA /20 Lemmatization Basic form of annotation involving identification of underlying lemmas (lexemes) of the words in.
Experimental study of morphological priming: evidence from Russian verbal inflection Tatiana Svistunova Elizaveta Gazeeva Tatiana Chernigovskaya St. Petersburg.
Jelena Mirković and Maryellen C. MacDonald Language and Cognitive Neuroscience Lab, University of Wisconsin-Madison Introduction How to Study Subject-Verb.
Morphology: Lexical category
Program Evaluation. Program evaluation Methodological techniques of the social sciences social policy public welfare administration.
Introduction Pinker and colleagues (Pinker & Ullman, 2002) have argued that morphologically irregular verbs must be stored as full forms in the mental.
Chapter 2 Developmental Psychology A description of the general approach to behavior by developmental psychologists.
CSA2050: Introduction to Computational Linguistics Part of Speech (POS) Tagging II Transformation Based Tagging Brill (1995)
Psycholinguistic Theory
Adele E. Goldberg. How argument structure constructions are learned.
The Role of Phonological Distance and Relative Support on the Productivity of the Dutch Simple Past Tense Bram Vandekerckhove, Emmanuel Keuleers, & Dominiek.
Statistical Learning in Infants (and bigger folks)
ANOVA and Linear Regression ScWk 242 – Week 13 Slides.
Modelling Language Evolution Lecture 1: Introduction to Learning Simon Kirby University of Edinburgh Language Evolution & Computation Research Unit.
Morphology An Introduction to the Structure of Words Lori Levin and Christian Monson Grammars and Lexicons Fall Term, 2004.
Educational Research Chapter 13 Inferential Statistics Gay, Mills, and Airasian 10 th Edition.
Aron, Aron, & Coups, Statistics for the Behavioral and Social Sciences: A Brief Course (3e), © 2005 Prentice Hall Chapter 12 Making Sense of Advanced Statistical.
1 Cross-language evidence for three factors in speech perception Sandra Anacleto uOttawa.
Cognitive Processes PSY 334 Chapter 11 – Language Structure June 2, 2003.
Lexical and morphosyntactic minimal pairs. Evidence for different processing Luca Cilibrasi, Vesna Stojanovik, Patricia Riddell, School of Psychology,
Introduction Chapter 1 Foundations of statistical natural language processing.
Morphological typology
Natural Language Processing Chapter 2 : Morphology.
3 Phonology: Speech Sounds as a System No language has all the speech sounds possible in human languages; each language contains a selection of the possible.
October 2004CSA3050 NLP Algorithms1 CSA3050: Natural Language Algorithms Morphological Parsing.
Y Letson 2007 (Miell et al 2002) Social Constructivist Approach to Learning.
Usage-based phonology Why are lines in grocery store about equal?
Variability in Interlanguage Session 6. Variability Variability refers to cases where a second language learner uses two or more linguistic variants to.
Presenter: Grace M. Wholley Advisor: Jessica F. Hay Department of Psychology, The University of Tennessee, Knoxville
Usage-Based Phonology Anna Nordenskjöld Bergman. Usage-Based Phonology overall approach What is the overall approach taken by this theory? summarize How.
FIRST AND SECOND LANGUAGE ACQUISITION/ LEARNING
عمادة التعلم الإلكتروني والتعليم عن بعد
Morphology Morphology Morphology Dr. Amal AlSaikhan Morphology.
Introduction to Statistics: Probability and Types of Analysis
Making Sense of Advanced Statistical Procedures in Research Articles
Saidna Zulfiqar bin Tahir STATE UNIVERSITY OF MAKASSAR
Quantitative and Qualitative Approaches Dr. William M. Bauer
Calculating and Reporting
Introduction to Qualitative Research
Presentation transcript:

Experimental evidence for product- oriented generalizations (or not) Vsevolod Kapatsinski Indiana University Dept. of Linguistics Cognitive Science Program Speech Research Laboratory

Product-oriented vs. source-oriented generalizations Bybee (2001:126) “Generative rules express source-oriented generalizations. That is, they act on a specific input to change it in well-defined ways into an output of a certain form. Many, if not all, schemas are product- oriented rather than source-oriented. A product- oriented schema generalizes over forms of a specific category, but does not specify how to derive that category from some other.” Source oriented: k] sg  t  i] pl Product-oriented: ‘plurals must end in t  i’

Present study Given an artificial lexicon and a particular training paradigm what generalizations do the learners extract?

The paradigm (Bybee & Newman 1995)

The artificial languages BLUERED {k;g}  {t  ;d  }i 100% 30 {t;d;p;b}  {t;d;p;b}i25% 8 75% 24 {t;d;p;b}  {t;d;p;b}a75% 24 25% 8 Two plural suffixes –i and -a If –i attached to a velar ({k;g}), the velar changes to an alveopalatal This is velar palatalization

Velar palatalization The process: k  t  /_i Productivity: p(k  t  i) / ( p(k  t  i) + p(k  ki) ) Coding scheme: BLUE – velar palatalization applies RED – velar palatalization fails

Research question Does the productivity of velar palatalization differ in the BLUE language and the RED language? Depends on your model of grammar.

The possible grammars BLUERED {k;g}  {t  ;d  }i 100% 30 {t;d;p;b}  {t;d;p;b}i25% 8 75% 24 {t;d;p;b}  {t;d;p;b}a75% 24 25% 8 /62

Non-competing rules BLUERED {k;g}  {t  ;d  }i 100% 30 {t;d;p;b}  {t;d;p;b}i25% 8 75% 24 {t;d;p;b}  {t;d;p;b}a75% 24 25% 8 Triggers velar palatalization Does not compete with anything Equally supported in both languages BLUE = RED e.g., Hale and Reiss 2008, Plag 2003

Simple positive product- oriented generalizations BLUERED {k;g}  {t  ;d  }i 100% 30 {t;d;p;b}  {t;d;p;b}i25% 8 75% 24 {t;d;p;b}  {t;d;p;b}a75% 24 25% 8 Triggers velar palatalization BLUE = RED Equally supported in both languages Bybee & Slobin 1982, Bybee & Moder 1983, Bybee 2001

Negative product-oriented generalizations BLUERED {k;g}  {t  ;d  }i 100% 30 *ki0 {t;d;p;b}  {t;d;p;b}i25% 8 75% 24 Ci3854 {t;d;p;b}  {t;d;p;b}a75% 24 25% 8 Triggers velar palatalization /ki/ less expected in the blue language  its absence is less notable BLUE < RED

Competing weighted rules BLUERED {k;g}  {t  ;d  }i 100% 30 C  Ci25% 8 75% 24 C  Ca75% 24 25% 8 Triggers velar palatalization BLUE > RED Competes with Competition stronger in red Albright & Hayes 2003 Iff the choice between the rules is stochastic.

Conditional product- oriented generalizations BLUERED {k;g}  {t  ;d  } | i 30/3830/54 {t;d;p;b}  {t;d;p;b} | i8/3824/54 {t;d;p;b}  {t;d;p;b} | a11 BLUE > RED Triggers velar palatalization More reliable in blue Aslin et al. 1998

Result BLUE RED

Results * 100% 30 BLUE RED Non-competing rules Simple positive product-oriented Negative product-oriented Competing weighted rules Conditional product-oriented

Individual subject data

Competing weighted rules BLUERED {k;g}  {t  ;d  }i 100% 30 C  Ci25% 8 75% 24 C  Ca75% 24 25% 8 Albright & Hayes 2003 {p;b;t;d}

22 Results *** Competing weighted rules ANCOVA: This correlation is significant F(1,27)=14.23, p<.001, while Language is not, F(1,27)=.082, p>.5). The predicted explanatory variable accounts for all the variance in velar palatalization rate attributable to the artificial language

The competing rules look good. Can we pit them against (conditional) product-oriented generalizations directly?

BLUERED {k;g}  {t  ;d  } | i Support vel.pal Support vel.pal {t  ;d  }  {t  ;d  } | i {t;d;p;b}  {t;d;p;b} | i8/3824/54 {t;d;p;b}  {t;d;p;b} | a11 Conditional product-oriented generalizations

Competing weighted rules BLUERED {k;g}  {t  ;d  }i 100% 30 {t  ;d  }  {t  ;d  }i Oppose vel.pal Oppose vel.pal C  Ci C  Ca75% 24 25% 8

Competing weighted rules r(33) = -.49, p=.003

The addition of t   t  i hurts vel.pal t(33)=2.88, p=.007 Competing weighted rules (Conditional) product-oriented

Something that looks product-oriented but isn’t Result: All subjects attach –i rather than –a to singulars ending in {t  ;d ʒ } In the Blue Language even more than to singulars ending in {k;g} H: Because both languages have plurals ending in {t  ;d ʒ }i, not {t  ;d ʒ }a. A product-oriented schema? ‘Plurals must end in {t  ;d ʒ }i’. BLUE RED

If –{t  ;d ʒ }  {t  ;d ʒ }i because of ‘Plurals must end in {t  ;d ʒ }i’, and this is the schema that does {k;g}  {t  ;d ʒ }i, Then –there should be a positive correlation between rate of {k;g}  {t  ;d ʒ }i and rate of {t  ;d ʒ }  {t  ;d ʒ }i But r=-.03, n.s. Why not?

It’s categorization of source forms. The more a subject attaches –i to velars, the more s/he attaches it to alveopalatals. Why more –i with {t  ;d ʒ } than with {k;g}? Subjects have a bias against stem changes.

Prior experimental evidence for product- oriented generalizations Frequent output patterns get ‘overused’, being derived from inputs in ways that are not attested in the lexicon, e.g., [v  n]  [v  ] (Bybee & Moder, 1983). (Also see Köpcke 1988, Lobben 1991, Wang and Derwing 1994, Albright and Hayes 2003) H: because the subjects have generalized ‘past tense forms must end in [  ]’

An alternative interpretation H’: The production of an output primes sublexical chunks occurring in that output. Lobben (1991) –“the plurals [that don’t obey the rules but all end in ooCii] are appearing concentrated and subsequently… and… this is a typical characteristic of all other plural patterns’ (Lobben 1991:173), – ‘[In this example] the second syllable of the singular is left out in the plural form, which never happens with real nouns… The surrounding… plurals, two preceding and seven following… are trisyllabic [in accordance with source-oriented rules]. This… provides an explanation as to why the plural [in this example], which, if produced according to the rule…, would have four syllables, is made to have three syllables in a very unorthodox way’ (Lobben 1991:182) Presupposition: the output is more salient than the input  chunks from the output are more likely to persist and be reused than chunks from the input

Are products more salient than sources? During training, subjects repeated the word pairs they heard. Subjects have a bias against stem changes  If the product is more salient, they should tend to erroneously make the shape of the singular fit the shape of the plural.  If the source is more salient, the plural should fit the singular.

Products are more salient t   t  i k  ki k  t  i repeated as χ 2 =28.9, p<.0001

35 In this AGL paradigm (Bybee & Newman 1995), Learners extract competing rules The outcome of competition is influenced by reliability or type frequency (Albright and Hayes 2003, Pierrehumbert 2006) The choice between rules is stochastic No evidence for product-oriented generalizations. Future work: Role of the training paradigm. Summary

36 Product forms are more salient than source forms. Thus creating a product may prime chunks and patterns that occur in that product for immediate reuse. This product priming may result in ‘overuse’ of frequent product patterns (found by Bybee & Moder 1983, Koepcke 1988, Lobben 1991, Wang & Derwing 1994, Albright & Hayes, 2003). If long-lasting, it may also result in the emergence of product-oriented schemas over time. Summary

References Albright, A., and B. Hayes Rules vs. analogy in English past tenses: A computational/experimental study. Cognition, 90, Aslin, R. N., J. R. Saffran, & E. L. Newport Computation of conditional probability statistics by 8-month-old infants. Psychological Science, 9, Bybee, J. L Phonology and language use. CUP. Bybee, J. L., & C. L. Moder Morphological classes as natural categories. Language, 59, Bybee, J. L., & J. E. Newman Are stem changes as natural as affixes? Linguistics, 33, Bybee, J. L., & D. I. Slobin Rules and schemas in the development and use of the English past. Language 58: Hale, M., & C. Reiss The phonological enterprise. OUP. Köpcke, K.-M Schemas in German plural formation. Lingua, 74, Lobben, M Pluralization of Hausa nouns, viewed from psycholinguistic experiments and child language data. M.Phil Thesis, University of Oslo. Pierrehumbert, J. B The statistical basis of an unnatural alternation. In Laboratory Phonology 8, Mouton de Gruyter. Plag, I Word formation in English. Mouton de Gruyter. Wang, H. S., & B. L. Derwing Some vowel schemas in three English morphological classes: Experimental evidence. In M. Y. Chen & O. C. L. Tseng, eds. In honor of Professor William S.-Y. Wang: Interdisciplinary studies on language and language change, Taipei: Pyramid Press.