Learning Parametric Stress without Domain-Specific Mechanisms

Aleksei Nazarov, Gaja Jarosz

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

State-of-the-art learning mechanisms for stress in Optimality Theory (see, e.g., Tesar and Smolensky 2000; Boersma and Pater 2016; Jarosz 2013) make use of probabilistic mechanisms that are domain-general in that they do not refer to the content of constraints and must not be in UG. By contrast, Pearl (2007, 2011) has argued that domain-general probabilistic learners of parametric grammars (Yang 2002) are insufficient for word stress, and, instead, domain-general learning mechanisms must be stipulated in UG alongside the parameters themselves. We propose a modification of Yang’s (2002) learner based on Jarosz’s (2015) learner for Optimality Theory: the Expectation Driven Parameter Learner, and show that this modification yields a dramatic improvement in accuracy (from 4.3% to 96%) on a representative typology generated by Dresher and Kaye’s (1990) parameter set. This suggests that domain-general learning mechanisms may be sufficient for learning stress after all, contra Pearl (2007, 2011), regardless of which grammatical representation (parameters or violable constraints) is a better reflection of the human language capacity.
Original languageEnglish
Title of host publicationProceedings of the 2016 Meeting on Phonology
EditorsKaren Jesney, Charlie O'Hara, Caitlin Smith, Rachel Walker
Place of PublicationWashington, DC
PublisherLinguistic Society of America
Number of pages12
DOIs
Publication statusPublished - 2017
Externally publishedYes
Event2016 Annual Meeting on Phonology - University of Southern California, Los Angeles, United States
Duration: 21 Oct 201623 Oct 2016
https://amp2016usc.wordpress.com/ (Link to Conference Website)

Publication series

NameProceedings of the Annual Meeting on Phonology
PublisherLinguistic Society of America
ISSN (Print)2377-3324

Conference

Conference2016 Annual Meeting on Phonology
CountryUnited States
CityLos Angeles
Period21/10/1623/10/16
Internet address

Fingerprint

learning
grammar
typology
language

Cite this

Nazarov, A., & Jarosz, G. (2017). Learning Parametric Stress without Domain-Specific Mechanisms. In K. Jesney, C. O'Hara, C. Smith, & R. Walker (Eds.), Proceedings of the 2016 Meeting on Phonology (Proceedings of the Annual Meeting on Phonology). Washington, DC: Linguistic Society of America. https://doi.org/10.3765/amp.v4i0.4010
Nazarov, Aleksei ; Jarosz, Gaja. / Learning Parametric Stress without Domain-Specific Mechanisms. Proceedings of the 2016 Meeting on Phonology. editor / Karen Jesney ; Charlie O'Hara ; Caitlin Smith ; Rachel Walker. Washington, DC : Linguistic Society of America, 2017. (Proceedings of the Annual Meeting on Phonology).
@inproceedings{4e38190a8de143058c4d31c0c20e30a5,
title = "Learning Parametric Stress without Domain-Specific Mechanisms",
abstract = "State-of-the-art learning mechanisms for stress in Optimality Theory (see, e.g., Tesar and Smolensky 2000; Boersma and Pater 2016; Jarosz 2013) make use of probabilistic mechanisms that are domain-general in that they do not refer to the content of constraints and must not be in UG. By contrast, Pearl (2007, 2011) has argued that domain-general probabilistic learners of parametric grammars (Yang 2002) are insufficient for word stress, and, instead, domain-general learning mechanisms must be stipulated in UG alongside the parameters themselves. We propose a modification of Yang’s (2002) learner based on Jarosz’s (2015) learner for Optimality Theory: the Expectation Driven Parameter Learner, and show that this modification yields a dramatic improvement in accuracy (from 4.3{\%} to 96{\%}) on a representative typology generated by Dresher and Kaye’s (1990) parameter set. This suggests that domain-general learning mechanisms may be sufficient for learning stress after all, contra Pearl (2007, 2011), regardless of which grammatical representation (parameters or violable constraints) is a better reflection of the human language capacity.",
keywords = "Principles and Parameters, Domain-General Learning, Expectation Maximization",
author = "Aleksei Nazarov and Gaja Jarosz",
note = "Copyright (c) 2017 Aleksei Nazarov, Gaja Jarosz",
year = "2017",
doi = "10.3765/amp.v4i0.4010",
language = "English",
series = "Proceedings of the Annual Meeting on Phonology",
publisher = "Linguistic Society of America",
editor = "Karen Jesney and Charlie O'Hara and Caitlin Smith and Rachel Walker",
booktitle = "Proceedings of the 2016 Meeting on Phonology",
address = "United States",

}

Nazarov, A & Jarosz, G 2017, Learning Parametric Stress without Domain-Specific Mechanisms. in K Jesney, C O'Hara, C Smith & R Walker (eds), Proceedings of the 2016 Meeting on Phonology. Proceedings of the Annual Meeting on Phonology, Linguistic Society of America, Washington, DC, 2016 Annual Meeting on Phonology, Los Angeles, United States, 21/10/16. https://doi.org/10.3765/amp.v4i0.4010

Learning Parametric Stress without Domain-Specific Mechanisms. / Nazarov, Aleksei; Jarosz, Gaja.

Proceedings of the 2016 Meeting on Phonology. ed. / Karen Jesney; Charlie O'Hara; Caitlin Smith; Rachel Walker. Washington, DC : Linguistic Society of America, 2017. (Proceedings of the Annual Meeting on Phonology).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Learning Parametric Stress without Domain-Specific Mechanisms

AU - Nazarov, Aleksei

AU - Jarosz, Gaja

N1 - Copyright (c) 2017 Aleksei Nazarov, Gaja Jarosz

PY - 2017

Y1 - 2017

N2 - State-of-the-art learning mechanisms for stress in Optimality Theory (see, e.g., Tesar and Smolensky 2000; Boersma and Pater 2016; Jarosz 2013) make use of probabilistic mechanisms that are domain-general in that they do not refer to the content of constraints and must not be in UG. By contrast, Pearl (2007, 2011) has argued that domain-general probabilistic learners of parametric grammars (Yang 2002) are insufficient for word stress, and, instead, domain-general learning mechanisms must be stipulated in UG alongside the parameters themselves. We propose a modification of Yang’s (2002) learner based on Jarosz’s (2015) learner for Optimality Theory: the Expectation Driven Parameter Learner, and show that this modification yields a dramatic improvement in accuracy (from 4.3% to 96%) on a representative typology generated by Dresher and Kaye’s (1990) parameter set. This suggests that domain-general learning mechanisms may be sufficient for learning stress after all, contra Pearl (2007, 2011), regardless of which grammatical representation (parameters or violable constraints) is a better reflection of the human language capacity.

AB - State-of-the-art learning mechanisms for stress in Optimality Theory (see, e.g., Tesar and Smolensky 2000; Boersma and Pater 2016; Jarosz 2013) make use of probabilistic mechanisms that are domain-general in that they do not refer to the content of constraints and must not be in UG. By contrast, Pearl (2007, 2011) has argued that domain-general probabilistic learners of parametric grammars (Yang 2002) are insufficient for word stress, and, instead, domain-general learning mechanisms must be stipulated in UG alongside the parameters themselves. We propose a modification of Yang’s (2002) learner based on Jarosz’s (2015) learner for Optimality Theory: the Expectation Driven Parameter Learner, and show that this modification yields a dramatic improvement in accuracy (from 4.3% to 96%) on a representative typology generated by Dresher and Kaye’s (1990) parameter set. This suggests that domain-general learning mechanisms may be sufficient for learning stress after all, contra Pearl (2007, 2011), regardless of which grammatical representation (parameters or violable constraints) is a better reflection of the human language capacity.

KW - Principles and Parameters

KW - Domain-General Learning

KW - Expectation Maximization

UR - https://journals.linguisticsociety.org/proceedings/index.php/amphonology/article/view/4010

UR - https://journals.linguisticsociety.org/proceedings/index.php/amphonology/issue/view/157

U2 - 10.3765/amp.v4i0.4010

DO - 10.3765/amp.v4i0.4010

M3 - Conference contribution

T3 - Proceedings of the Annual Meeting on Phonology

BT - Proceedings of the 2016 Meeting on Phonology

A2 - Jesney, Karen

A2 - O'Hara, Charlie

A2 - Smith, Caitlin

A2 - Walker, Rachel

PB - Linguistic Society of America

CY - Washington, DC

ER -

Nazarov A, Jarosz G. Learning Parametric Stress without Domain-Specific Mechanisms. In Jesney K, O'Hara C, Smith C, Walker R, editors, Proceedings of the 2016 Meeting on Phonology. Washington, DC: Linguistic Society of America. 2017. (Proceedings of the Annual Meeting on Phonology). https://doi.org/10.3765/amp.v4i0.4010