Rules vs. analogy in English past tenses: A computational/experimental study. Albright, A. & Hayes, B. Cognition, 90(2):119-61, 2003. abstract bibtex Are morphological patterns learned in the form of rules? Some models deny this, attributing all morphology to analogical mechanisms. The dual mechanism model (Pinker, S., & Prince, A. (1998). On language and connectionism: analysis of a parallel distributed processing model of language acquisition. Cognition, 28, 73-193) posits that speakers do internalize rules, but that these rules are few and cover only regular processes; the remaining patterns are attributed to analogy. This article advocates a third approach, which uses multiple stochastic rules and no analogy. We propose a model that employs inductive learning to discover multiple rules, and assigns them confidence scores based on their performance in the lexicon. Our model is supported over the two alternatives by new "wug test" data on English past tenses, which show that participant ratings of novel pasts depend on the phonological shape of the stem, both for irregulars and, surprisingly, also for regulars. The latter observation cannot be explained under the dual mechanism approach, which derives all regulars with a single rule. To evaluate the alternative hypothesis that all morphology is analogical, we implemented a purely analogical model, which evaluates novel pasts based solely on their similarity to existing verbs. Tested against experimental data, this analogical model also failed in key respects: it could not locate patterns that require abstract structural characterizations, and it favored implausible responses based on single, highly similar exemplars. We conclude that speakers extend morphological patterns based on abstract structural properties, of a kind appropriately described with rules.
@Article{Albright2003,
author = {Albright, Adam and Hayes, Bruce},
journal = {Cognition},
title = {Rules vs. analogy in {E}nglish past tenses: {A} computational/experimental study.},
year = {2003},
number = {2},
pages = {119-61},
volume = {90},
abstract = {Are morphological patterns learned in the form of rules? Some models
deny this, attributing all morphology to analogical mechanisms. The
dual mechanism model (Pinker, S., & Prince, A. (1998). On language
and connectionism: analysis of a parallel distributed processing
model of language acquisition. Cognition, 28, 73-193) posits that
speakers do internalize rules, but that these rules are few and cover
only regular processes; the remaining patterns are attributed to
analogy. This article advocates a third approach, which uses multiple
stochastic rules and no analogy. We propose a model that employs
inductive learning to discover multiple rules, and assigns them confidence
scores based on their performance in the lexicon. Our model is supported
over the two alternatives by new "wug test" data on English past
tenses, which show that participant ratings of novel pasts depend
on the phonological shape of the stem, both for irregulars and, surprisingly,
also for regulars. The latter observation cannot be explained under
the dual mechanism approach, which derives all regulars with a single
rule. To evaluate the alternative hypothesis that all morphology
is analogical, we implemented a purely analogical model, which evaluates
novel pasts based solely on their similarity to existing verbs. Tested
against experimental data, this analogical model also failed in key
respects: it could not locate patterns that require abstract structural
characterizations, and it favored implausible responses based on
single, highly similar exemplars. We conclude that speakers extend
morphological patterns based on abstract structural properties, of
a kind appropriately described with rules.},
keywords = {Computing Methodologies, Human, Language, Learning, Mental Processes, Models, Theoretical, Stochastic Processes, Support, U.S. Gov't, Non-P.H.S., 14599751},
}
Downloads: 0
{"_id":"QGCSxbd2Bkri5LZFd","bibbaseid":"albright-hayes-rulesvsanalogyinenglishpasttensesacomputationalexperimentalstudy-2003","authorIDs":[],"author_short":["Albright, A.","Hayes, B."],"bibdata":{"bibtype":"article","type":"article","author":[{"propositions":[],"lastnames":["Albright"],"firstnames":["Adam"],"suffixes":[]},{"propositions":[],"lastnames":["Hayes"],"firstnames":["Bruce"],"suffixes":[]}],"journal":"Cognition","title":"Rules vs. analogy in English past tenses: A computational/experimental study.","year":"2003","number":"2","pages":"119-61","volume":"90","abstract":"Are morphological patterns learned in the form of rules? Some models deny this, attributing all morphology to analogical mechanisms. The dual mechanism model (Pinker, S., & Prince, A. (1998). On language and connectionism: analysis of a parallel distributed processing model of language acquisition. Cognition, 28, 73-193) posits that speakers do internalize rules, but that these rules are few and cover only regular processes; the remaining patterns are attributed to analogy. This article advocates a third approach, which uses multiple stochastic rules and no analogy. We propose a model that employs inductive learning to discover multiple rules, and assigns them confidence scores based on their performance in the lexicon. Our model is supported over the two alternatives by new \"wug test\" data on English past tenses, which show that participant ratings of novel pasts depend on the phonological shape of the stem, both for irregulars and, surprisingly, also for regulars. The latter observation cannot be explained under the dual mechanism approach, which derives all regulars with a single rule. To evaluate the alternative hypothesis that all morphology is analogical, we implemented a purely analogical model, which evaluates novel pasts based solely on their similarity to existing verbs. Tested against experimental data, this analogical model also failed in key respects: it could not locate patterns that require abstract structural characterizations, and it favored implausible responses based on single, highly similar exemplars. We conclude that speakers extend morphological patterns based on abstract structural properties, of a kind appropriately described with rules.","keywords":"Computing Methodologies, Human, Language, Learning, Mental Processes, Models, Theoretical, Stochastic Processes, Support, U.S. Gov't, Non-P.H.S., 14599751","bibtex":"@Article{Albright2003,\n author = {Albright, Adam and Hayes, Bruce},\n journal = {Cognition},\n title = {Rules vs. analogy in {E}nglish past tenses: {A} computational/experimental study.},\n year = {2003},\n number = {2},\n pages = {119-61},\n volume = {90},\n abstract = {Are morphological patterns learned in the form of rules? Some models\n\tdeny this, attributing all morphology to analogical mechanisms. The\n\tdual mechanism model (Pinker, S., & Prince, A. (1998). On language\n\tand connectionism: analysis of a parallel distributed processing\n\tmodel of language acquisition. Cognition, 28, 73-193) posits that\n\tspeakers do internalize rules, but that these rules are few and cover\n\tonly regular processes; the remaining patterns are attributed to\n\tanalogy. This article advocates a third approach, which uses multiple\n\tstochastic rules and no analogy. We propose a model that employs\n\tinductive learning to discover multiple rules, and assigns them confidence\n\tscores based on their performance in the lexicon. Our model is supported\n\tover the two alternatives by new \"wug test\" data on English past\n\ttenses, which show that participant ratings of novel pasts depend\n\ton the phonological shape of the stem, both for irregulars and, surprisingly,\n\talso for regulars. The latter observation cannot be explained under\n\tthe dual mechanism approach, which derives all regulars with a single\n\trule. To evaluate the alternative hypothesis that all morphology\n\tis analogical, we implemented a purely analogical model, which evaluates\n\tnovel pasts based solely on their similarity to existing verbs. Tested\n\tagainst experimental data, this analogical model also failed in key\n\trespects: it could not locate patterns that require abstract structural\n\tcharacterizations, and it favored implausible responses based on\n\tsingle, highly similar exemplars. We conclude that speakers extend\n\tmorphological patterns based on abstract structural properties, of\n\ta kind appropriately described with rules.},\n keywords = {Computing Methodologies, Human, Language, Learning, Mental Processes, Models, Theoretical, Stochastic Processes, Support, U.S. Gov't, Non-P.H.S., 14599751},\n}\n\n","author_short":["Albright, A.","Hayes, B."],"key":"Albright2003","id":"Albright2003","bibbaseid":"albright-hayes-rulesvsanalogyinenglishpasttensesacomputationalexperimentalstudy-2003","role":"author","urls":{},"keyword":["Computing Methodologies","Human","Language","Learning","Mental Processes","Models","Theoretical","Stochastic Processes","Support","U.S. Gov't","Non-P.H.S.","14599751"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://endress.org/publications/ansgar.bib","creationDate":"2020-01-31T01:09:12.403Z","downloads":0,"keywords":["computing methodologies","human","language","learning","mental processes","models","theoretical","stochastic processes","support","u.s. gov't","non-p.h.s.","14599751"],"search_terms":["rules","analogy","english","past","tenses","computational","experimental","study","albright","hayes"],"title":"Rules vs. analogy in English past tenses: A computational/experimental study.","year":2003,"dataSources":["SzgNB6yMASNi6tysA","xPGxHAeh3vZpx4yyE","5qXSH7BrePnXHtcrf","TXa55dQbNoWnaGmMq"]}