Open AccessLexical knowledge boosts statistically-driven speech segmentation.
Author(s) -
Shekeila D. Palmer,
James Hutson,
Laurence White,
Sven L. Mattys
Publication year - 2019
Publication title -
journal of experimental psychology. learning, memory, and cognition
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.758
H-Index - 156
eISSN - 1939-1285
pISSN - 0278-7393
DOI - 10.1037/xlm0000567
Subject(s) - natural language processing , text segmentation , computer science , speech segmentation , word (group theory) , artificial intelligence , bootstrapping (finance) , task (project management) , segmentation , syllable , speech recognition , contrast (vision) , mechanism (biology) , linguistics , mathematics , philosophy , management , epistemology , economics , econometrics
The hypothesis that known words can serve as anchors for discovering new words in connected speech has computational and empirical support. However, evidence for how the bootstrapping effect of known words interacts with other mechanisms of lexical acquisition, such as statistical learning, is incomplete. In 3 experiments, we investigated the consequences of introducing a known word in an artificial language with no segmentation cues other than cross-syllable transitional probabilities. We started with an artificial language containing 4 trisyllabic novel words and observed standard above-chance performance in a subsequent recognition memory task. We then replaced 1 of the 4 novel words with a real word (tomorrow) and noted improved segmentation of the other 3 novel words. This improvement was maintained when the real word was a different length to the novel words (philosophy), ruling out an explanation based on metrical expectation. The improvement was also maintained when the word was added to the 4 original novel words rather than replacing 1 of them. Together, these results show that known words in an otherwise meaningless stream serve as anchors for discovering new words. In interpreting the results, we contrast a mechanism where the lexical boost is merely the consequence of attending to the edges of known words, with a mechanism where known words enhance sensitivity to transitional probabilities more generally. (PsycINFO Database Record (c) 2018 APA, all rights reserved).