Transitional probabilities and positional frequency phonotactics in a hierarchical model of speech segmentation

Abstract

The present study explored the influence of a new metrics of phonotactics on adults' use of transitional probabilities to segment artificial languages. We exposed French native adults to continuous streams of trisyllabic nonsense words. High-frequency words had either high or low congruence with French phonotactics, in the sense that their syllables had either high or low positional frequency in French trisyllabic words. At test, participants heard lowfrequency words and part-words, which differed in their transitional probabilities (high for words, low for partwords) but were matched for frequency and phonotactic congruency. Participants' preference for words over partwords was found only in the high-congruence languages. These results establish that subtle phonotactic manipulations can influence adults' use of transitional probabilities to segment speech and unambiguously demonstrate that this prior knowledge interferes directly with segmentation processes, in addition to affecting subsequent lexical decisions. Implications for a hierarchical theory of segmentation cues are discussed.

Learning novel phonotactics from exposure to continuous speech

“…That is, even though performance in the ABC condition was not significant from chance, this does not necessarily imply that no learning took place in the ABC condition. Several studies have shown that artificial language learning from continuous speech is affected by phonological patterns in the native language (Boll-Avetisyan & Kager, 2014;Finn & Hudson Kam, 2008;Mersad & Nazzi, 2011;Onnis et al, 2005). If participants would somehow be biased towards BCA patterns, then ABC participants could have un-learned this bias (resulting in chancelevel performance), and BCA participants could have developed a preference for BCA patterns while ignoring statistical cues in the artificial language.…”

Statistical learning of language: Theory, validity, and predictions of a statistical learning account of language acquisition

“…Some mechanistic accounts of word segmentation conceptualize sensitivity to conditional statistical structure as arising from distinct mechanisms from those that produce sensitivity to acoustic cues such as lexical stress and phonotactic rules (e.g., Mersad & Nazzi, 2011). For example, an account put forth by Shukla et al (2007) suggested that computations are performed separately on transitional probability information and prosodic information.…”

“…Some models explicitly compute transitional probabilities (e.g., Gambell & Yang, 2004;Mersad & Nazzi, 2011), based on the typical assumption that learners compute transitional probabilities between elements . In contrast, other models do not explicitly compute transitional probabilities, despite the fact that they invoke processes that lead to the appearance of sensitivity to transitional probability (e.g., Perruchet & Vinter, 1998;Servan-Schreiber & Anderson, 1990).…”

“…One such criterion is psychological plausibility, based on what is known about the cognitive architecture in which these models ultimately must operate. One advantage of a chunking model such as PARSER (Perruchet & Vinter, 1998) is that it can explain sensitivity to conditional structure without explicitly calculating transitional probabilities as some models do (Mersad & Nazzi, 2011). PARSER simply pulls out chunks from the input and can learn the structure of input that can be described using transitional probabilities without the model ever learning anything about transitional probabilities.…”

“…For example, in the StaGE model (Adriaans & Kager, 2010), learners identify words via calculation of transitional probabilities and then deduce phonological and phonotactic constraints via a hierarchical ranking of constraints similar to that of Optimality Theory (Tesar & Smolensky, 2000). Similarly, Mersad and Nazzi (2011) proposed that learners use a hierarchical ranking of phonological cues to parse the speech stream, relying on sequential statistics only in cases where these cues are uninformative. In addition to invoking very different kinds of learning mechanisms and representations than the Extraction and Integration Framework, both of these accounts explain segmentation in terms of a hierarchy of cues, such that higherranking cues should always outrank lower-ranking cues.…”