An acoustic-phonetic account of phonotactic perceptual assimilation

Abstract

Previous research has identified a coronal-to-dorsal ‘perceptual assimilation’ in which English and French listeners identify Hebrew word-initial /tl/ and /dl/ as beginning with /k/ and /g/, respectively (Hallé and Best, 2007). However, the acoustic-phonetic factors that contribute to this misperception have not been thoroughly identified, and previous results indicate that /tl/ is misperceived more often than /dl/—an asymmetry that is surprising on phonological grounds. The present study further explored this perceptual assimilation in two experiments with English listeners and Hebrew stop-liquid-vowel syllables ([t,k,d,g] × [l,ʁ] × [a,o,u]). The first experiment, which used the same stimuli as Hallé & Best, replicated previous findings, including the asymmetry between /tl/ and /dl/. The second experiment employed stimuli produced by a different native Hebrew speaker. While coronal-to-dorsal assimilation was observed, the previous /tl/-/dl/ asymmetry was not found: /dl/ was perceived as dorsal-initial somewhat more often than /tl/, suggesting that there can be no consistent phonemic or phonotactic explanation of the rate of assimilation. In support of a phonetic account, we find that misperception rates in both experiments are highly correlated (r > 0.65) with the stimulus-specific degree of anticipatory coarticulation of the lateral, as reflected in the spectral shape of the stop burst.

Long words in maximum entropy phonotactic grammars

“…Besides the admittedly metalinguistic task of acceptability judgments, phonotactic well-formedness has been implicated in various aspects of speech perception and speech production. For example, Davidson, Wilson and colleagues have made fairly explicit proposals for a Bayesian model of native and non-native speech perception that integrates acoustics and phonotactics (Davidson & Shaw 2012, Wilson & Davidson 2013, Chodroff & Wilson 2014, Wilson et al 2014). In Chodroff & Wilson (2014), the phonotactic grammar plays the role of a ‘prior model’, which calculates the likelihood of encountering e.g.…”

“…For example, Davidson, Wilson and colleagues have made fairly explicit proposals for a Bayesian model of native and non-native speech perception that integrates acoustics and phonotactics (Davidson & Shaw 2012, Wilson & Davidson 2013, Chodroff & Wilson 2014, Wilson et al 2014). In Chodroff & Wilson (2014), the phonotactic grammar plays the role of a ‘prior model’, which calculates the likelihood of encountering e.g. word-initial [bl] vs. word-initial [bəl]; this information is integrated into the acoustic model, which determines how well [bl] matches the observed acoustic sequence vs. how well [bəl] does.…”