The Possible-Word Constraint in the Segmentation of Continuous Speech

Abstract

We propose that word recognition in continuous speech is subject to constraints on what may constitute a viable word of the language. This Possible-Word Constraint (PWC) reduces activation of candidate words if their recognition would imply word status for adjacent input which coultl not be a word -for instance, a single consonant. In two word-spotting experiments, listeners found it much harder to detect apple, for example, in fapple (where [f] alone would be an impossible word), than in vujfapple (where vuff could be a word of English). We demonstrate that the PWC can readily be implemented in a competition-based model of continuous speech recognition, as a constraint on the process of competition between candidate words; where a stretch of speech between a candidate word and a (known or likely) word boundary is not a possible word, activation of the candidate word is reduced. This implementation accu rately simulates both the present results and data from a range of earlier studies of speech segmentation. © 1997 Academic Press INTRODUCTIONWhen we listen to continuous speech, we apprehend a sequence of discrete words. The effortlessness with which we understand speech might suggest that recognition of spoken words is a rather simple task; but in fact it is a complex process, in which a very variable and largely continuous speech signal has to be mapped onto a lexicon of tens of thousands of words. TheWe thank the Longman Group, U.K. Inc. for allowing us to use a machine-readable version of the Longman Dictionary o f Contemporary English (Procter, 1975 Owned by Academic Press, 525 B Street, Suite 1900, San Diego, CA 92101-4495 Known bondholders, mortgagees, and other security holders owning or holding 1 percent or more of total amount o f bonds, mortgages, and other securities None Paragraphs 2 and 3 include, in cases where the stockholder or security holder appears upon the books o f the company as trustee or in any other fiduciary relation, the name o f the person or corporation for whom such trustee is acting, also the statements in the two paragraphs show the affiant's full knowledge and belief as to the circumstances and conditions under which stockholders and security holders who do not appear upon the books o f the company as trustees, hold stock and securities in a capacity other than that o f a bona fide owner Names and addresses of individuals who are stockholders o f a corporation which itself is a stockholder or holder of bonds, mortgages, or other securities of the publishing corporation have been included in paragraphs 2 and 3 when the interests o f such individuals are equivalent to I percent or more of the total amount of the stock or securities of the publishing corporation Total no copies printed average no copies each issue during preceding 12 months 2249, single issue nearest to filing date 2247 Paid circulation (a) to term subscribers by mail, carrier delivery, or by other means average no copies each issue during preceding 12 months 1054, single issue nearest to filing date 1050 (b) Sal...

Syntactic predictability in the recognition of carefully and casually produced speech.

“…We may therefore find that listeners are less able to Syntactic Predictability 9 make use of predictive syntactic information when the words that carry it are acoustically reduced because listeners may not have enough time to recover from the reduction of the auxiliary verb before hearing the participle. The idea that the importance of some linguistic cues depends on the availability of other cues is part of both accounts and is consistent with Mattys et al's (2005) and Norris et al's (1997) frameworks on cue integration in speech segmentation.…”

The Role of the Syllable in Lexical Segmentation in French: Word-Spotting Data

“…The finding of a target position effect is not specific to our study (Banel & Bacri, 1995;McQueen, 1998;McQueen et al, 1994;Norris, McQueen, Cutler, & Butterfield, 1997;van der Lugt, 2001). Assuming that target identification takes about the same time in both embedding positions, the RT difference suggests that for initial embedding, other processes take place after listeners have identified the target.…”

Lexical viability constraints on speech segmentation by infants

“…Although Norris et al (1997) found that adult segmentation was constrained by the PWC regardless of whether target words occurred at the end of the longer nonsense item (apple in ''fapple'' or ''vuffapple'') or at the beginning (sea in ''seash'' or ''seashub''), the response time advantage of possible over impossible contexts which they observed was largest for words in final position, with the context preceding the word. Studies of adult listening with other techniques have also shown greater inhibition of activation for finalembedded words preceded by impossible contexts than for initial-embedded words followed by impossible contexts (Vroomen & de Gelder, 1997).…”

“…Further research on adult word recognition has attempted to account for how everyday listeners successfully deal with unknown lexical items and rule out implausible segmentations of utterances containing such items. Norris, McQueen, Cutler, and Butterfield (1997) proposed that online segmentation and recognition of words is guided by a bias against considering any segmentation that leaves a single consonant as a residue.…”