Distributional regularity and phonotactics are useful for early lexical acquisition

In the course of language acquisition infants must segment connected speech into sound sequences that can be stored in the lexicon and eventually paired with meanings However there is no known acoustic analog of the blank space that separates printed words so it is not clear how infants can segment speech into words at the stage where most words are unfamiliar to them This paper investi gates two sources of information that might be useful for speech segmentation at the onset of lexical acquisition distributional regularity and phonotactics Infor mally distributional regularity refers to the intuition that sound sequences that occur frequently and in a variety of contexts are better candidates for the lexicon than those that occur rarely and in few contexts This paper begins by formalizing that intuition This formalization makes possible the development of algorithms that starting out without any lexical items acquire a lexicon from unsegmented utterances In addition three types of phonotactic constraints are investigated One of the three types is exempli ed by the fact that bigdog cannot be segmented into the two words bi and gdog because gdog is not a possible syllable of En glish and therefore it is not a possible word The other two constraints state that every word and every syllable must have a syllabic vowel like segment By applying computer implementations of lexical acquisition algorithms to phonetic transcripts of child directed English we show that both distributional analysis and two of the three phonotactic constraints can be used to signi cantly improve lex ical acquisition Further the contributions of these two information sources are not redundant so using both yields better lexical acquisition than using either one Correspondence to Michael Brent Dept of Cognitive Science Johns Hopkins University Baltimore MD USA Internet michael mail cog jhu edu This work has bene tted greatly from discussions with Lila Gleitman Peter Jusczyk Brian MacWhinney Gary Marcus Jacques Mehler Je Siskind and Paul Smolensky Special thanks are due to Sandor Brent for his careful reading of several drafts and thoughtful comments on them M R Brent T A Cartwright and A Gafos