Distributional regularity and phonotactic constraints are useful for segmentation

“Word-based Strategies Olivier (1968) proposed a procedure that maintains a lexicon and uses it for segmenting.”

“At one extreme, the phonological component might segment perfectly, as assumed in most attempts to explain the syntactic and semantic components of lexical acquisition (e.g., Pinker, 1984; Siskind, this issue).”

“It has been proposed that such regularities play a role in early word discovery (Cutler, 1994, 1996; Cutler, Mehler, Norris, & Segui, 1994).”

“Jusczyk and Aslin (1995) familiarized 7.5-month-old American infants with two diierent monosyllabic words and then presented them with passages that either did or did not contain the familiar words in sentences.”

“Jusczyk, Cutler, and Redanz (1993) have found that 9-month-old American infants listen longer to words with strong-weak stress (the predominant pattern in English) than to words with weak-strong stress, but 6-month-olds show no preference.”

“Similar studies conducted on French-learning infants at 12 months revealed a very different pattern: these infants were found to react to the onset of both the first and the second syllable of bisyllabic words, a pattern of data compatible with the use of the syllabic unit for word segmentation in…”

“…have been used for the evaluation of a number of segmentation models (Batchelder, 2002; Bernstein-Ratner, 1987; Blanchard et al., 2010; Brent, 1999; Brent & Cartwright, 1996; Goldwater et al., 2009; Johnson, 2008b; Johnson & Goldwater, 2009; Venkataraman, 2001), and we use it for consistency with…”

“This type of investigation will require an evaluation of the natural speech signal as opposed to transcribed speech, which has usually been favored in computational studies designed to examine the usefulness of multiple cues (e.g., Brent & Cartwright, 1996; Christiansen et al., 1998).”

“There is also extensive work on segmentation of the linguistic signal that does not rely on neural methods, and is not directly relevant here, (e.g., Brent and Cartwright, 1996; Goldwater et al., 2009; Kamper et al., 2016, and references there).”

“From an acoustic perspective, there are no obvious cues indicating where individual words begin and end (Brent & Cartwright, 1996; van de Weijer, 1998).”

“The process of word segmentation has received considerable attention also from the computational linguistics community, where various computational models have been proposed (e.g. (Brent and Cartwright, 1996; Goldwater et al., 2009)).”

“…work has shown that phonotactic constraints are learnable from transcriptions of continuous speech, and that such constraints have a positive effect on word segmentation (Adriaans & Kager, 2010; Blanchard et al., 2010; Brent & Cartwright, 1996; Cairns et al., 1997; Daland & Pierrehumbert, 2011).”

“Following this idea, Brent and coauthors (Brent, 1997; Brent and Cartwright, 1996) proposed that if a particular number of words are heard in isolation and are successfully retained, those words then will allow the language learner to bootstrap a lexical segmentation device.”

“…been successfully utilized in perceptual organization (Chater 1996; Hochberg and McAllister 1953; Leeuwenberg 1971; Pothos and Ward 2000), aspects of language acquisition (Brent and Cartwright 1996; Wolff 1977), memory (Attneave 1959; Garner 1974), and theories of similarity (Chater and Hahn 1997).”

“Elman (1990) introduced the word segmentation task as a simplified form of lexical acquisition, and Brent and Cartwright (1996) and Brent (1999) introduced the unigram model of word segmentation, which forms the basis of the model used here. Goldwater et al. (2009) described a non-parametric Bayesian model of word seg…”

“We discuss the relationship between our model and other accounts of lexical segmentation and word recognition, and compare the model’s performance to psycholinguistic data on the recognition of onset-embedded words.”

“Our line of work is further related to the one of Brent and Cartwright [58], who have shown that phonotactic constraints as top-level influence can guide segmentation in addition to distributional regularities.”

“Learning to segment words from raw speech is itself an unsolved problem [26–28], but understanding how segmentation occurs may be important to word learning if meaning can be ascribed to phrases before individual words.”

“Machine learning has contributed much to this discussion by showing that predictive models of language are capable of inferring the existence of word boundaries solely based on statistical properties of the input (Elman, 1990; Brent & Cartwright, 1996; Goldwater et al., 2009).”

“The approach has been used to analyse language learnability and acquisition in a number of contexts (e.g. Ellison, 1992; Brent & Cartwright, 1996; Grünwald, 1996; Clark, 2001; Goldsmith, 2001; Onnis, Roberts & Chater, 2002).”

“…constraints, i.e., constraints in the order of phones, are also observed to be helpful in word segmentation (Aslin, Saffran & Newport, 1998; Brent & Cartwright, 1996; Saffran, Newport & Aslin, 1996; among others, though see also Cairns, Shillcock, Chater & Levy, 1997; Mattys, Jusczyk, Luce…”

“…with distributional cues to many levels of linguistic structure, including phoneme distinctions and phoneme combinations, word boundaries, syntactic categories, and permissible orderings of words (e.g., Brent and Cartwright 1996; Cartwright and Brent 1997; Mintz et al. 2002; Vallabha et al. 2007).”

“Early influential datasets like CHILDES [30] facilitated initial explorations into language acquisition through linguistic recordings across diverse languages [1, 8].”

“…driven INC-DROP model (INcremental Distributional Regularity OPtimization) where familiar units drive segmentation of novel adjacent units in an iterative manner (Brent & Cartwright, 1996; Brent, 1997; see also Dahan & Brent, 1999 for behavioral validation of the model in adult segmentation).”

“Imposed constraints, such as the constraint that legal segmentations must contain a vowel (featured in previous models, such as that of Brent & Cartwright, 1996) were found to be less effective than the simple boundary constraint in further pilot runs.”

“The reoccurring phoneme sequences constituting a word can be isolated/distinguished from the rest of the acoustic/phoneme stream based on statistical discrimination [46, 47, 48].”

“…typically been generated by taking an orthographic transcription of speech and replacing each word with its pronunciation according to a dictionary (e. g., Brent & Cartwright, 1996; see, for example, B ¨ orschinger, Johnson, & Demuth, 2013; Elsner, Goldwater, Feldman, & Wood, 2013 for exceptions).”

“…text exhibit distributional regularities that can be computationally exploited for word boundary detection (Aslin, Woodward, LaMendola, & Bever, 1996; Cairns, Shillcock, Chater, & Levy, 1997; Christiansen, Allen, & Seidenberg, 1998; Elman, 1990; see also Brent & Cartwright, 1996; de Marcken, 1996).”

“…related to Bayesian induction, has been used for various learning and prediction tasks, including for natural language, in the works of Horning (1969), Berwick (1982), Ris-sanen and Ristad (1994), Stolcke (1994), Brent and Cartwright (1996), Gr¨unwald (1996), and de Marcken (1996), among others.”

“As this growth proceeds, words (or hypothesized proto-words) can aid in �nding additional words, through “segmentation by default” (Cutler, 1994), that is, if a word is identi�ed for sure, whatever comes after it is the start of another word (Brent and Cartwright, 1996).”

“Listeners would apply knowledge about phonotactics (Brent & Cartwright, 1996; Mattys, Jusczyk, Luce, & Morgan, 1999; Tagliapietra, Fanari, De Candia, & Tabossi, 2009), allophonic variation (Jusczyk, Hohne, & Bauman, 1999), constraints on possible word forms such as vowel harmony (Suomi, McQueen, &…”

“There have been an increasing number of models of segmentation, particularly within the last two decades (e.g., Aslin, 1993; Brent, 1999; Brent & Cartwright, 1996; Cairns, Shillcock, Chater, & Levy, 1994; Christiansen, Allen, & Seidenberg, 1998; Elman, 1990; Fleck, 2008; Goldwater, Griffiths, &…”

“Elles réfèrent à l’intuition que « les séquences de sons qui se produisent fréquemment et dans de nombreux contextes sont de meilleurs candidats pour le lexique que celles qui se produisent rarement et dans peu de contextes » (Brent & Cartwright, 1996).”

“Various symbolic segmentation models have been proposed (e.g., Brent & Cartwright, 1996; Goldwater, Griffiths, & Johnson, 2009) that use as input a phonemic transcription and return a hypothesized segmentation, to be compared against (orthographic) word boundaries.”

“…Saffran et al., 1996; Jusczyk et al., 1999; Suomi et al., 1997; van Kampen et al., 2008) and computational modeling (Cairns et al., 1994; Christiansen et al., 1998; Brent and Cartwright, 1996; Brent, 1999; Venkataraman, 2001; Xanthos, 2004; Goldwater et al., 2009; Johnson and Gold-water, 2009).”

“Unsupervised algorithms have been developed to learn a variety of levels of linguistic structure, such as word segmentation (Brent and Cartwright 1996), morphology (Goldsmith 2001, 2006), distributional part of speech classes (Schütze 1993; Redington et al.”

“Most existing models of language acquisition have been evaluated by artificially derived data of speech and semantics (Brent & Cartwright, 1996; Siskind, 1996; Regier, 1996; Cohen, Oates, Adams, & Beal, 2001) (but also see Roy & Pentland, 2002).”

“…Mehler, 2007; Soderstrom, Seidl, Nelson, & Jusczyk, 2003), sentence position (Seidl & Johnson, 2006), language-specific patterns of word formation (Brent & Cartwright, 1996; Toro, Pons, Bion, & Sebastián-Gallés, 2011), lexical stress (Houston, Jusczyk, Kuijpers, Coolen, & Cutler, 2000; Johnson &…”

“The Brent/Ratner corpus of English child-directed speech (Brent and Cartwright, 1996) has now become the standard dataset for the evaluation of computational models of early language acquisition.”

“…Allen, & Seidenberg, 1998), while other models focus on learning the lexicon itself, identifying the boundaries as a side effect (de Marcken, 1995; Brent & Cartwright, 1996; Perruchet & Vinter, 1998; Brent, 1999a; Venkataraman, 2001; Swingley, 2005; Goldwater, Griffiths, & Johnson, 2006a, 2009).”

“These models essentially learn language-specific information without pre-specifying what that information should be, as some have done (Brent & Cartwright, 1996; Gambell & Yang, 2005; Mattys, White & Melhorn, 2005).”

“Wikipedia + g2p me S ikoko iriburuko espet S e batean sartu zuten eta me S iko . . . English (Brent and Cartwright, 1996) ju want tu si D@ b U k l U k DErz @ b OI w ID h I z h æ t . . . Finnish (Aho, 1884) + g2p vai oli eilen kolmekymment æ kotoap æ ink ø se matti ajelee . . .”

“Goldwater, Griffiths, and Johnson (2009) generalized Brent and Cartwright (1996) ’ s proposal by showing that their implementation was formally equivalent to a unigram adaptor grammar, which in itself is a special case of adaptor grammars.”

“…been suggested that the distribution of phonetic-acoustic cues in a given language is tracked by infants in the early stages of language acquisition to form a native-speech segmentation mechanism prior to the emergence of word meaning (Aslin, Saffran and Newport 1998; Brent and Cartwright 1996).”

“…last two decades (just to exemplify a few, Elman, 1990, Aslin, 1993, Cairns et al., 1994, Christiansen, Allen, and Seidenberg, 1998, Fleck, 2008, Brent and Cartwright, 1996, Brent, 1999, Venkataraman, 2001, Xanthos, 2004, Gold-water, Griffiths, and M. Johnson, 2009, M. Johnson and Goldwater,…”

“…(2018, two experiments), and French et al. (2011), Equal Transitional-Probability experiment), as well as adult data from Perruchet and Desaulty (2008, two experiments), Giroux and Rey (2009), Frank, Goldwa-ter, Griffiths, and Tenenbaum (2010, two experiments), and Brent and Cartwright (1996).”

“Amongst other alternative statistical measures with varying levels of sophistication, we employ a technnique developed by [10], which relies on relatively uncontroversial phonotactic and learning strategies, as our focus is empirical plausibility in word detection.”

“Consonant sequences that violate word-internal phonotactic constraints provide cues to where word boundaries exist (Brent & Cartwright, 1996; Jusczyk, 1999).”

“Phonotactics play an important role in explaining patterns of word recognition in native-speaking adults (Brent & Cartwright 1996a,b; Li 1996; McQueen 1998; Mattys et al. 2005).”