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.”

“Also, because the listener’s goal is to identify sequences of sounds in fluent speech that match stored representations, the mental lexicon itself may play an important role in segmentation (Brent, 1999; Brent & Cartwright, 1996; Suomi, 1993).”

“Indeed, Brent and Cartwright (1996) suggested just such a constraint on initial word learning.”

“Brent and Cartwright (1996) propose that phonotactic constraints can be learned through inspection of consonant clusters that appear at utterance boundaries.”

“Many existing models of language acquisition have been evaluated by artificially derived data of speech and semantics (Brent and Cartwright 1996, Siskind 1996, Bailey et al. 1998, Cohen et al. 2001).”

“If the PWC has the same form across languages, it could assist infants in their initial attempts to segment continuous speech and acquire their rst words (Johnson, Jusczyk, Cutler & Norris, 2000; cf. Brent & Cartwright, 1996).”

“…that this combinatorial property of affixes should provide for highly efficient segmentation through chunking as demonstrated in similar theories of speech segmentation (e.g., Brent, 1997; Brent & Cartwright, 1996; see also Dahan & Brent, 1999) and visual word recognition (Davis, 1999, 2010).”

“For a human, knowing a language implies having a mental lexicon, i.e. a memorised set of associations among sound sequences, their meanings and their syntactic privileges (Brent and Cartwright 2002).”

“Moreover, the Simplicity Principle has been used practically in a wide range of models of language processing and structure (e.g., Brent & Cartwright, 1996; Dowman, 2000; Ellison, 1992; Goldsmith, 2001; Onnis, Roberts & Chater, 2002; Wolff, 1988).”

“Fourthly, it has long been recognized that the statistical regularities of phoneme co-occurrences are also reliable indicators of morpheme and word boundaries (Harris 1955; Brent and Cartwright 1996).”

“…lexical searches (e.g., Shukla et al. 2007, 2011), constraints on minimal word lengths such that all words contain at least one vowel (e.g., Brent & Cartwright 1996, Johnson et al. 2003), and possibly the Unique Stress Constraint (no word can contain more than one syllable with primary…”

“B&S further proposed that words discovered by being heard in isolation could then go on to assist in the discovery of other words using a kind of “remainder-is-a-new-potential-word” assumption (formalized in Brent & Cartwright, 1996).”

“Several computational demonstrations of word segmentation in acquisition ( Brent & Cartwright, 1996; Cairns, Shillcock,Chater, & Levy, 1997; Christiansen, Allen, & Seidenberg, 1998) have also confirmed the utility of sequential probabilities as cues to the location of word boundaries.”

“For example, Brent and Cartwright (1996) proposed a model based on Minimal Description Length.”

“The BR corpus is the same one that Brent (1999), Venkataraman (2001), Goldwater (2007), Fleck (2008) and Johnson & Goldwater (2009) used to evaluate their models, and it has become the de facto standard for segmentation testing ever since it was phonemicized by Brent & Cartwright (1996).”

“This was necessary in order to avoid having to deal with hypotheses that generate infinite sequences (which are possible, although more challenging technically), and also may better match the observation that utterance boundaries provide an important cue in segmentation (136).”

“…were construed so as to have a high to low vowel-consonant ratio and then arranging the letters into sequences that either did or did not conform to German phonotactic constraints (Brent and Cartwright, 1996; Saffran, 2003; Warker and Dell, 2006) according to the authors’ subjective judgment.”

“Most of the systems from the NLP community take as input a speech corpus that has been transcribed phonemically (gold transcription), but where the word boundaries have been deleted (Brent and Cartwright, 1996; Brent, 1999; Johnson et al., 2007; Goldwater et al., 2009).”

“Although smaller and larger constituents are sometimes considered as well (Batchelder, 2002; Bonatti, Pe~na, Nespor & Mehler, 2007; Brent & Cartwright, 1996; Monaghan et al., 2005), syllable has been predominantly recognized as a minimal perceptual unit in speech and has therefore been used as a…”

“Interestingly, the PWC is similar to a proposal made by Brent and Cartwright (1996) to improve performance of their computational model of vocabulary acquisition with a corpus of child-directed speech.”

“…principle for the scientist or as a learning criterion for the learner—notably, Berwick (1982), Rissanen and Ristad (1994), Stolcke (1994), Brent and Cartwright (1996), Grünwald (1996), de Marcken (1996), Clark (2001), Goldsmith (2001), Dowman (2007), Hsu and Chater (2010), Hsu, Chater,…”

“…task may involve numerous strategies incorporating acoustic (e.g. Christophe, Gout, Peperkamp, & Morgan, 2003; Jusczyk, Houston, & Newsome, 1999), phonotactic (e.g. Friederici & Wessels, 1993), distributional (e.g. Saffran, Newport, & Aslin, 1996) and pragmatic cues (e.g. Brent & Cartwright, 1996).”

“Instead of looking for units’ boundaries, the general strategy shared by all the chunk-based models (e.g., Brent & Cartwright, 1996; Frank, Goldwater, Griffiths, & Tenenbaum, 2010; Perruchet & Vinter, 1998; Robinet et al., 2011; Servan-Schreiber & Anderson, 1990) is that a large number of potential…”

“Likewise, it is clear that infants could use the phonotactic regularities of their languages, such as the fact that English words cannot begin with gd, to avoid erroneous segmentations such as bi gdog (e.g., Brent & Cartwright, 1996 ).”

“…(e.g., Aslin et al., 1998; Gleitman & Rozin, 1977; Mehler, 1981; Savin & Bever, 1970; Swingley, 2005; Thiessen & Saffran, 2003), that phonemes are (Brent & Cartwright, 1996; Lamel & Zue, 1984; Mattys & Jusczyk, 2001; McQueen, 1998; Norris, McQueen, Cutler, & Butterfield, 1997; among many others),…”

“…account, they detect the regular alternation and map it onto onsets, potentially because onsets are privileged perceptually and ⁄ or lexically (e.g., Brent & Cartwright, 1996; Jusczyk, Jusczyk, Kennedy , Schomberg, & Koenig, 1995; Magnuson, Dixon, Tanenhaus, & Aslin, 2007; Marslen-Wilson &…”

“The question is, Why does Parser find out the remaining third of the words, instead of the part words, which have the same distributional properties? Parser exploits here a property that is also exploited in other nonconnectionist models of word segmentation, such as the Brent and Cartwright (1996) model.”

“…tested on phonetic transcription data include a connectionist model by Christiansen, Allen, and Seidenberg (1998), a further MDL-based algorithm by Brent and Cartwright (1996), the somewhat similar but incremental probabilistic approach known as the Model-based dynamic programming (MBDP-1)…”

“TP-words and part-words, we know that although this factor is potentially important to word discovery by infants ( Brent & Cartwright, 1996 ) and can explain participants’ performance in some studies (see, e.g., Dahan & Brent, 1999), it cannot always explain the impact of TPs (Aslin, Saffran, & Newport, 1998).”

“Other investigators have pointed to the potential utility of phonotactic cues in word segmentation (Brent & Cartwright, 1996; Cairns, Shillcock, Chater, & Levy, 1997).”

“(2002) and Bulf et al. (2011). This study shows that there need not have been any statistical learning in those studies.”

“…and cognition (e.g., Attneave & Frost, 1969; Jacob Feldman, 2000; Hochberg & McAlister, 1953; Leeuwenberg, 1969), including language (e.g., Brent & Cartwright, 1996; Dowman, 2000; Ellison, 1992; Goldsmith, 2001; Onnis, Roberts & Chater, 2002; Vousden, Ellefson, Solity & Chater, 2011;…”

“Brent & Cartwright (1996) used it in order to account for the ability of children to segment a stream of phonemes into words.”

“Speech consists of a largely contiguous stream of auditory signals; the demarcations between the beginning and ending of words are frequently ambiguous in a purely acoustic analysis (Brent and Cartwright 1996; Christiansen et al. 1998; Klatt 1980).”

“…corpus is smaller than the Brent corpus that was used for the analysis of reduplicated words presented earlier, but it was selected here because of the availability of a phonologically represented derived corpus prepared by Brent and Cartwright (1996) and contributed to CHILDES by Sharon Goldwater.”

“Finally, the model was able to scale up to more realistic linguistic corpora , in particular, the Brent & Cartwright [21] data.”

“MDL-based systems have been relatively successful for tasks including word segmentation (de Marcken, 1996; Brent and Cartwright, 1996), morphological Figure 1: An example signature segmentation (Goldsmith, 2001; Creutz and Lagus, 2002), discovery of syllabicity and sonority (Elli-son, 1993), and…”

“Computational studies have confirmed that CDS constitutes input for word segmentation models that is superior to that of ADS (Aslin, Woodward, LaMendola, & Bever, 1996; Batchelder, 2002; Brent & Cartwright, 1996; Christiansen, Allen, & Seidenberg, 1998).”

“Early investigators, including Broadbent (1955), reported that listeners perceived a single “fused” syllable when a synthetic first formant was presented to one ear and a synthetic second formant (or second and third formant) to the other.”

“…between statistical learning and lexical knowledge in speech segmentation is lacking, there is at least one computational model directly relevant to this issue, namely the INCDROP (INCremental Distributional Regularity OPtimization; Brent & Cartwright, 1996; Brent, 1997; Dahan & Brent, 1999).”

“Instead, listeners were likely using multiple acoustic segmentation cues including allophonic variation, phonotactic constraints, and language-specific rhythmic properties (Brent & Cartwright, 1996; Cutler, Mehler, Norris, & Sequi, 1992; Jusczyk, 1999).”

“…has been used to develop learnability proofs for non-finite classes of grammar (e.g. Muggleton, 1996) and to develop theoretical and computational models of lexical and grammatical acquisition (e.g. Brent and Cartwright, 1996; de Marcken, 1996; Rissanen and Ristad, 1994; Osborne and Briscoe, 1997).”

“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 and Cartwright, 1996; Goldwater et al., 2009).”

“All of the cues we examined have the potential to aid word segmentation, and of course, might boost the effect when infants can draw on a combination of cues, as is the case in natural language (Brent & Cartwright, 1996; Matzinger et al., 2019).”

“The prevalence of short utterances, repetition, and paraphrase in IDS/CDS is one reason why computer simulations of language learning do very well when provided with IDS/CDS corpora to “learn” from.(19,20) Speech to children differs in the types of utterances that are most prevalent, as well.”

“Peters (1983) as well as Brent and Cartwright (1996) posit that children segment words from running speech using known words (or sequences of words) as access points. For example, in the phrase cognitive wugness, knowledge of the word cognitive can be used to segment and identify the word, wugness. In this way, each t…”

“Let us illustrate the point using the model of word segmentation developed by Brent and Cartwright (1996), in which segmentation is construed as an optimization problem.”

“Computational models support the hypothesis that transitional information can be sufficient for correct word segmentation (8, 9).”

“For example, even if it has been shown that distributional cues can be very effective for segmenting utterances into words (Brent and Cartwright 1996), there is no clear equivalent to semantically opaque morphemes in the domain of syntactic segmentation.”

“…a variety of guises might be used for extracting re-curring sequences has been implemented in numerous computational models (e.g., Batchelder, 2002; Brent & Cartwright, 1996; Christiansen, Allen, & Seidenberg, 1998; Frank, Goldwater, Griffiths, & Tenenbaum, 2010; Orb´an, Fiser, Aslin, & Lengyel,…”

“Indeed, the importance of UBs in word segmentation is supported by several computational models of early word segmentation (e.g., Aslin et al., 1996; Brent & Cartwright, 1996; Monaghan & Christiansen, 2010).”

“LaMendola & Bever, 1996; Brent & Cartwright, 1996; Christiansen, Allen & Seidenberg,”

“To do so, they must solve a word segmentation problem, since words are rarely uttered in isolation but come up in multi-word utterances (Brent and Cartwright, 1996).”

“…to the discovery of prosodic regularities (e.g., Jusczyk et al. 1993a, 1993b, 1994), to word segmentation with or without prosodic cues (e.g., Brent and Cartwright 1996; Christiansen et al. 1998; Sa¤ran et al. 1996a; Aslin et al. 1998; for adults Sa¤ran et al. 1996b), to word sequencing and…”

“In section 4 we explore research showing that young infants can track statistical regularities of language (Brent & Cartwright, 1996; Mattys & Jusczyk, 2001; Maye et al., 2003; Saffran, 2001, 2002, 2003; Saffran, Aslin, & Newport, 1996).”

“Brent and Cartwright (1996) improved the performance of their distributional model for word segmentation by requiring that all possible parses contain a vowel.”