From phoneme to morpheme

“When computed across word boundaries, probabilities of sound pairs should generally be lower, reflecting the decreased constraints at boundaries noted by Harris (1955).”

“More-least to Harris (1955), who proposed that linguists working on unfamiliar languages might over, the structure of their test was such that even these results might be in part the outcomediscover the individual word units by counting how many phones can succeed a given string. of test strategies…”

“Indeed, most words in natural languages are comprised of consistent sound patterns among adjacent syllables, and the transitional probability relations we examined in our miniature language studies were similar to those exhibited in real human languages ( Harris, 1955 ).”

“This was first proposed by Zellig Harris (1955, 1967), and further developed by others, notably by Hafer and Weiss (1974).”

“These procedures were useful heuristic guides rather than algorithms (though see Harris, 1955).”

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

“Most previous work on statistical word segmentation is based on the observation that transitions from one syllable or phoneme to the next tend to be less predictable at word boundaries than within words (Harris, 1955; Saffran et al., 1996a).”

“Such a sequence of probabilities can indicate that /mamiz/ forms a word—Mommy’s—and the sounds that follow belong to other words (Harris, 1955; Hayes & Clark, 1970; Saffran et al., TABLE 2”

“Brent & Cartwright, 1996; Goodsitt, Morgan, & Kuhl, 1993; Harris, 1955; Hayes & Clark, 1970; Saffran, Aslin, & Newport, 1996.”

“Jacquemin [1997] discovers morphological variants of multiword collocations, for example, ‘longitudinal recording’ vs.”

“Indeed, much of this work is reminiscent of themes found in early associationist approaches to language acquisition (e.g., Harris, 1955 ).”

“The idea that words may be identified via distributional analysis is not new (e.g. Brent & Cartwright, 1996; Harris, 1955; Hayes & Clark, 1970; Pinker, 1984) but infants discovery of words has received less attention than infants discovery of phonetic categories.”

“Proponents of structural linguistics (Harris 1955) and information theory (Shannon 1948) have long recognized that the statistical information encoded in language provides cues to its constituent units (e.g., morphemes and words) and structural patterns.”

“Another distributional account assumes that chunking the speech stream into frequently occurring phoneme sequences will produce segmentation into linguistically coherent units (Brent & Cartwright, 1996; Perruchet & Vinter, 1998; Wolff, 1977; see Harris, 1955, for a related approach in linguistics).”

“Whereas within a word, a phoneme is followed by one out of 5–20 other phonemes (Harris, 1955), the number of possible successors of a word can reach the order of 10 4 .”

“How do infants learn to segment fluent speech into a series of words? Statistical strategies were proposed for the segmentation of words, based on distributional properties over sub-lexical units like phonemes or syllables (e.g., Harris, 1955; Brent & Cartwright, 1996; Dahan & Brent, 1999; Batchelder, 2002; Swingley,…”

“…sensitivity to statistical cues in the speech signal, in particular to transition probabilities (TPs), could in principle solve this problem (e.g., Harris, 1955; Hayes & Clark, 1970, but see Yang, 2004).1 Recent research has shown that adults and infants can indeed segment a continuous speech…”

“Studdert-Kennedy (1974) provided one of the earliest explicit articulations of this kind of mediated lexical access model, which itself drew inspiration from linguistic theory (see Bloomfield, 1933; Chomsky & Halle, 1968; Harris, 1955; Kenstowicz & Kisseberth, 1979; see also Jusczyk & Luce, 2002, for a discussion).”

“In the field of linguistics, a major goal was to devise rigorous scientific procedures that would yield a correct structural description of a particular language when provided with a detailed corpus of utterances (Bloomfield, 1933; Harris, 1955).”

“Usually and justifiedly, the work of Harris (1955, 1967) is given as the starting point of ULM.”

“Harris (1955) described a phoneme-based distributional learning approach that could help field linguists find morphemes in an unfamiliar language.”

“A further cue, suggested by early work in distributional linguistics (e.g. Harris, 1955), and utilised in the models described later, is the redundancy, or predictability of the speech stream.”

“The problem has been the focus of many years of intense research, and there are a large variety of proposals in the literature (Harris, 1955; Saffran et al., 1996a; Harris, 1955; Olivier, 1968; Saffran et al., 1996b; Brent, 1999b; Frank et al., 2010; Frank et al., 2013).”

“Rhythm class: According to Pike (1945) and Abercrombie (1967).”

“Across languages, syllable pairs with higher statistical coherence tend to belong to the same word (Harris, 1955; Hayes & Clark, 1970).”

“…linguists had long provided suggestions for what these cues might be: the words of the language are those sound sequences that regularly recur within a corpus, in contrast to word boundaries, which vary depending on what words happen to follow one another in a particular sentence (Harris, 1955 ).”

“…retain word-forms on the basis of how cohesive they are in terms of co-occurring sounds (cf. Saffran, Newport, & Aslin, 1996; Swingley, 2005a; the idea dates back to Harris, 1955), speech chunks such as gâteau and le gâteau might indeed be both retained as possible word-forms in a “protolexicon.””

“For instance, Harris (1955) develops a strategy for identifying morpheme boundaries that checks whether the number of different letters following a sequence of letters exceeds some given threshold.”

“In principle, Harris’s idea underlies both AV and BE; that is, the branching uncertainty is higher at a boundary in between language units such as morphemes or words than at other places.”

“One early proposal on how they might do this, originally due to Harris (1955) , was that they might be sensitive to the transition statistics of syllables or phonemes in natural language.”

“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).”

“One approach is using sequences of characters to identify morpheme boundaries— an idea first presented by Harris (1955)—either by an algorithmic approach (Hafer and Weiss 1974; Keshava and Pitler 2006) or developing n-gram-based statistical models (e.g., Spiegler et al. 2009).”

“Also relevant are studies by linguists in which the main motive is the development of tools for linguistic analysis (Gammon, 1969; Harris, 1955) and there is a sizeable literature on discovery procedures intended to have practical utility in artificial intelligence, robotics and pattern recognition…”

“The oldest published work on unsupervised word segmentation is due to Zelig Harris who suggested in the 1950 s a process to break a phonetic text into morphemes (Harris 1955) using only successors numbering.”

“Our algorithm gathers different approaches already applied to the word segmentation problem: a) use predictability of phonemes: the basic idea here, first suggested by [9], is that the number of distinct phonemes that are possible successors of the preceding string reduces rapidly with the length of that string unless…”

“This has been investigated especially with respect to segmenting words embedded in fluent speech, because sounds within a word are more predictable than sounds across word boundaries (e.g., Harris, 1955).”

“Other algorithms for unsupervised learning of word segmentation using corpus statistics are also known (Harris, 1955).”

“Harris (1955) suggested that the probability of hearing any two syllables in succession is higher if the syllables both occur within a word (for example, “baby”) than if the syllables occur as the last sound in one word and the first sound in the next (for example “-by doll”).”

“…have enjoyed a fair amount of attention for the last several decades (e.g., Brent & Cartwright, 1996; Gambell & Yang, 2005; Goldwater et al., 2009; Harris, 1955), as they reveal that the latter may be close enough to infants’ segmentation to make predictions that can be validated via direct…”

“To filter out left-leaning clusters, ParaMor adapts a strategy due to Harris (1955), who detects morpheme boundaries by examining word internal character variation.”

“Especially automatic approaches to the unsupervised detection of morphemes date back to the 1950s [116], and many different methods have been proposed over the last decade [117–119].”

“This idea is recurrent in research on the segmentation of words into morphemes (see for instance [4,5,6]).”

“For example, merely for the CRF model alone, there exist several proposed semi-supervised learning approaches (Jiao et al. 2006; Mann and McCallum 2008; Wang et al. 2009).”

“One possibility is that infants rely on a cross-linguistically consistent characteristic of words, namely, that syllables that comprise words cluster together with greater reliability than syllables that occur incidentally across word boundaries (Harris, 1955; Hayes & Clark, 1970).”

“2006]), and unsupervised methods (for example, [Harris 1955; Goldsmith 2001; Adda-Decker 2003; Creutz and Lagus 2005]).”

“-PA Author M anuscript N IH -PA Author M anuscript clusters of co-occurring sounds (Harris, 1955).”

“based algorithm [ 13 , 12, 10] described and implemented previously [4, 5]. The additional component of morpheme analysis is based on a prototypical implementation described in [7].”

“Finally, some conclusions and perspectives are given.”

“Another well-known conflation technique is the successor stemmer, based on the work of Harris (1955), segmenting utterances that are spelt phonetically, and creating trees whose labels correspond to a single character (Knuth, 1973).”

“This is interesting as it accords with some very early attempts to devise methods of segmenting speech into morphemes (Harris 1955).”

“This connection between morpheme segmentation, transitional probabilities, and mutual information goes back at least to Harris (1955).”

“Harris (1955) was the first to suggest that linguists can use distributional cues of segments to find word boundaries.”

“The character transition approach is exemplified by [8], which builds forward and backward character tries”

“After the pioneering work of Harris in the fifties (see, e.g., Harris 1955) and until very recently, modeling morpheme discovery has been a relatively unpopular research domain.”

“Second, the results of the learning algorithms developed here should be systematically evaluated against the predictions made by segmentation algorithms based purely on the distribution of word forms and segments (Harris 1955, 1967; Langer 1991; Gold-smith 2001).”

“Indeed, Harris [1] described such a system in the fifties.”

“This has be done through edit distance measures [8], maximum affix recognition [37,31,71], word insertion tests [39,28,10], and analogies [43,35,49]; – Another method relies on entropy models based on Harris’ hypothesis [34].”

“Conditional statistics measure the predictive relationships between consecutive events (Harris, 1955; Saffran, Aslin, & Newport, 1996).”