Words In Fluent Speech Research Articles

Hebbian learning can explain rhythmic neural entrainment to statistical regularities.

In many domains, learners extract recurring units from continuous sequences. For example, in unknown languages, fluent speech is perceived as a continuous signal. Learners need to extract the underlying words from this continuous signal and then memorize them. One prominent candidate mechanism is statistical learning, whereby learners track how predictive syllables (or other items) are of one another. Syllables within the same word predict each other better than syllables straddling word boundaries. But does statistical learning lead to memories of the underlying words-or just to pairwise associations among syllables? Electrophysiological results provide the strongest evidence for the memory view. Electrophysiological responses can be time-locked to statistical word boundaries (e.g., N400s) and show rhythmic activity with a periodicity of word durations. Here, I reproduce such results with a simple Hebbian network. When exposed to statistically structured syllable sequences (and when the underlying words are not excessively long), the network activation is rhythmic with the periodicity of a word duration and activation maxima on word-final syllables. This is because word-final syllables receive more excitation from earlier syllables with which they are associated than less predictable syllables that occur earlier in words. The network is also sensitive to information whose electrophysiological correlates were used to support the encoding of ordinal positions within words. Hebbian learning can thus explain rhythmic neural activity in statistical learning tasks without any memory representations of words. Learners might thus need to rely on cues beyond statistical associations to learn the words of their nativelanguage. RESEARCH HIGHLIGHTS: Statistical learning may be utilized to identify recurring units in continuous sequences (e.g., words in fluent speech) but may not generate explicit memory forwords. Exposure to statistically structured sequences leads to rhythmic activity with a period of the duration of the underlying units (e.g., words). I show that a memory-less Hebbian network model can reproduce this rhythmic neural activity as well as putative encodings of ordinal positions observed in earlier research. Direct tests are needed to establish whether statistical learning leads to declarative memories for words.

Lexical stress constrains English-learning infants’ segmentation in a non-native language

Infants’ ability to segment words in fluent speech is affected by their language experience. In this study we investigated the conditions under which infants can segment words in a non-native language. Using the Head-turn Preference Procedure, we found that monolingual English-learning 8-month-olds can segment bisyllabic words in Spanish (trochees and iambs) but not French (iambs). Our results are incompatible with accounts that rely on distributional learning, language rhythm similarity, or target word prosodic shape alone. Instead, we show that monolingual English-learning infants are able to segment words in a non-native language as long as words have stress, as is the case in English. More specifically, we show that even in a rhythmically different non-native language, English-learning infants can find words by detecting stressed syllables and treating them as word onsets or offsets.

Do Infants Learn Words From Statistics? Evidence From English‐Learning Infants Hearing Italian

Infants are sensitive to statistical regularities (i.e., transitional probabilities, or TPs) relevant to segmenting words in fluent speech. However, there is debate about whether tracking TPs results in representations of possible words. Infants show preferential learning of sequences with high TPs (HTPs) as object labels relative to those with low TPs (LTPs). Such findings could mean that only the HTP sequences have a word-like status, and they are more readily mapped to a referent for that reason. But these findings could also suggest that HTP sequences are easier to encode, just like any other predictable sequence. Here we aimed to distinguish between these explanations. To do so, we built on findings that infants become resistant to learning labels that are not typical of their native language as they approach 2years of age and add words to their lexicons. If tracking TPs in speech results in identifying candidate words, at this age TPs may have reduced power to confer lexical status when they yield a unit that is very dissimilar to word forms that are typical of infants' native language. Indeed, we found that at 20months, English-learning infants with relatively small vocabularies learned HTP Italian words (but not LTP words) as object labels, while infants with larger vocabularies resisted learning HTP Italian words. These findings suggest that the HTP sequences may be represented as candidate words, and more broadly, that TP statistics are relevant to word learning.

Canadian oats and Canadian goats: Comparing distal cues to segmentation and segments

Distal timing cues (specifically, the speech rate of sentences temporally removed from a point of ambiguity in speech) have been shown to weakly modulate segmental perception [i.e., the perception of basic speech sounds, like “p” and “b”; Newman and Sawusch (1996)] but strongly affect the perception of word boundaries [i.e., the perception of separation between words in fluent speech; Dilley and Pitt (2010)]. However, no study as of yet has directly compared the two classes of percept using identical manipulations. In this study, we will examine the role of distal timing cues to segmental perception and word segmentation using the same distal contexts (e.g., “The merchant sold Canadian oats/notes,” a word boundary distinction, or “The merchant sold Canadian coats/goats,” a voicing-based segmental distinction). Distal speech rate will be artificially slowed to an identical extent for both types of contrast. We predict that categorical perception leads distal context effects to be much stronger on word segmentation (“…Canadian oats/notes”) than on segmental perception (“…Canadian coats/goats”). This research may help bridge speech perception theories that have been developed for each class of percept, and further clarify the role of distal cues in speech perception.

Discovering words in fluent speech: the contribution of two kinds of statistical information.

To efficiently segment fluent speech, infants must discover the predominant phonological form of words in the native language. In English, for example, content words typically begin with a stressed syllable. To discover this regularity, infants need to identify a set of words. We propose that statistical learning plays two roles in this process. First, it provides a cue that allows infants to segment words from fluent speech, even without language-specific phonological knowledge. Second, once infants have identified a set of lexical forms, they can learn from the distribution of acoustic features across those word forms. The current experiments demonstrate both processes are available to 5-month-old infants. This demonstration of sensitivity to statistical structure in speech, weighted more heavily than phonological cues to segmentation at an early age, is consistent with theoretical accounts that claim statistical learning plays a role in helping infants to adapt to the structure of their native language from very early in life.

A cross-linguistic investigation of word segmentation by monolingual infants learning Spanish and English

Infants' ability to find words in fluent speech is affected by their language experience. The focus of this study was the extent to which monolingual Spanish- and monolingual English-learning 8-month-old infants' segmentation abilities are tied to their familiarity with the rhythmic properties of their native language. Spanish and English are considered to belong to different rhythm classes; the former is syllable-timed, whereas the latter is stress-timed. However, in both languages CVC.CV words are predominantly trochaic—i.e., words with stress on the initial syllable (Cutler Carter, 1984; Pons Bosch, 2010). We tested segmentation of trochaic words of the form CVC.CV in Spanish and English. Spanish word segmentation was tested using the Headturn Preference Procedure and English word segmentation using the visual fixation procedure. According to Jusczyk et al.'s rhythm hypothesis, infants should fail to segment words in a rhythmically different language. Contrary to the rhythm hypothesis, preliminary data indicate that Spanish-learning infants are successful in segmenting trochees in both Spanish and English. Thus, successful transfer of word segmentation abilities by infants seems to depend upon a match in the prosodic shape of the target word across the two languages.

Cues and cue interactions in segmenting words in fluent speech

Fluent speech does not contain obvious breaks to word boundaries, yet there are a number of cues that listeners can use to help them segment the speech stream. Most of these cues have been investigated in isolation from one another. In previous work, Norris, McQueen, Cutler, and Butterfield (1997) suggested that listeners use a Possible Word Constraint when segmenting fluent speech into individual words. This constraint limits the word recognition system to consider only those parsings that could conceivably be words in the language (that is, those that do not strand illegal sequences). The present paper examines how this constraint interacts with other cues to segmentation, such as junctural and allophonic cues and neighborhood probabilities. Segmentation was influenced both by the PWC and by the presence of acoustic cues to juncture, such as the acoustic results of a speaker’s intention to produce a particular phoneme as the end of one syllable vs. as the start of another (vuff-apple vs. vuh-fapple). In contrast, segmentation was not affected by the legality of a syllable-final vowel (tense vs. lax), or by the similarity of a sequence to words. This suggests that acoustic cues in the signal play a far larger role in segmentation than do sources of bias from the lexicon, and that probabilistic lexical information from the lexicon (such as neighborhood information) is unlikely to be used in the process of word segmentation.

Is early word-form processing stress-full? How natural variability supports recognition

In a series of studies, we examined how mothers naturally stress words across multiple mentions in speech to their infants and how this marking influences infants’ recognition of words in fluent speech. We first collected samples of mothers’ infant-directed speech using a technique that induced multiple repetitions of target words. Acoustic analyses revealed that mothers systematically alternated between emphatic and nonemphatic stress when talking to their infants. Using the headturn preference procedure, we then tested 7.5-month-old infants on their ability to detect familiarized bisyllabic words in fluent speech. Stress of target words (emphatic and nonemphatic) was systematically varied across familiarization and recognition phases of four experiments. Results indicated that, although infants generally prefer listening to words produced with emphatic stress, recognition was enhanced when the degree of emphatic stress at familiarization matched the degree of emphatic stress at recognition.

Overcoming the Effects of Variation in Infant Speech Segmentation: Influences of Word Familiarity

Previous studies have shown that 7.5-month-olds can track and encode words in fluent speech, but they fail to equate instances of a word that contrast in talker gender, vocal affect, and fundamental frequency. By 10.5 months, they succeed at generalizing across such variability, marking a clear transition period during which infants' word recognition skills become qualitatively more mature. Here we explore the role of word familiarity in this critical transition and, in particular, whether words that occur frequently in a child's listening environment (i.e., "Mommy" and "Daddy") are more easily recognized when they differ in surface characteristics than those that infants have not previously encountered (termed nonwords). Results demonstrate that words are segmented from continuous speech in a more linguistically mature fashion than nonwords at 7.5 months, but at 10.5 months, both words and nonwords are segmented in a relatively mature fashion. These findings suggest that early word recognition is facilitated in cases where infants have had significant exposure to items, but at later stages, infants are able to segment items regardless of their presumed familiarity.

Influences of high and low variability on infant word recognition

Although infants begin to encode and track novel words in fluent speech by 7.5 months, their ability to recognize words is somewhat limited at this stage. In particular, when the surface form of a word is altered, by changing the gender or affective prosody of the speaker, infants begin to falter at spoken word recognition. Given that natural speech is replete with variability, only some of which determines the meaning of a word, it remains unclear how infants might ever overcome the effects of surface variability without appealing to meaning. In the current set of experiments, consequences of high and low variability are examined in preverbal infants. The source of variability, vocal affect, is a common property of infant-directed speech with which young learners have to contend. Across a series of four experiments, infants’ abilities to recognize repeated encounters of words, as well as to reject similar-sounding words, are investigated in the context of high and low affective variation. Results point to positive consequences of affective variation, both in creating generalizable memory representations for words, but also in establishing phonologically precise memories for words. Conversely, low variability appears to degrade word recognition on both fronts, compromising infants’ abilities to generalize across different affective forms of a word and to detect similar-sounding items. Findings are discussed in the context of principles of categorization that may potentiate the early growth of a lexicon.

English-Learning Infants' Representations of Word Forms With Iambic Stress.

Retaining detailed representations of unstressed syllables is a logical prerequisite for infants' use of probabilistic phonotactics to segment iambic words from fluent speech. The head-turn preference study was used to investigate the nature of English-learners' representations of iambic word onsets. Fifty-four 10.5-month-olds were familiarized to passages containing the nonsense iambic word forms ginome and tupong. Following familiarization, infants were either tested on familiar (ginome and tupong) or near-familiar (pinome and bupong) versus unfamiliar (kidar and mafoos) words. Infants in the familiar test group (familiar vs. unfamiliar) oriented significantly longer to familiar than unfamiliar test items, whereas infants in the near-familiar test group (near-familiar vs. unfamiliar) oriented equally long to near-familiar and unfamiliar test items. Our results provide evidence that infants retain fairly detailed representations of unstressed syllables and therefore support the hypothesis that infants use phonotactic cues to find words in fluent speech.

Some influences on the realization of for and four in American English

Words in fluent speech frequently differ from their citation forms. This paper investigates some factors influencing the realization of the preposition for and the number four. Realizations of these words in read and spontaneous speech show that instead of viewing function words as having discrete weak, strong, or even intermediate allomorphs, a continuum resulting from gestural interaction better represents the facts. In the speech examined here, numerous factors influence different aspects of the realization of for and four. The duration of the onset /f/ is affected by the preceding prosodic boundary while the duration and composition of the rime is affected by foot structure and segmental content. The preposition exhibits extreme reduction possibilities not seen in the number. In even this small set of studies, the read and spontaneous speech highlight different generalizations, pointing out the need to examine both kinds of speech. While this paper cannot locate all factors responsible for the variation in these words, it begins to unravel some of them and their effects on different portions of function words.

Some implications from language development for merge

Recent investigations indicate that, around 7-months-of-age, infants begin to show some ability to recognize words in fluent speech. In segmenting and recognizing words, infants rely on information available in the speech signal. We consider what implications these findings have for adult word recognition models in general, and for Merge, in particular.

The role of talker-specific information in word segmentation by infants.

Infants' representations of the sound patterns of words were explored by examining the effects of talker variability on the recognition of words in fluent speech. Infants were familiarized with isolated words (e.g., cup and dog) from 1 talker and then heard 4 passages produced by another talker, 2 of which included the familiarized words. At 7.5 months of age, infants attended longer to passages with the familiar words for materials produced by 2 female talkers or 2 male talkers but not for materials by a male and a female talker. These findings suggest a strong role for talker-voice similarity in infants' ability to generalize word tokens. By 10.5 months, infants could generalize different instances of the same word across talkers of the opposite sex. One implication of the present results is that infants' initial representations of the sound structure of words not only include phonetic information but also indexical properties relating to the vocal characteristics of particular talkers.

The Beginnings of Word Segmentation in English-Learning Infants

A series of 15 experiments was conducted to explore English-learning infants' capacities to segment bisyllabic words from fluent speech. The studies in Part I focused on 7.5 month olds' abilities to segment words with strong/weak stress patterns from fluent speech. The infants demonstrated an ability to detect strong/weak target words in sentential contexts. Moreover, the findings indicated that the infants were responding to the whole words and not to just their strong syllables. In Part II, a parallel series of studies was conducted examining 7.5 month olds' abilities to segment words with weak/strong stress patterns. In contrast with the results for strong/weak words, 7.5 month olds appeared to missegment weak/strong words. They demonstrated a tendency to treat strong syllables as markers of word onsets. In addition, when weak/strong words co-occurred with a particular following weak syllable (e.g., “guitar is”), 7.5 month olds appeared to misperceive these as strong/weak words (e.g., “taris”). The studies in Part III examined the abilities of 10.5 month olds to segment weak/strong words from fluent speech. These older infants were able to segment weak/strong words correctly from the various contexts in which they appeared. Overall, the findings suggest that English learners may rely heavily on stress cues when they begin to segment words from fluent speech. However, within a few months time, infants learn to integrate multiple sources of information about the likely boundaries of words in fluent speech.

Rapid gains in the speed and efficiency of word recognition by infants in the second year

Toward the end of the second year, during the period of rapid development in productive vocabulary known as the ‘‘vocabulary explosion,’’ infants also make dramatic gains in the speed and efficiency with which they are able to recognize familiar words in fluent speech. Here several studies using an on-line measure of speech processing by infants from 15 to 24 months of age are presented. In this research, an auditory-visual matching procedure, in which infants looked at two familiar pictures while hearing a target word that matched one of the pictures, was used. Monitoring infants’ eye movements from the onset of the target word, word recognition was assessed by measuring infants’ latency to look to the matching picture in response to the spoken word. These studies show that by the end of the second year, children are progressing toward the highly efficient performance of adults, using incomplete acoustic information to make rapid and accurate decisions about the identity of spoken words. [Work supported by NIMH.]

Effects of phonotactic probability on segmentation of words in continuous speech

At the last meeting of the Society, evidence was presented regarding the role of phonotactic probability in the segmentation of spoken words in continuous speech. Results of further studies examining this issue will be presented. Participants made speeded word detection responses to sequences of spoken stimuli composed of target words preceded and followed by nonwords (i.e., NONWORD–TARGET WORD–NONWORD). Speed and accuracy of detection were examined as a function of the nonword contexts of the target words. In particular, probabilities of segmental transitions from the nonword contexts to the target words were manipulated: Pairs of segments composed of the last segment of the preceding nonword and the first segment of the target word, as well as pairs composed of the last segment of the target word and the first segment of the following nonword, were varied in terms of intraword transition probability (HIGH and NONE) and position specific segment probability (HIGH and LOW). Only CC transitions were used. Both spliced and coarticulated stimuli were used. The implications of our results for the use of phonotactic probabilities in the identification of words in fluent speech will be discussed.

Effects of phonotactic probability on segmentation of words in continuous speech

The role of phonotactic probability in the segmentation of spoken words in continuous speech was investigated. Participants made speeded word detection responses to sequences of spoken stimuli composed of target words preceded and followed by nonwords (i.e., NONWORD–TARGET WORD–NONWORD). Speed and accuracy of detection were examined as a function of the nonword contexts of the target words. In particular, probabilities of segmental transitions from the nonword contexts to the target words were manipulated: Pairs of segments composed of the last segment of the preceding nonword and the first segment of the target word—as well as pairs composed of the last segment of the target word and the first segment of the following nonword—were varied in terms of intraword transition probability (HIGH and NONE) and transition type (CC, CV, VC, and VV). Both intraword co-occurrence probabilities and transitional probabilities were manipulated. The implications of the results for the use of phonotactic probabilities in the identification of words in fluent speech will be discussed.

Possible word boundary constraints on multiple activation of form‐based representations of spoken words.

Most current theories of spoken word perception (e.g., Cohort theory, TRACE, Shortlist, Neighborhood Activation Model) propose that multiple form‐based representations of words are activated in memory during recognition. Few of these theories, however, specify the precise nature of multiple activation of words in fluent speech or the constraints on activation imposed by information within the speech signal. The degree to which word boundary information in two‐word sequences may limit lexical hypotheses during recognition was examined. Using a cross‐modal priming technique, spoken two‐word utterances, such as NOTE‐RAIL, were presented and the activation of possible lexical items spanning the word boundary (e.g., TRAIL) was investigated. The implications of these results for current theories of spoken word recognition will be discussed. [Work supported by NIDCD.]

Extracting sentence structure from infant-directed speech

Although infants begin to encode and track novel words in fluent speech by 7.5 months, their ability to recognize words is somewhat limited at this stage. In particular, when the surface form of a word is altered, by changing the gender or affective prosody of the speaker, infants begin to falter at spoken word recognition. Given that natural speech is replete with variability, only some of which determines the meaning of a word, it remains unclear how infants might ever overcome the effects of surface variability without appealing to meaning. In the current set of experiments, consequences of high and low variability are examined in preverbal infants. The source of variability, vocal affect, is a common property of infant-directed speech with which young learners have to contend. Across a series of four experiments, infants’ abilities to recognize repeated encounters of words, as well as to reject similar-sounding words, are investigated in the context of high and low affective variation. Results point to positive consequences of affective variation, both in creating generalizable memory representations for words, but also in establishing phonologically precise memories for words. Conversely, low variability appears to degrade word recognition on both fronts, compromising infants’ abilities to generalize across different affective forms of a word and to detect similar-sounding items. Findings are discussed in the context of principles of categorization that may potentiate the early growth of a lexicon.