Short Utterances Research Articles

Basic word segmentation abilities emerge earlier in infancy than previously thought.

English‐learning 7.5‐ but not 6‐month‐olds extract word forms from fluent speech (Jusczyk et al., 1999). Thus, English learners are thought to begin segmenting words from speech by 7.5 months. However, recent research has shown that when target words are flanked by a frequent and emotionally salient word (e.g., the infant’s name), even 6‐month‐olds can extract words from speech (e.g., Bortfeld et al., 2005). This suggests that basic word segmentation capabilities may emerge earlier than past studies have suggested. Using the Headturn Preference Procedure, we tested 6‐month‐olds’ ability to segment utterance‐flanked words from speech, e.g., “geff” from “At the circus we met a silly geff.” Infants were familiarized with passages containing target words in utterance initial and final position, and then tested on their recognition of these words in isolation. A significant looking time difference to familiar versus unfamiliar words was found, indicating that 6‐month‐olds segmented the target words from speech. Six‐month‐olds’ success at segmenting utterance‐flanked words from speech is particularly interesting because infant‐directed speech consists of short utterances containing many utterance‐flanked words. Segmentation of utterance‐flanked words could help infants learn the cues needed to extract harder utterance‐medial words from speech (see Seidl and Johnson, 2006).

Acoustical cues and grammatical units in speech to two preverbal infants

The current study examines the syntactic and prosodic characteristics of the maternal speech to two infants between six and ten months. Consistent with previous work, we find infant-directed speech to be characterized by generally short utterances, isolated words and phrases, and large numbers of questions, but longer utterances are also found. Prosodic information provides cues to grammatical units not only at utterance boundaries, but also at utterance-internal clause boundaries. Subject-verb phrase boundaries in questions also show reliable prosodic cues, although those of declaratives do not. Prosodic information may thus play an important role in providing preverbal infants with information about the grammatically relevant word groupings. Furthermore, questions may play an important role in infants' discovery of verb phrases in English.

Patterns of Discourse Coherence: Variations in Genre Performance in Children with Language Impairment

Background: The purpose of this investigation was to compare the discourse coherence of 36 children with language impairment (LI) who produced 3 types of genres: scripts, personal narratives, and procedural discourse. Method: The children described in random order a routine activity, personal experience, and a favorite game. The genres were analyzed for length, syntactic complexity, topic maintenance, informativeness, and fluency. Results: Scripts resulted in short, simple, and fluent utterances. Personal narratives and procedural discourse were similar in their length, informativeness, and fluency. Procedures were more syntactically complex and on topic than personal narratives. Conclusions: Children with LI are influenced by discourse genre. Clinical Implications: Different discourse genres should be compared in clinical assessments. Intervention should include different discourse genres in order to maximize a child's social, communicative, and classroom discourse.

Emotion Recognition in Speech Using Neural Network

Emotion recognition in speech studies has grown after the growing of speech recognition technologies. The aim of such studies is to make language interfaces in human-computer interaction applications more wide in use and to make it efficient. And it may help the studiers of the human sound areas. This study deals with four spoken sentences of sixteen short utterance expressing five emotions: a happiness, anger, sadness, fear, and normal (unemotional) state. The feature extraction techniques are used to capture the most important information of the signal, this process is applied in the time domain. The extracted feature vector contain 12 LPC (linear predictive coding) parameters, the pitch, the power, and the three first formant frequencies. While in the second part of the system (the emotion recognizer) the artificial neural network technology is used, the designing network is with feedforward backpropagation. This design used ten separated nets. each one make a decision whether the spoken sentence is belong to one of only tow emotions. So the computational time is increased with this system, while such a system has better performance with respect to other systems. The designed system could recognize all emotions with ratio 75% except the fear one which recognized in ratio of 50%.

Automatic Language Identification with Discriminative Language Characterization Based on SVM

Robust automatic language identification (LID) is the task of identifying the language from a short utterance spoken by an unknown speaker. The mainstream approaches include parallel phone recognition language modeling (PPRLM), support vector machine (SVM) and the general Gaussian mixture models (GMMs). These systems map the cepstral features of spoken utterances into high level scores by classifiers. In this paper, in order to increase the dimension of the score vector and alleviate the inter-speaker variability within the same language, multiple data groups based on supervised speaker clustering are employed to generate the discriminative language characterization score vectors (DLCSV). The back-end SVM classifiers are used to model the probability distribution of each target language in the DLCSV space. Finally, the output scores of back-end classifiers are calibrated by a pair-wise posterior probability estimation (PPPE) algorithm. The proposed language identification frameworks are evaluated on 2003 NIST Language Recognition Evaluation (LRE) databases and the experiments show that the system described in this paper produces comparable results to the existing systems. Especially, the SVM framework achieves an equal error rate (EER) of 4.0% in the 30-second task and outperforms the state-of-art systems by more than 30% relative error reduction. Besides, the performances of proposed PPRLM and GMMs algorithms achieve an EER of 5.1% and 5.0% respectively.

Revising segmentation hypotheses in first and second language listening

Abstract Any on-line processing that takes place while an utterance is unfolding is extremely tentative, with early-formed hypotheses having to be revised as the utterance proceeds. The hypotheses in question relate not only to the words that are present but also to where their boundaries fall. This study examines how first and second language listeners adjust their segmentation assumptions as new perceptual evidence comes in. It employs a variant of the gating task in which subjects transcribe a short utterance presented in sections of gradually increasing length. The first two presentations were phonetically ambiguous and could be segmented in any one of three ways. The third and fourth presentations provided disambiguating input; subjects’ responses were examined at these points to see how quickly they switched from a wrong interpretation to one that fitted the perceptual evidence. The results indicated a significant difference in the way in which first and second language listeners deal with incorrect segmentation hypotheses. Whereas native listeners are quick to change their interpretations on the basis of incoming evidence, non-native listeners are considerably more reluctant to do so.

Narrating and naming positive agents: Storytelling by Philadelphia postwar political elite

This paper connects research in social psychology, psycholinguistics, and discourse analysis to develop two hypotheses about character description. Interviews conducted in the late 1970s with political elites of Philadelphia, Pennsylvania, USA confirm predictions. Speakers describe co-group members as praiseworthy agents disproportionately, in comparison to out-group members, and refer to co-group members with individually specific language disproportionately. Extrapolating from research of similar practices in other contexts, the author explains why these two patterns may be important to identity formation. Images of positive agents are central to collective identity, so repeated stories portraying group members as such would reinforce those identities. Social psychological research suggests that specific, individual naming draws attention away from group identity; explicit group naming does the opposite. The use of individually specific names, rather than group names, may therefore mask the symbolic power created by the stories’ repeated descriptions of co-group members as positive agents. The language practice is both ubiquitous and opaque; it is common, and its power emerges from the aggregation of very short utterances. The combined practices of “narrating and naming” may exemplify the kind of misrecognition imagined by Pierre Bourdieu and Passeron (1990 [1977]). Together, these practices can reproduce or challenge social inequalities without being acknowledged for doing so.

Using SVM as Back-End Classifier for Language Identification

Robust automatic language identification (LID) is a task of identifying the language froma short utterance spoken by an unknown speaker. One of the mainstream approaches named parallel phone recognition language modeling (PPRLM) has achieved a very good performance. The log-likelihood radio (LLR) algorithm has been proposed recently to normalize posteriori probabilities which are the outputs of back-end classifiers in PPRLM systems. Support vector machine (SVM) with radial basis function (RBF) kernel is adopted as the back-end classifier. But for the conventional SVM classifier, the output is not probability. We use a pairwise posterior probability estimation (PPPE) algorithm to calibrate the output of each classifier. The proposed approaches are evaluated on the 2005 National Institute of Standards and Technology (NIST). Language recognition evaluation databases and experiments show that the systems described in this paper produce comparable results to the existing arts.

Learning to show you're listening

Good listeners generally produce back-channel feedback, that is, short utterances such as uh-huh which signal active listening. As the rules governing back-channeling vary from language to language, second-language learners may need help acquiring this skill. This paper is an initial exploration of how to provide this. It presents a training sequence which enables learners to acquire a basic Arabic back-channel skill, namely, that of producing feedback immediately after the speaker produces a syllable or two with a sharply falling pitch. This training sequence includes an explanation, audio examples, the use of visual signals to highlight occurrences of this pitch downslope, auditory and visual feedback on learners' attempts to produce the cue themselves, and feedback on the learners' performance as they play the role of an attentive listener in response to one side of a pre-recorded dialog. Experiments show that this enables learners to better approximate proper Arabic back-channeling behavior.

Nodding, aizuchi, and final particles in Japanese conversation: How conversation reflects the ideology of communication and social relationships

It has been noted that Japanese differs markedly from languages like English and Mandarin in the use of head nods and aizuchis (short utterances roughly equivalent to English “uh huh” and “yeah”). In Japanese conversation, such behaviors are extremely frequent, and their placement is often unexpected from the viewpoint of speakers of languages like English and Mandarin. For example, these behaviors often occur in non-transition relevant places. Sometimes aizuchis can even be uttered by the turn-holder. In such cases, the conventional technical terms such as “back-channel”, “continuer”, and “reactive token” are hardly applicable. Furthermore, the turn-holder often actively elicits aizuchis from the listener. Final particles, which are very frequent in spoken discourse, play an important role in the elicitation. Finally, there is a discussion of how the Japanese ideology of communication and social relationships may provide motivations for the above phenomena.

Robust speaker modeling using perceptually motivated feature

This paper introduces a novel method to extract robust features for text-independent speaker identification from short utterances. This method is perceptually motivated and inspired by the perceptual linear prediction (PLP) technique. The new feature is called perceptual log area ratio (PLAR). It is perceptual in the sense that it depends on notions from psychoacoustics where the robustness can be assured. Also, the log area ratio is an effective feature for recognizing speakers as it embodies the geometry and dynamics of the vocal tract, which are very much person-dependent. This research thus focuses on providing a reliable vocal biometric from speakers, which can be used effectively with full-band and telephone-band speech in noisy environments. Intensive performance analysis has been performed to benchmark the proposed method against the commonly-used features using different databases in different noisy environments. In almost all usable cases the PLAR proved its superiority over the commonly-used features such as MFCC and LPCC.

Morphological modifications in Lithuanian child-directed speech

Child-directed speech (CDS) is very important for the language development of a child. When talking to children adults modify their language by simplifying it and adapting it to the language acquisition stage of the child. Words and constructions that are used in CDS are aqcuired earlier by the child. This paper analyses the morphological modifications of Lithuanian CDS. The analysis proves that the most frequent morphological modifications of CDS include: 1) the higher frequency of verb tokens over noun tokens, 2) the low frequency of indefinite pronouns, demonstrative pronouns, and adjectives, and 3) the high frequency of diminutives and interjections. Prevalence of the denominative parts of speech has been influenced by the cognitive function; the abundance of verbs is conditioned by short utterances. The high frequency of diminutives is related to pragmatic function, intimacy, and emotions. DOI: http://dx.doi.org/10.5128/ERYa3.10

Cross-language perception of voice and affect

The paper reports on a cross-language study of how voice quality and f0 combine in the signaling of affect. Speakers of Irish-English and Japanese participated in perception tests. The stimuli consisted of a short utterance where f0 and voice source parameters were varied using the LF-model implementation of the KLSyn88a formant synthesizer, and were of three types: (1) VQ only involving voice quality variations and a neutral f0 contour; (2) f0 only, with different affect-related f0 contours and modal voice; (3) VQ+f0 stimuli, where the voice qualities of (1) combine with specific f0 contours from (2). Overall, stimuli involving voice quality variation were consistently associated with affect. In (2) only stimuli with high f0 yielded high affective ratings. Striking differences emerge between the ratings obtained from the two language groups. The results show that not only were some affects consistently perceived by one language group and not the other, but also that specific voice qualities and pitch contours were associated with very different affects across the two groups. The results have important implications for expressive speech synthesis, indicating that language/culture-specific differences need to be considered. [This work is supported by the EU-funded Network of Excellence on Emotion, HUMAINE.]

Is nonmodal phonation at the end of utterances a cue for speaker recognition by humans?

Many studies of nonmodal phonation report a high rate of interspeaker variation. These experiments also showed that the rate and type of glottalization are characteristic to the speaker. Still, it remains an open question whether human listeners use this information as a cue for recognizing familiar voices. A listening experiment was conducted to investigate this issue, concentrating on utterance-final irregularities. Short utterances produced by four speakers were copy synthesized with the Klatt synthesizer in two conditions: with modal and with nonmodal voice quality in their final regions. Because two of the speakers were reliable ‘‘glottalizers’’ and two seldom ‘‘glottalized,’’ one of the two synthesis conditions reflected the speaker’s usual utterance-final voice quality and the other did not. These pairs of stimuli were used to test the effect of the presence or absence of utterance-final nonmodal phonation on the identification of voices familiar to the subjects. When asked to choose which one of the two variants is the speaker’s voice, listeners preferred the stimuli with the speaker’s typical voice quality at the end. Results will be discussed in comparison with a well-known perceptual cue on speaker identity, the F0 contour. [Work supported by a Fulbright Fellowship.]

Measurement of temporal changes in vocal tract area function from 3D cine-MRI data

A 3D cine-MRI technique was developed based on a synchronized sampling method [Masaki et al., J. Acoust. Soc. Jpn. E 20, 375-379 (1999)] to measure the temporal changes in the vocal tract area function during a short utterance /aiueo/ in Japanese. A time series of head-neck volumes was obtained after 640 repetitions of the utterance produced by a male speaker, from which area functions were extracted frame-by-frame. A region-based analysis showed that the volumes of the front and back cavities tend to change reciprocally and that the areas near the larynx and posterior edge of the hard palate were almost constant throughout the utterance. The lower four formants were calculated from all the area functions and compared with those of natural speech sounds. The mean absolute percent error between calculated and measured formants among all the frames was 4.5%. The comparison of vocal tract shapes for the five vowels with those from the static MRI method suggested a problem of MRI observation of the vocal tract: data from static MRI tend to result in a deviation from natural vocal tract geometry because of the gravity effect.

Language identification using acoustic log-likelihoods of syllable-like units

Automatic spoken language identification (LID) is the task of identifying the language from a short utterance of the speech signal uttered by an unknown speaker. The most successful approach to LID uses phone recognizers of several languages in parallel [Zissman, M.A., 1996. Comparison of four approaches to automatic language identification of telephone speech. IEEE Trans. Speech Audio Process. 4 (1), 31–44]. The basic requirement to build a parallel phone recognition (PPR) system is segmented and labeled speech corpora. In this paper, a novel approach is proposed for the LID task which uses parallel syllable-like unit recognizers, in a frame work similar to the PPR approach in the literature. The difference is that the sub-word unit models for each of the languages to be recognized are generated in an unsupervised manner without the use of segmented and labeled speech corpora. The training data of each of the languages is first segmented into syllable-like units and language-dependent syllable-like unit inventory is created. These syllable-like units are then clustered using an incremental approach. This results in a set of syllable-like units models for each language. Using these language-dependent syllable-like unit models, language identification is performed based on accumulated acoustic log-likelihoods. Our initial results on the Oregon Graduate Institute Multi-language Telephone Speech Corpus [Muthusamy, Y.K., Cole, R.A., Oshika, B.T., 1992. The OGI multi-language telephone speech corpus. In: Proceedings of Internat. Conf. Spoken Language Process., October 1992, pp. 895–898] show that the performance is 72.3%. We further show that if only a subset of syllable-like unit models that are unique (in some sense) are considered, the performance improves to 75.9%.

Metrical phonology in HPSG

This paper proposes a new approach to the prosody-syntax interface in HPSG. Previous approaches to prosody in HPSG (Klein, 2000; Haji-Abdolhosseini, 2003) represent prosodic information by constructing metrical constituent structure in the tradition of (Selkirk, 1980; Liberman and Prince, 1977). One drawback of this approach is that it does not allow for a direct representation of purely metrical constraints, which are relegated to an unformalized performance component. By contrast, so called 'grid only' approaches (Prince, 1983; Selkirk, 1984; Delais-Roussarie, 2000) use a single data structure, a metrical grid, to encode prosodic constraints resulting from syntax and constraints of a rhythmic nature. We first review relevant data from French showing that prosodic constituency is much less constrained by syntactic structure than is predicted by existing approaches. In all but very short utterances, many different prosodic groupings are possible for a given sentence with a determinate information structure, and rhythmic factors determine a preference ordering on the possible groupings. We then present an HPSG implementation of the metrical grid, and propose minimal syntactic constraints on relative prominence, leaving room for noncategorical rythmic constraints to choose between alternatives. We finish by discussing the interaction of the metrical grid with the rest of the prosodic grammar.

Semantic data mining of short utterances

This paper introduces a methodology for speech data mining along with the tools that the methodology requires. We show how they increase the productivity of the analyst who seeks relationships among the contents of multiple utterances and ultimately must link some newly discovered context into testable hypotheses about new information. While, in its simplest form, one can extend text data mining to speech data mining by using text tools on the output of a speech recognizer, we have found that it is not optimal. We show how data mining techniques that are typically applied to text should be modified to enable an analyst to do effective semantic data mining on a large collection of short speech utterances. For the purposes of this paper, we examine semantic data mining in the context of semantic parsing and analysis in a specific situation involving the solution of a business problem that is known to the analyst. We are not attempting a generic semantic analysis of a set of speech. Our tools and methods allow the analyst to mine the speech data to discover the semantics that best cover the desired solution. The coverage, in this case, yields a set of Natural Language Understanding (NLU) classifiers that serve as testable hypotheses.

Neutralization or truncation? The perception of two Russian pitch accents on utterance-final syllables

This paper presents the results of a perception experiment that was carried out to verify the hypothesis that in Russian the contrast between pitch accents LH*L and LH* on utterance-final syllables is neutralized. Recordings for the experiment were 10 sets of three short utterances with word stress in the ultimate, penultimate and antepenultimate syllable of the utterance-final word. These utterances were read aloud by four female and four male native speakers. They were asked to realize accents LH*L and LH* in the utterance-final word. After instructions and rehearsing, recordings were made separately for each of the two types. In the perception experiment, 30 native subjects listened to short utterances selected from the recordings and presented in 180 pairs: 120 pairs with ultimate stress and, in order to test whether listeners can hear the difference at all, 60 pairs with penultimate and antepenultimate word stress in utterance-final position. The 180 stimuli pairs consisted of short utterances with realizations of LH*L and LH* on the final word, each pair containing two same or two different types of pitch accent. The task was to compare two stimuli in a pair and to indicate on a score form whether two realizations in a stimulus pair count as passable imitations of each other and thus belong to the same type of pitch accent. The same/different judgments indicate that listeners successfully distinguished between the two pitch accents in the antepenultimate and penultimate conditions, but much less so in the ultimate condition. This suggests that the two accents are truncated in final position, but not neutralized.

Argumentation with restricted linguistic ability: Performing a role play with aphasia or in a second language

The main objective of this study is to illustrate how adaptation to linguistic limitations takes place in a specific activity and is affected by factors pertaining to the social activity or the individuals. A man with aphasia is compared to an adult immigrant L2 learner. An argumentative role play was video‐recorded, transcribed and analysed. Both subjects have a very limited vocabulary and produce short utterances. The L2 learner often uses words that are semantically related to the target word, while the subject with aphasia uses more general and vague words, like pronouns, in combination with adverbs and set phrases. Both subjects use gesturing as strategy, and it is suggested that the semantic specificity of words as well as gestures is important in determining the role of gesture. Apart from gesture, he L2 learner uses mainly simplification and appeal strategies, while the subject with aphasia uses mainly fluency and sociolinguistic strategies.