Consecutive Utterances Research Articles

The effects of discourse coherence on the persistence of sentence structures.

We studied the role of discourse coherence relations on structure formulation in sentence production by examining whether a connective, an essential signal of coherence relations, modulates the tendency for speakers to reuse sentence structures (i.e., structural priming). We further examined three possible modulating factors: the type of connectives (additive vs. adversative connective), event similarity (similar event vs. different event), and topic cohesion (with or without available anaphoric antecedent). In four structural priming experiments, native Dutch participants were asked to read either a Dutch double object sentence or a prepositional object sentence and describe pictures that depicted ditransitive events. Critically, the prime and the target either were linked by a connective (en "and" or maar "but") or were not linked. The verb overlap between the prime and the target was also manipulated. In Experiment 1, the presence of en facilitated structural priming, but only when the verbs were different. In Experiment 2, maar reduced structural priming when the verbs were repeated. Experiment 3 replicated the findings of Experiments 1 and 2 in a within-subjects design. In Experiment 4, there was no referential link between the sentences. Now there was no connective effect on structural priming. Taken together, we demonstrated that the insertion of a connective influences syntactic persistence. The connective effects vary across semantic properties of the connectives, event similarity, and referential continuity, suggesting that the production of sentence structure is modulated by speakers' prediction about listeners' inference of coherence relations between consecutive utterances. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Mechanism of Theme-Formation in a Text as it was Represented by Boris Tomashevsky and Georgij Shengeli (Based on the Works of 1925)

In the mid-1930s, linguists began to spot problems, which cannot be solved without analysis of textual and discursive phenomena. Up to that moment some of such problems have been already pointed out and considered in the adjacent fields of Humanitarian theory and practice. The article deals with one of such anticipations made simultaneously by prominent Russian theorists Boris Tomashevsky and Georgij Shengeli in their studies of literary composition. They discovered that theme-formation in a text is based on the unity of meanings in neighboring sentences (Tomashevsky) or on semantic interaction of consecutive utterances (Shengeli).

A comparison of clinician assessment of speech versus automated speech analysis in mild cognitive impairment and Alzheimer’s dementia

AbstractBackgroundLanguage impairment is an important marker of neurocognitive disorders. Despite this, there is no universal system of terminology used to describe speech impairment and large inter‐rater variability can exist between clinicians assessing speech. The role of automated speech analysis is emerging as a novel and potentially more objective method to assess speech in individuals with mild cognitive impairment (MCI) and Alzheimer’s dementia (AD). No studies have analyzed whether variables extracted through automated speech analysis can also be correlated to speech characteristics identified by a clinician. We sought to: (1) investigate whether clinician‐identified speech characteristics could be correlated with acoustic and linguistic variables identified through automated speech analysis, and (2) through automated speech analysis, identify novel acoustic and linguistic variables that may be associated with MCI or AD.MethodUsing the DementiaBank speech corpus (Cookie Theft picture description task), audio recordings from patients with possible/probable AD (n=10), MCI (n=10), and controls (n=10) were rated by clinicians. Four characteristics of speech were rated by clinicians, including: word‐finding difficulty, coherence, perseveration, and speech errors. Clinicians were blinded to each other’s scoring. Where scores differed, a consensus rating was established. The speech recordings were then transcribed, and linguistic and acoustic variables were extracted through automated speech analysis. The correlation between clinician‐identified speech characteristics and the acoustic and linguistic variables were then compared.ResultA significant correlation (p < 0.05) was found between clinician‐identified speech characteristics and variables extracted through automated analysis. Average word length was correlated with word‐finding difficulty (ρ = 0.74); changes in syntactic construction (use of past tense, third person singular verbs, and subordinate clauses) were correlated with coherence (ρ = 0.51) and errors in speech (ρ = 0.58); similarity between consecutive utterances was correlated to perseveration (ρ = 0.68).ConclusionIn this exploratory study, variables extracted through automated acoustic and linguistic analysis of MCI and AD speech were strongly correlated to clinician‐identified speech characteristics. Our results suggest further investigation for using an automated, data‐driven approach to define and monitor subjective clinical speech descriptors in neurocognitive disorders.

Maternal discourse continuity and infants' actions organize 12-month-olds' language exposure during object play.

Infant language learning depends on the distribution of co-occurrences within language-between words and other words-and between language content and events in the world. Yet infant-directed speech is not limited to words that refer to perceivable objects and actions. Rather, caregivers' utterances contain a range of syntactic forms and expressions with diverse attentional, regulatory, social, and referential functions. We conducted a distributional analysis of linguistic content types at the utterance level, and demonstrated that a wide range of content types in maternal speech can be distinguished by their distribution in sequences of utterances and by their patterns of co-occurrence with infants' actions. We observed free-play sessions of 38 12-month-old infants and their mothers, annotated maternal utterances for 10 content types, and coded infants' gaze target and object handling. Results show that all content types tended to repeat in consecutive utterances, whereas preferred transitions between different content types reflected sequences from attention-capturing to directing and then descriptive utterances. Specific content types were associated with infants' engagement with objects (declaratives, descriptions, object names), with disengagement from objects (talk about attention, infant's name), and with infants' gaze at the mother (affirmations). We discuss how structured discourse might facilitate language acquisition by making speech input more predictable and/or by providing clues about high-level form-function mappings.

Dynamic control over the allocation of listening effort in speech perception

Unlike typical laboratory and clinical testing situations, everyday speech communication involves hearing multiple consecutive utterances. Previous work has shown that effort allocated to perceiving degraded speech remains elevated until well after sentence offset, presumably to allow cognitive restoration of words that were missed, via contextual cues. This restorative process potentially conflicts with the perception of subsequent sentences. In this study we introduce various follow-up signals after a target sentence with the intent of interacting with the restorative processes used to recover intelligibility in difficult listening conditions. Pupillometry was used to gauge listening effort for a sentence followed by a critical 2-second window filled by silence, noise, or a random sequence of three digits; sentences were always repeated, and digits were repeated or ignored, in blocked sessions. Speech followed by noise elicited a pattern of effort that paused until the noise was over, and then resumed. Digits elicited a modest elevation in effort when they were ignored, but a considerable elevation when repeated along with the sentence. Notably, effort allocated to the sentence appears to be suppressed when the listener plans to attend to upcoming signals. Results suggest that responses to single utterances might not reflect dynamic cognitive demands in communication.

Parsing entire discourses as very long strings: Capturing topic continuity in grounded language learning

Grounded language learning, the task of mapping from natural language to a representation of meaning, has attracted more and more interest in recent years. In most work on this topic, however, utterances in a conversation are treated independently and discourse structure information is largely ignored. In the context of language acquisition, this independence assumption discards cues that are important to the learner, e.g., the fact that consecutive utterances are likely to share the same referent (Frank et al., 2013). The current paper describes an approach to the problem of simultaneously modeling grounded language at the sentence and discourse levels. We combine ideas from parsing and grammar induction to produce a parser that can handle long input strings with thousands of tokens, creating parse trees that represent full discourses. By casting grounded language learning as a grammatical inference task, we use our parser to extend the work of Johnson et al. (2012), investigating the importance of discourse continuity in children’s language acquisition and its interaction with social cues. Our model boosts performance in a language acquisition task and yields good discourse segmentations compared with human annotators.

Right-dislocation in Catalan: Its discourse function and counterparts in English

This paper presents a corpus study of right dislocation (RD) in Catalan and discusses crosslinguistic differences of information packaging between English and Catalan. The Catalan corpus consists of 93 RDs which have been coded according to three parameters: (1) the point where the entity in the right-dislocated constituent had appeared in the discourse, (2) consequences of eliminating the right-dislocated constituent and (3) consequences of restoring the canonical order. I argue that RD in Catalan is a means to structure information in a coherent way by displacing old information from the main clause. Three main types of RDs can be found: (1) RDs which activate an entity which was no longer accessible in the discourse and make it highly salient, while still marking its discourse-old status; (2) RDs which make explicit an implicit, never textually mentioned, referent and places it in a discourse-old information position. (3) RDs referring to entities mentioned in the previous sentence. Such RDs convey an additional meaning, some ‘emotional content’, having to do with the expression of opposition or emphasis. In order to analyse crosslinguistic differences, an English text and its Catalan translation have been used. The Catalan translation contained 42 instances of RD, while the English text contained none, which shows that the two languages use different strategies to encode information packaging. The Catalan translation uses RDs mostly in cases in which the English original repeats the same phrase in two consecutive utterances and in utterances which convey contrast or opposition.

Verification of Speech Recognition Results Incorporating In-domain Confidence and Discourse Coherence Measures

Conventional confidence measures for assessing the reliability of ASR (automatic speech recognition) output are typically derived from low-level information which is obtained during speech recognition decoding. In contrast to these approaches, we propose a novel utterance verification framework which incorporates high-level knowledge sources. Specifically, we investigate two application-independent measures: in-domain confidence, the degree of match between the input utterance and the application domain of the back-end system, and discourse coherence, the consistency between consecutive utterances in a dialogue session. A joint confidence score is generated by combining these two measures with an orthodox measure based on GPP (generalized posterior probability). The proposed framework was evaluated on an utterance verification task for spontaneous dialogue performed via a (English/Japanese) speech-to-speech translation system. Incorporating the two proposed measures significantly improved utterance verification accuracy compared to using GPP alone, realizing reductions in CER (confidence error-rate) of 11.4% and 8.1% for the English and Japanese sides, respectively. When negligible ASR errors (that do not affect translation) were ignored, further improvement was achieved for the English side, realizing a reduction in CER of up to 14.6% compared to the GPP case.