The disfluent discourse: Effects of filled pauses on recall

Attentional Orienting and Disfluency-Related Memory Boost Are Intact in Adults With Moderate-Severe Traumatic Brain Injury

PURPOSE

Traumatic brain injury (TBI) is associated with a range of cognitive-communicative deficits that interfere with everyday communication and social interaction. Considerable effort has been directed at characterizing the nature and scope of cognitive-communication disorders in TBI, yet the underlying mechanisms of impairment are largely unspecified. The present research examines sensitivity to a common communicative cue, disfluency, and its impact on memory for spoken language in TBI.

METHOD

Fifty-three participants with moderate-severe TBI and 53 noninjured comparison participants listened to a series of sentences, some of which contained disfluencies. A subsequent memory test probed memory for critical words in the sentences.

RESULTS

Participants with TBI successfully remembered the spoken words (b = 1.57, p < .0001) at a similar level to noninjured comparison participants. Critically, participants with TBI also exhibited better recognition memory for words preceded by disfluency compared to words from fluent sentences (b = 0.57, p = .02).

CONCLUSIONS

These findings advance mechanistic accounts of cognitive-communication disorder by revealing that, when isolated for experimental study, individuals with moderate-severe TBI are sensitive to attentional orienting cues in speech and exhibit enhanced recognition of individual words preceded by disfluency. These results suggest that some aspects of cognitive-communication disorders may not emerge from an inability to perceive and use individual communication cues, but rather from disruptions in managing (i.e., attending, weighting, integrating) multiple cognitive, communicative, and social cues in complex and dynamic interactions. This hypothesis warrants further investigation.

Disfluency across the lifespan: an individual differences investigation

This study had two research objectives. The first was to examine age-related differences in the fluency of speech outputs, as prior research contains conflicting findings concerning whether older adults produce more disfluency than younger adults. The second was to examine cognitive individual differences, and their relationship with the production of disfluency. One hundred and fifty-four adults completed a story re-telling task, and a battery of cognitive measures. Results showed that younger adults produced more um's and fewer repetitions. For individual differences, results showed that inhibition and set shifting were related to the production of repetitions, and inhibition and working memory were related to uh production. Our results provide clarification about mixed findings with respect age and disfluency production. The individual differences provide clarification on theoretical arguments for disfluent speech in aging (e.g. Inhibition Deficit Hypothesis), and also sheds light on the role of executive functions in models of language production.

Linguistic disfluencies in Russian-speaking typically and atypically developing children: individual variability in different contexts

Disfluency in children and adults seems to occur like errors of speech but, at the same time, is an essential feature of spontaneous (unprepared) speech. The present study aimed to evaluate linguistic disfluencies in typically and atypically developing Russian-speaking children from the perspective of the dynamic adaptive model of self-monitoring in speech production. The study collected four language samples from 10 six-year-old children with developmental language disorder and 14 typically developing peers: two storytelling tasks, structured conversation, and a play argument. After transcribing audio-recordings and marking linguistic disfluencies, the authors conducted structured distributional analysis. The distribution of several indexes of disfluency was estimated to assess the prevalence and profiles of different (sub)types of disfluencies. The disfluency rate statistics were similar between the typically developing children and children with developmental language disorder. The distributional indexes score showed that tasks significantly impacted the rate of different (sub)types of disfluencies. Task-related patterns in a set of the distributional indexes significantly distinguished the groups. Thus, changes in the disfluency profile related to different external factors, as a sign of a flexibility of an adaptive self-monitoring system, may be limited in children with developmental language disorder.

Delayed onset facilitates subsequent retrieval of words during language comprehension

Prior research has shown that during language comprehension, memory representations associated with premodified words (e.g., the injured and dangerous bear) are retrieved faster from memory than those associated with unmodified words (e.g., the bear). Current explanations attribute this effect to the semantic richness of modified words. However, it is not clear whether the presence of modifying words are in fact necessary for a retrieval benefit. Premodifiers necessarily delay the onset of the target word (i.e., bear), and temporal delays may heighten attention to upcoming stimuli, and/or strengthen encoding by producing free time during encoding, facilitating subsequent retrieval. We therefore examined whether a simple delay in the onset of the target can produce a retrieval benefit. Our results show that delayed onset facilitates the subsequent retrieval of target words in the absence of any modifying information. These results lend support to models of language comprehension according to which delays may enhance attention to upcoming words, and also to models of working memory based on which free time replenishes encoding resources, strengthening the memory trace of encoded information and facilitating its retrieval at a subsequent point. Our results also contribute to current memory-based theories of sentence comprehension by showing that retrieval from memory may be affected by nonlinguistic factors such as delay-induced attention enhancement, or free time during encoding.

Do you say uh or uhm? A cross-linguistic approach to filler particle use in heritage and majority speakers across three languages

Filler particles like uhm in English or ähm in German show subtle language-specific differences and their variation in form is related to socio-linguistic variables like gender. The use of fillers in a second language has been shown to differ from monolinguals' filler particle use in both frequency and form in different language contexts. This study investigates the language-specific use of filler particles by bilingual heritage speakers in both their languages, looking at the dominant majority language in the society and their minority heritage language spoken at home. This is done based on heritage Russian and German data and majority German and English data from the RUEG corpus. Language-specific fillers were extracted from the corpus and analyzed for their occurrence and segmental form. The frequency analysis suggests an influence of bilingualism, age group, and formality of the situation on the filler frequency across all languages. The number of filler particles is higher in formal, older, and bilingual speech. The form analysis reveals an effect of language and gender on the type of filler particle. The vocalic-nasal filler particles (e.g., uhm) are more frequently found in German and English and in female speech of these languages. Heritage speakers of Russian in contact with German and English show higher use of vocalic-nasal forms also in their Russian while producing similar gender related patterns to monolingual speakers in both their languages. The higher frequency of filler particles in formal situations, older speakers and in bilingual speech, is discussed related to cognitive load which is assumed to be higher in these contexts while speech style which differs between situations and social groups is also considered as explanation. The higher use of vocalic-nasal filler particles in German and English suggests language specific filler particle preferences also related to the socio-linguistic variable gender in these languages. The results from heritage speakers suggest and influence on filler particle form in their heritage language, while also revealing socio-linguistic usage patterns related to gender which are produced by heritage speakers similarly to monolinguals in their respective language.

Exploring the Effects of Online Physician Voice Pitch Range and Filled Pauses on Patient Satisfaction in Mobile Health Communication

The convenience of mobile devices has driven the widespread use of voice technology in mobile health communication, significantly improving the timeliness of online service. However, the issue of listening to therapeutic content, which requires great cognitive effort and may exceed the patient's information processing capacity (i.e., information overload), is of concern. Based on information processing theory, this study reports how online physicians' voice characteristics (pitch range and filled pauses) affect patient satisfaction. We obtained 10,585 mobile voice consultation records of 1,416 doctors from China's largest mHealth platform and analyzed them using audio mining and empirical methods. Results showed that pitch range (β = 0.0539, p < .01) and filled pauses (β = 0.0365, p < .01) in doctors' voice positively influenced online patient satisfaction. However, the effect of filled pauses becomes weaker for patients with higher health literacy and higher disease risk. This suggests that there is heterogeneity in the way different patients process audio information. This study provides important insights for guiding online physician behaviors, enhancing patient satisfaction, and improving mobile health platform management.

Linguistic features of spontaneous speech predict conversational recall

Empirical studies of conversational recall show that the amount of conversation that can be recalled after a delay is limited and biased in favor of one's own contributions. What aspects of a conversational interaction shape what will and will not be recalled? This study aims to predict the contents of conversation that will be recalled based on linguistic features of what was said. Across 59 conversational dyads, we observed that two linguistic features that are hallmarks of interactive language use-disfluency (um/uh) and backchannelling (ok, yeah)-promoted recall. Two other features-disagreements between the interlocutors and use of "like"-were not predictive of recall. While self-generated material was better remembered overall, both hearing and producing disfluency and backchannels improved memory for the associated utterances. Finally, the disfluency-related memory boost was similar regardless of the number of disfluencies in the utterance. Overall, we conclude that interactional linguistic features of conversation are predictive of what is and is not recalled following conversation.

Self-Repair Increases Referential Coordination

When interlocutors repeatedly describe referents to each other, they rapidly converge on referring expressions which become increasingly systematized and abstract as the interaction progresses. Previous experimental research suggests that interactive repair mechanisms in dialogue underpin convergence. However, this research has so far only focused on the role of other-initiated repair and has not examined whether self-initiated repair might also play a role. To investigate this question, we report the results from a computer-mediated maze task experiment. In this task, participants communicate with each other via an experimental chat tool, which selectively transforms participants' private turn-revisions into public self-repairs that are made visible to the other participant. For example, if a participant, A, types "On the top square," and then before sending, A revises the turn to "On the top row," the server automatically detects the revision and transforms the private turn-revisions into a public self-repair, for example, "On the top square umm I meant row." Participants who received these transformed turns used more abstract and systematized referring expressions, but performed worse at the task. We argue that this is due to the artificial self-repairs causing participants to put more effort into diagnosing and resolving the referential coordination problems they face in the task, yielding better grounded spatial semantics and consequently increased use of abstract referring expressions.

Listener- Versus Speaker-Oriented Disfluencies in Autistic Adults: Insights From Wearable Eye-Tracking and Skin Conductance Within a Live Face-to-Face Paradigm

PURPOSE

Our study addresses three main questions: (a) Do autistics and neurotypicals produce different patterns of disfluencies, depending on the experimenter's direct versus averted gaze? (b) Are these patterns correlated to gender, skin conductance responses, fixations on the experimenter's face, alexithymia, or social anxiety scores? Lastly, (c) can eye-tracking and electrodermal activity data be used in distinguishing listener- versus speaker-oriented disfluencies?

METHOD

Within a live face-to-face paradigm combining a wearable eye-tracker with electrodermal activity sensors, 80 adults (40 autistics, 40 neurotypicals) defined words in front of an experimenter who was either staring at their eyes (direct gaze condition) or looking elsewhere (averted gaze condition).

RESULTS

Autistics produce less listener-oriented (uh, um) and more speaker-oriented (prolongations, breath) disfluencies than neurotypicals. In both groups, men produce less um than women. Both autistics' and neurotypicals' speech are influenced by whether their interlocutor systematically looks at them in the eyes or not, but their reactions go in opposite directions. Disfluencies seem to primarily be linguistic phenomena as experienced stress, social attention, alexithymia, and social anxiety scores do not influence any of the reported results. Finally, eye-tracking and electrodermal activity data suggest that laughter could be a listener-oriented disfluency.

CONCLUSIONS

This article studies disfluencies in a fine-grained way in autistic and neurotypical adults while controlling for social attention, experienced stress, and experimental condition (direct vs. averted gaze). It adds to current literature by (a) enlightening our knowledge of speech in autism, (b) opening new perspectives on disfluency patterns as important signals in social interaction, (c) addressing theoretical issues on the dichotomy between listener- and speaker-oriented disfluencies, and (d) considering understudied phenomena as potential disfluencies (e.g., laughter, breath).

SUPPLEMENTAL MATERIAL

https://doi.org/10.23641/asha.23549550.

EEG Correlates of Distractions and Hesitations in Human–Robot Interaction: A LabLinking Pilot Study

In this paper, we investigate the effect of distractions and hesitations as a scaffolding strategy. Recent research points to the potential beneficial effects of a speaker’s hesitations on the listeners’ comprehension of utterances, although results from studies on this issue indicate that humans do not make strategic use of them. The role of hesitations and their communicative function in human-human interaction is a much-discussed topic in current research. To better understand the underlying cognitive processes, we developed a human–robot interaction (HRI) setup that allows the measurement of the electroencephalogram (EEG) signals of a human participant while interacting with a robot. We thereby address the research question of whether we find effects on single-trial EEG based on the distraction and the corresponding robot’s hesitation scaffolding strategy. To carry out the experiments, we leverage our LabLinking method, which enables interdisciplinary joint research between remote labs. This study could not have been conducted without LabLinking, as the two involved labs needed to combine their individual expertise and equipment to achieve the goal together. The results of our study indicate that the EEG correlates in the distracted condition are different from the baseline condition without distractions. Furthermore, we could differentiate the EEG correlates of distraction with and without a hesitation scaffolding strategy. This proof-of-concept study shows that LabLinking makes it possible to conduct collaborative HRI studies in remote laboratories and lays the first foundation for more in-depth research into robotic scaffolding strategies.

Can filled pauses be represented as linguistic items? Investigating the effect of exposure on the perception and production of um

The current paper presents three studies that investigated the effect of exposure on the mental representations of filled pauses (um/uh). In Study 1, a corpus analysis identified the frequency of co-occurrence of filled pauses with words located immediately before or after them in naturalistic spoken adult British English (BNC2014). Based on the collocations identified in Study 1, in Study 2, 22 native British English-speaking adults heard sentences in which the location of filled pauses and the co-occurring words were manipulated and the participants were asked to judge the acceptability of the sentences heard. Study 3 was a sentence recall experiment in which we asked 29 native British English adults to repeat a similar set of sentences as used in Study 2. We found that frequency-based distributional patterns of filled pauses (Study 1) affected the sentence judgments (Study 2) and repetition accuracy (Study 3), in particular when the filled pause followed its collocate. Thus, the current study provides converging evidence for the account maintaining that filled pauses are linguistic items. In addition, we suggest filled pauses in certain locations could be considered as grammatical items, such as suffixes.

Aging, Gesture Production, and Disfluency in Speech: A Comparison of Younger and Older Adults

Age-related changes are observed in the speech and gestures of neurotypical individuals. Older adults are more disfluent in speech and use fewer representational gestures (e.g., holding two hands close to each other to mean small), compared to younger adults. Using gestures, especially representational gestures, is common in difficult tasks to aid the conceptualization process and to facilitate lexical access. This study investigates how aging can affect gesture production and the co-occurrence between gesture and speech disfluency. We elicited speech and gesture samples from younger and older adults (N = 60) by using a painting description task that provided concrete and abstract contexts. Results indicated that albeit the two age groups revealed comparable overall speech disfluency and gesture rates, they differed in terms of how their disfluencies and gestures were distributed across specific categories. Moreover, the proportion of speech disfluencies that occur with a gesture was significantly higher for younger than older adults. However, the two age groups were comparable in terms of the proportion of gestures that were accompanied by a speech disfluency. These findings suggest that younger adults' language production system might be better at benefiting from other modalities, that is, gesture, to resolve temporary problems in speech planning. However, from a gesture perspective, it might be difficult to differentiate between gestures' self-oriented and communicative functions and understand their role in speech facilitation. Focusing on specific cases where speech disfluency and gestures co-occur and considering individual differences might bring insight into multimodal communication.

Bilingual Children's Narrative Comprehension: Do Pauses during Retelling Buy Time for Understanding?

This study extended research on bilingual children's narrative comprehension to text processing during retelling. A cross- and monolinguistic design allowed for the investigation of the association of various aspects of narrative comprehension with two types of pause occurring during oral retelling of a story heard in one language and retold in a second or in the same language. A sample of 95 bilingual (L1 German/L2 English) fourth-graders participated in the experiment. Comprehension was predominantly accomplished during listening, as children with L1 input outperformed children with L2 input, but they did not benefit from L1 in retelling. Children's comprehension performance and corresponding pause patterns suggested that the duration of filled pauses dedicated to gist formation indicated efficiency of comprehension: The duration decreased with increasing comprehension. The findings allowed the conclusion that longer pauses during retelling did not buy time for comprehension.

How the Probabilistic Structure of Grammatical Context Shapes Speech

Does systematic covariation in the usage patterns of forms shape the sublexical variance observed in conversational speech? We address this question in terms of a recently proposed discriminative theory of human communication that argues that the distribution of events in communicative contexts should maintain mutual predictability between language users, present evidence that the distributions of words in the empirical contexts in which they are learned and used are geometric, and thus support this. Here, we extend this analysis to a corpus of conversational English, showing that the distribution of grammatical regularities and the sub-distributions of tokens discriminated by them are also geometric. Further analyses reveal a range of structural differences in the distribution of types in parts of speech categories that further support the suggestion that linguistic distributions (and codes) are subcategorized by context at multiple levels of abstraction. Finally, a series of analyses of the variation in spoken language reveals that quantifiable differences in the structure of lexical subcategories appears in turn to systematically shape sublexical variation in speech signal.

Processing of Self-Repairs in Stuttered and Non-Stuttered Speech

Previous research suggests that listeners can use the presence of speech disfluencies to predict upcoming linguistic input. But how is the processing of typical disfluencies affected when the speaker also produces atypical disfluencies, as in the case of stuttering? We addressed this question in a visual-world eye-tracking experiment in which participants heard self-repair disfluencies while viewing displays that contained a predictable target entity. Half the participants heard the sentences spoken by a speaker who stuttered, and half heard the sentences spoken by the same speaker who produced the sentences without stuttering. Results replicated previous work in demonstrating that listeners engage in robust predictive processing when hearing self-repair disfluencies. Crucially, the magnitude of the prediction effect was reduced when the speaker stuttered compared to when the speaker did not stutter. Overall, the results suggest that listeners' ability to model the production system of a speaker is disrupted when the speaker stutters.

I see what you meant to say: Anticipating speech errors during online sentence processing

Everyday speech is rife with errors and disfluencies, yet processing what we hear usually feels effortless. How does the language comprehension system accomplish such an impressive feat? The current experiment tests the hypothesis that listeners draw on relevant contextual and linguistic cues to anticipate speech errors and mentally correct them, even before receiving an explicit correction from the speaker. In the current visual-world eye-tracking experiment, we monitored participants' eye movements to objects in a display while they listened to utterances containing reparandum-repair speech errors (e.g., . . . his cat, uh I mean his dog . . .). The contextual plausibility of the misspoken word and the certainty with which the speaker uttered this word were systematically manipulated. Results showed that listeners immediately exploited these cues to generate top-down expectations regarding the speaker's communicative intention. Crucially, listeners used these expectations to constrain the bottom-up speech input and mentally correct perceived speech errors, even before the speaker initiated the correction. The results provide powerful evidence regarding the joint process of correcting speech errors that involves both the speaker and the listener. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

RETRACTED: Auditory (dis-)fluency triggers sequential processing adjustments

An increasing amount of studies indicates that experiencing increased task demands, triggered for example by conflicting stimulus features or low perceptual fluency, lead to processing adjustments. While these demand-triggered processing adjustments have been shown for different paradigms (e.g., response conflict tasks, perceptual disfluency, task switching, dual tasking), most of them are restricted to the visual modality. The present study investigated as to whether the challenge to understand speech signals in normal-hearing subjects would also lead to sequential processing adjustments if the processing fluency of the respective auditory signals changes from trial to trial. To that end, we used spoken number words (one to nine) that were either presented with high (clean speech) or low perceptual fluency (i.e., vocoded speech as used in cochlear implants-Experiment 1; speech embedded in multi-speaker babble noise as typically found in bars-Experiment 2). Participants had to judge the spoken number words as smaller or larger than five. Results show that the fluency effect (performance difference between high and low perceptual fluency) in both experiments was smaller following disfluent words. Thus, if it's hard to understand, you try harder.

Disfluencies signal reference to novel objects for adults but not children

Speech disfluencies can guide the ways in which listeners interpret spoken language. Here, we examined whether three-year-olds, five-year-olds, and adults use filled pauses to anticipate that a speaker is likely to refer to a novel object. Across three experiments, participants were presented with pairs of novel and familiar objects and heard a speaker refer to one of the objects using a fluent ("Look at the ball/lep!") or disfluent ("Look at thee uh ball/lep!") expression. The salience of the speaker's unfamiliarity with the novel referents, and the way in which the speaker referred to the novel referents (i.e., a noun vs. a description) varied across experiments. Three- and five-year-olds successfully identified familiar and novel targets, but only adults' looking patterns reflected increased looks to novel objects in the presence of a disfluency. Together, these findings demonstrate that adults, but not young children, use filled pauses to anticipate reference to novel objects.

Self Addressed Questions and Filled Pauses: A Cross-linguistic Investigation

There is an ongoing debate whether phenomena of disfluency (such as filled pauses) are produced communicatively. Clark and Fox Tree (Cognition 84(1):73-111, 2002) propose that filled pauses are words, and that different forms signal different lengths of delay. This paper evaluates this Filler-As-Words hypothesis by analyzing the distribution of self-addressed-questions or SAQs (such as "what's the word") in relation to filled pauses. We found that SAQs address different problems in different languages (most frequently about memory-retrieval in English and Chinese, and about appropriateness in Japanese). In relation to filled pauses, British but not American English uses "um" to signal a more severe problem than "uh". Chinese uses different filled pauses to signal the syntactic category of the problem constituent. Japanese uses different filled pauses to signal levels of interaction with the interlocuter. Overall, our data supports the Filler-As-Words hypothesis that filled pauses are used communicatively. However, the dimensions of its meanings vary across languages and dialects.

A Comparison of Coverbal Gesture Use in Oral Discourse Among Speakers With Fluent and Nonfluent Aphasia

PURPOSE

Coverbal gesture use, which is affected by the presence and degree of aphasia, can be culturally specific. The purpose of this study was to compare gesture use among Cantonese-speaking individuals: 23 neurologically healthy speakers, 23 speakers with fluent aphasia, and 21 speakers with nonfluent aphasia.

METHOD

Multimedia data of discourse samples from these speakers were extracted from the Cantonese AphasiaBank. Gestures were independently annotated on their forms and functions to determine how gesturing rate and distribution of gestures differed across speaker groups. A multiple regression was conducted to determine the most predictive variable(s) for gesture-to-word ratio.

RESULTS

Although speakers with nonfluent aphasia gestured most frequently, the rate of gesture use in counterparts with fluent aphasia did not differ significantly from controls. Different patterns of gesture functions in the 3 speaker groups revealed that gesture plays a minor role in lexical retrieval whereas its role in enhancing communication dominates among the speakers with aphasia. The percentages of complete sentences and dysfluency strongly predicted the gesturing rate in aphasia.

CONCLUSIONS

The current results supported the sketch model of language-gesture association. The relationship between gesture production and linguistic abilities and clinical implications for gesture-based language intervention for speakers with aphasia are also discussed.

Conflict and metacognitive control: the mismatch-monitoring hypothesis of how others' knowledge states affect recall

Information about others' success in remembering is frequently available. For example, students taking an exam may assess its difficulty by monitoring when others turn in their exams. In two experiments, we investigated how rememberers use this information to guide recall. Participants studied paired associates, some semantically related (and thus easier to retrieve) and some unrelated (and thus harder). During a subsequent cued recall test, participants viewed fictive information about an opponent's accuracy on each item. In Experiment 1, participants responded to each cue once before seeing the opponent's performance and once afterwards. Participants reconsidered their responses least often when the opponent's accuracy matched the item difficulty (easy items the opponent recalled, hard items the opponent forgot) and most often when the opponent's accuracy and the item difficulty mismatched. When participants responded only after seeing the opponent's performance (Experiment 2), the same mismatch conditions that led to reconsideration even produced superior recall. These results suggest that rememberers monitor whether others' knowledge states accord or conflict with their own experience, and that this information shifts how they interrogate their memory and what they recall.

Thee, uhh disfluency effect in preschoolers: A cue to discourse status

Speech disfluencies, such as filled pauses (ummm, uhhh), are increasingly recognized as an informative element of the speech stream. Here, we examined whether 2- and 3-year-olds expected that the presence of filled pause would signal reference to objects that are new to a discourse. Children viewed pairs of familiar objects on a screen and heard a speaker refer to one of the objects twice in succession. Next, children heard a critical utterance and were asked to look and point at either the discourse-given (i.e., previously mentioned) or discourse-new (i.e., previously unmentioned) object using a fluent ('Look at the ball!') or disfluent ('Look at thee uh ball!') expression. The results indicated that 3-year-old children, but not 2-year-old children, initially expected the speaker to continue to refer to given information in the critical utterance. Upon hearing a filled pause, however, both 2- and 3-year-old children's looking patterns reflected increased looks to discourse-new objects, although the timing of the effect differed between the age groups. Together, these findings demonstrate that young children have an emerging understanding of the role of filled pauses in speech.

The effect of filled pauses on the processing of the surface form and the establishment of causal connections during the comprehension of spoken expository discourse

The purpose of this study was to examine the effect of filled pauses (uh) on the verification of words and the establishment of causal connections during the comprehension of spoken expository discourse. With this aim, we asked Spanish-speaking students to listen to excerpts of interviews with writers, and to perform a word-verification task and a question-answering task on causal connectivity. There were two versions of the excerpts: filled pause present and filled pause absent. Results indicated that filled pauses increased verification times for words that preceded them, but did not make a difference on response times to questions on causal connectivity. The results suggest that, as signals of delay, filled pauses create a break with surface information, but they do not have the same effect on the establishment of meaningful connections.

Filled pause refinement based on the pronunciation probability for lecture speech

Nowadays, although automatic speech recognition has become quite proficient in recognizing or transcribing well-prepared fluent speech, the transcription of speech that contains many disfluencies remains problematic, such as spontaneous conversational and lecture speech. Filled pauses (FPs) are the most frequently occurring disfluencies in this type of speech. Most recent studies have shown that FPs are widely believed to increase the error rates for state-of-the-art speech transcription, primarily because most FPs are not well annotated or provided in training data transcriptions and because of the similarities in acoustic characteristics between FPs and some common non-content words. To enhance the speech transcription system, we propose a new automatic refinement approach to detect FPs in British English lecture speech transcription. This approach combines the pronunciation probabilities for each word in the dictionary and acoustic language model scores for FP refinement through a modified speech recognition forced-alignment framework. We evaluate the proposed approach on the Reith Lectures speech transcription task, in which only imperfect training transcriptions are available. Successful results are achieved for both the development and evaluation datasets. Acoustic models trained on different styles of speech genres have been investigated with respect to FP refinement. To further validate the effectiveness of the proposed approach, speech transcription performance has also been examined using systems built on training data transcriptions with and without FP refinement.

Alice's adventures in um-derland: Psycholinguistic sources of variation in disfluency production

This study tests the hypothesis that three common types of disfluency (fillers, silent pauses, and repeated words) reflect variance in what strategies are available to the production system for responding to difficulty in language production. Participants' speech in a storytelling paradigm was coded for the three disfluency types. Repeats occurred most often when difficult material was already being produced and could be repeated, but fillers and silent pauses occurred most when difficult material was still being planned. Fillers were associated only with conceptual difficulties, consistent with the proposal that they reflect a communicative signal whereas silent pauses and repeats were also related to lexical and phonological difficulties. These differences are discussed in terms of different strategies available to the language production system.