Corpus Of Conversations Research Articles

Hierarchical prosody modeling for Mandarin spontaneous speech

In this paper, a hierarchical prosody model (HPM)-based method for Mandarin spontaneous speech is proposed. First, an HPM is designed for describing relations among acoustic features of utterances, linguistic features of texts, and prosodic tags representing the underlying hierarchical prosodic structures of utterances. Subsequently, a sequential optimization algorithm is employed to train the HPM based on a large conversational speech corpus, the Mandarin Conversational Dialogue Corpus (MCDC), which features orthographic transcriptions and prosodic event annotations. In this unsupervised training method, all utterances of the MCDC are labeled with two types of prosodic tags, namely, break and prosodic states, automatically and simultaneously. After training, the HPM parameters are examined to identify critical prosodic properties of Mandarin spontaneous speech, which are then compared with their counterparts in the read-speech HPM. The prosodic tags on the studied utterances enable mapping of various prosodic events onto the hierarchical prosodic structures of the utterances. Prosodic analyses of some disfluent events are conducted using the prosodic tags affixed to the MCDC. Finally, an application of the HPM to assist in Mandarin spontaneous-speech recognition is discussed. Significant relative error rate reductions of 9.0%, 9.2%, 15.6%, and 7.3% are obtained for base-syllable, character, tone, and word recognition, respectively.

Declarative questions in Polish student conversations

This paper focuses on the contextual factors influencing the status of declarative polar (information seeking) questions in Polish. The study is based on the two tailor-made corpora of Polish student conversations, specially collected in Kraków and Krosno from 2015 to 2017, out of which 28 examples of declaratives are analysed (out of 200 identified in the corpora). Polish question research tradition, contrary to English, does not recognize the notion of a declarative question, but it does recognize the issue in other terms. Declarative questions in Polish conversation corpora are here defined as statements requesting or demanding some information from the hearer (B-events in Labov's terminology), who actually responds or reacts to them, usually cued by rising intonation (with notable exceptions of e.g. ironic declaratives), and some emotive “bias” (rhetoric, irony, humour etc.); they are analysed in the article in accordance to the cognitive linguistic methodology, involving language-thought-communication continuum, advanced by Gerard Steen, broadly applied throughout this special issue. Major conventional markers (such as a/i and ale) occurring with declaratives are identified and analysed, followed by more extended, contextual usages (referred to here, as in the whole special issue, as ‘deliberate’, following Steen's approach to metaphors). Among the markers attention is also paid to various combinations of forms, including particle sequences and question clusters, all characteristic of Polish conversational declaratives. The analysis draws attention to the functions of declaratives, including the broad repair function established in other languages, as well rhetorical usages of declaratives (esp. repetitive, emphatic or ironic), which tend to be the non-canonical A, AB or D-events, typical of Polish communication style.

Acoustic analysis of intonation: Comparison between two dialects of Spanish from the north of the peninsula

This paper introduces the melodic characterisation of spontaneous Spanish presented by speakers from two regions in the north of Spain: Asturias and Navarre, based on a conversational corpus composed of 100 utterances emitted by 62 participants. It focuses on the intonation of the dialects, an aspect which is poorly investigated but highly relevant for characterising and distinguishing one from the other.The study is based on the Melodic Analysis of Speech method, a meticulous acoustic analysis through which exact values are obtained from spontaneous speech, which can then be compared. In short, the most notable coincident and divergent melodic features between the two dialects are: (1) the tendency of the two to present a first peak, more frequently tonic in Asturian Spanish and atonic in Navarrese; (2) an almost flat or slightly falling body of the contour with abundant internal inflections, which are also distinguished by the syllable where the rise ends: tonic in Asturian Spanish, and tonic and atonic in the final position of the word in Navarrese; and (3) more pronounced final inflections, with five different endings, and tonal fields, in the region of Navarre.

Slips of the tongue in the Seoul Korean Corpus of spontaneous speech

Although speech errors have served as the primary source of evidence for how the language production system works, there is a lack of studies on naturalistic slips of the tongue in Korean. The present study had two main goals: first, we introduced a new collection of slips produced by native speakers of Korean in a naturalistic setting and provided a general description of their main error patterns. The speech errors were drawn from a corpus of Korean conversations transcribed from tape recordings (Seoul Corpus; Yun et al., 2015), and classified according to the nature of the errors and the type of linguistic units with which each error type occurs. These findings were then compared to those reported from other languages such as some Germanic languages, Spanish, and Mandarin in order to make a preliminary cross-linguistic comparison of error patterns. It was found that Korean showed a wide range of universal language patterns, but also unique patterns which were attributed to the phonological and orthographic structures of Korean.

LANGUAGE ADJUSTMENT ON A CONTACT SITUATIONS BETWEEN NATIVE SPEAKERS AND ADVANCED JAPANESE LEARNERS: A COMPARISON WITH NATIVE SPEAKERS SITUATIONS (母語話者と上級日本語学習者の接続場面における言語調整：母語場面との比較)

The situations where both native and non-native speakers participate in a conversation are called contact situations. In these situations, both native and non-native speakers make verbal behaviour adjustments to achieve a smooth conversation. Adjustments by native speakers usually are made in form of Foreigner Talk (FT) while adjustments by non-native speakers made by performing communication strategies (CS). The common point between FT and CS comes from the lack of language ability or resource deficit. However, higher the non-native speakers’ second language proficiency, the adjustments made by both speakers are expected to be more complex. This paper is aiming to investigate those verbal adjustments made by native speakers in the contact situations with advanced learners / non-native speakers. To do it, this paper is employing n-gram method to gain the database of collocation and its occurrence frequency. The data for this research are Nagoya University Conversation Corpus and Nagoya University Japanese Learners Corpus. From the data, this paper found verbal behaviour adjustments which takes form in expressions such are “un” which coined at the end of own utterance, “tokaitte”, and “sounanda”.

Pragmatic Features of Tag Questions in Turkish

Tag Questions are considerably multifunctional in terms of their pragmatic meanings. This study, adopting a descriptive framework, aims to identify and categorize the pragmatic functions of tag questions in Turkish, based on naturally occuring data. It is a corpus based study and draws upon a comprehensive data, driven from spontaneous speech. METU spoken corpus and 20 hours of natural conversation are used as data. Kappa is used for data analysis in order to find out inter-rater agreement for the taxonomy of Turkish tag questions. The resulting typology for Turkish tag questions is mainly similar to those suggested by Holmes’ (1995) and Tottie and Hoffmann’s (2006) typology. However, there are also different pragmatic functions of Turkish tag questions other than those proposed in the literature. Whereas mocking and justification tags are observed in Turkish, these functions have not been proposed in languages such as English and Italian.

LA CONSTRUCCIÓN NO SÉ EN ESPAÑOL COLOQUIAL. VALORES PRAGMÁTICOS Y POSICIÓN DISCURSIVA

La expresión del ‘no saber’ en primera persona del singular del presente de indicativo posee, en español, diversas posibilidades de manifestación formal, algunas de las cuales entrañan valores alejados a los propios de su funcionamiento dentro de la estructura predicativa de la oración y próximos al desempeño de funciones relacionadas con el ámbito de la marcación discursiva. Este trabajo se centra en una de esas construcciones, la expresión no sé en los usos en que no forma parte de una estructura de predicación con complemento directo. A partir de su rastreo en un corpus de conversaciones orales espontáneas, se aborda su descripción formal y funcional con ayuda (i) del modelo de segmentación de la conversación coloquial del grupo Val.Es.Co., en la última visión de conjunto ofrecida a propósito de su sistema de unidades (Grupo Val.Es.Co., 2014), (ii) de la propuesta de figuras de sintaxis del grupo GARS (cf. Blanche-Benveniste, 1985, López Serena, 2007) y (iii) de la aproximación funcional al comportamiento de los marcadores del discurso en el continuum entre inmediatez y distancia comunicativa llevada a cabo por López Serena y Borreguero Zuloaga (2010). La triangulación de los resultados de los tres tipos de análisis efectuados permite identificar interesantes correlaciones entre ciertos valores pragmáticos y determinadas regularidades y complementar la información que sobre esta construcción ofrecían ya algunos diccionarios de partículas, marcadores u operadores del español. Desde un punto de vista teórico, el artículo pone de relieve el interés de introducir, en el análisis de construcciones discursivas que, como la estudiada aquí, van «más allá de los marcadores del discurso» (cf. López Serena, 2011a), la consideración del tipo de unidad discursiva en la que aparecen, así como de los tipos de figuras de sintaxis con que concurren.

Modeling Speech Acts in Asynchronous Conversations: A Neural-CRF Approach

Participants in an asynchronous conversation (e.g., forum, e-mail) interact with each other at different times, performing certain communicative acts, called speech acts (e.g., question, request). In this article, we propose a hybrid approach to speech act recognition in asynchronous conversations. Our approach works in two main steps: a long short-term memory recurrent neural network (LSTM-RNN) first encodes each sentence separately into a task-specific distributed representation, and this is then used in a conditional random field (CRF) model to capture the conversational dependencies between sentences. The LSTM-RNN model uses pretrained word embeddings learned from a large conversational corpus and is trained to classify sentences into speech act types. The CRF model can consider arbitrary graph structures to model conversational dependencies in an asynchronous conversation. In addition, to mitigate the problem of limited annotated data in the asynchronous domains, we adapt the LSTM-RNN model to learn from synchronous conversations (e.g., meetings), using domain adversarial training of neural networks. Empirical evaluation shows the effectiveness of our approach over existing ones: (i) LSTM-RNNs provide better task-specific representations, (ii) conversational word embeddings benefit the LSTM-RNNs more than the off-the-shelf ones, (iii) adversarial training gives better domain-invariant representations, and (iv) the global CRF model improves over local models.

Asynchronous and Event-Based Fusion Systems for Affect Recognition on Naturalistic Data in Comparison to Conventional Approaches

Throughout many present studies dealing with multi-modal fusion, decisions are synchronously forced for fixed time segments across all modalities. Varying success is reported, sometimes performance is worse than unimodal classification. Our goal is the synergistic exploitation of multimodality whilst implementing a real-time system for affect recognition in a naturalistic setting. Therefore we present a categorization of possible fusion strategies for affect recognition on continuous time frames of complete recording sessions and we evaluate multiple implementations from resulting categories. These involve conventional fusion strategies as well as novel approaches that incorporate the asynchronous nature of observed modalities. Some of the latter algorithms consider temporal alignments between modalities and observed frames by applying asynchronous neural networks that use memory blocks to model temporal dependencies. Others use an indirect approach that introduces events as an intermediate layer to accumulate evidence for the target class through all modalities. Recognition results gained on a naturalistic conversational corpus show a drop in recognition accuracy when moving from unimodal classification to synchronous multimodal fusion. However, with our proposed asynchronous and event-based fusion techniques we are able to raise the recognition system's accuracy by 7.83 percent compared to video analysis and 13.71 percent in comparison to common fusion strategies.

The headset as an interactional resource in a video relay interpreting (VRI) setting

AbstractVideo relay interpreting (VRI) enables communication between a signed language user, remotely connected to an interpreter by videophone, and an interlocutor in spoken contact with the interpreter by telephone. Both users of the service are physically separated from each other and from the interpreter, who is in a studio. Essential technical components of the system include such items as videophones, telephones, computers, software, and a headset. This article explores how the interpreter orients towards the headset, turning it into an interactional resource. Examples of how this is done are identified in extracts from a corpus of VRI conversations between users of Swedish Sign Language (SSL) and spoken Swedish. Ethical approval and all participants’ consent were obtained. Three practices were identified: pointing towards the headset, orienting towards it in other ways (positioning, gesturing, direction of gaze), and holding it. All these practices have concrete pragmatic implications for the various steps in communication, such as establishing reference, repairs, and turn allocation. Enhancing VRI interpreters’ awareness of how equipment like a headset helps to organize the interaction is important, with a view to ensuring that the available technology is used to best effect for purposes of communication.

Nasal place assimilation trades off inferrability of both target and trigger words

In English, nasal place assimilation occurs across word boundaries, such as ten bucks pronounced as te[m] bucks. Assimilation can be viewed as a reduction or loss of the assimilation target’s place cue (/n/ in ten), and simultaneously as an enhancement of the assimilation trigger’s place cue (/b/ in bucks) by spreading its place cue earlier in the signal. A message-oriented phonological approach predicts that assimilation is sensitive to the relative contextual inferrability of both the target and trigger words: More assimilation should be observed for more contextually predictable target words, while less assimilation should be observed for contextually more predictable trigger words. These predictions deviate from accounts that view assimilation solely as reduction. To test these predictions, sequences which license assimilation were extracted from a conversational speech corpus. Both categorical assimilation (based on close phonetic transcription) and gradient acoustic assimilation (based on F2) were analyzed. As predicted, assimilation was more likely both when a target like ten was high in predictability and when its trigger bucks was low in predictability. Assimilation thus serves as both reduction and enhancement, and can be used to manage redundancy in the speech signal. More broadly, this constitutes evidence for the influence of communicative pressures on phonology.

Text-Independent Speaker Verification Based on Triplet Convolutional Neural Network Embeddings

The effectiveness of introducing deep neural networks into conventional speaker recognition pipelines has been broadly shown to benefit system performance. A novel text-independent speaker verification (SV) framework based on the triplet loss and a very deep convolutional neural network architecture (i.e., Inception-Resnet-v1) are investigated in this study, where a fixed-length speaker discriminative embedding is learned from sparse speech features and utilized as a feature representation for the SV tasks. A concise description of the neural network based speaker discriminative training with triplet loss is presented. An Euclidean distance similarity metric is applied in both network training and SV testing, which ensures the SV system to follow an end-to-end fashion. By replacing the final max/average pooling layer with a spatial pyramid pooling layer in the Inception-Resnet-v1 architecture, the fixed-length input constraint is relaxed and an obvious performance gain is achieved compared with the fixed-length input speaker embedding system. For datasets with more severe training/test condition mismatches, the probabilistic linear discriminant analysis (PLDA) back end is further introduced to replace the distance based scoring for the proposed speaker embedding system. Thus, we reconstruct the SV task with a neural network based front-end speaker embedding system and a PLDA that provides channel and noise variabilities compensation in the back end. Extensive experiments are conducted to provide useful hints that lead to a better testing performance. Comparison with the state-of-the-art SV frameworks on three public datasets (i.e., a prompt speech corpus, a conversational speech Switchboard corpus, and NIST SRE10 10 s–10 s condition) justifies the effectiveness of our proposed speaker embedding system.

Adolescents’ social background and non-standard writing in online communication

AbstractIn a large corpus (2.9 million tokens) of chat conversations, we studied the impact of Flemish adolescents’ social background on non-standard writing. We found significant correlations between different aspects of social class (level of education, home language and profession of the parents) and all examined deviations from formal written standard Dutch. Clustering several social variables might not only lead to a better operationalization of the complex phenomenon of social class, it certainly allows for discriminating social groups with distinct linguistic practices: lower class teenagers used each of the non-standard features much more often and in some cases in a different way than their upper class peers. Possible explanations concern discrepancies in terms of both linguistic proficiency and linguistic attitudes. Our findings emphasize the importance of including social background as an independent variable in variationist studies on youngsters’ computer-mediated communication.

The combination of discourse markers in spontaneous conversations

AbstractThis paper attempts to establish some basis for the combination of discourse markers in conversations. The analysis presented here will be based on Spanish spontaneous conversations and will make use of the Val.Es.Co model of discourse units developed in the last ten years (Briz et al. 2003;Grupo Val.Es.Co 2014) and a corpus of spontaneous conversations. As a result, patterns of combination of DMs will be established; some types of combinations will be described, and a distinction will be drawn betweenlexicalizationof DMs,adjacencyof DMs in two neighboring discourse units, andcombinationproper. In addition, some places are favored for the combination of discourse markers: the initial positions of the discourse unitsactandintervention. Binary combinations are preferred to ternary or 4-ary combinations. Finally, DMs with textual functions are preferred in the combinations studied when compared to modal or interpersonal DMs.

Smell Is Coded in Grammar and Frequent in Discourse: Cha'palaa Olfactory Language in Cross‐Linguistic Perspective

AbstractIt has long been claimed that there is no lexical field of smell, and that smell is of too little validity to be expressed in grammar. We demonstrate both claims are false. The Cha'palaa language (Ecuador) has at least 15 abstract smell terms, each of which is formed using a type of classifier previously thought not to exist. Moreover, using conversational corpora we show that Cha'palaa speakers also talk about smell more than Imbabura Quechua and English speakers. Together, this shows how language and social interaction may jointly reflect distinct cultural orientations towards sensory experience in general and olfaction in particular.

Process models of interrelated speech intentions from online health-related conversations.

Being related to the adoption of new beliefs, attitudes and, ultimately, behaviors, analyzing online communication is of utmost importance for medicine. Multiple health care, academic communities, such as information seeking and dissemination and persuasive technologies, acknowledge this need. However, in order to obtain understanding, a relevant way to model online communication for the study of behavior is required. In this paper, we propose an automatic method to reveal process models of interrelated speech intentions from conversations. Specifically, a domain-independent taxonomy of speech intentions is adopted, an annotated corpus of Reddit conversations is released, supervised classifiers for speech intention prediction from utterances are trained and assessed using 10-fold cross validation (multi-class, one-versus-all and multi-label setups) and an approach to transform conversations into well-defined, representative logs of verbal behavior, needed by process mining techniques, is designed. The experimental results show that: (1) the automatic classification of intentions is feasible (with Kappa scores varying between 0.52 and 1); (2) predicting pairs of intentions, also known as adjacency pairs, or including more utterances from even other heterogeneous corpora can improve the predictions of some classes; and (3) the classifiers in the current state are robust to be used on other corpora, although the results are poorer and suggest that the input corpus may not sufficiently capture varied ways of expressing certain speech intentions. The extracted process models of interrelated speech intentions open new views on grasping the formation of beliefs and behavioral intentions in and from speech, but in-depth evaluation of these conversational models is further required.

Estudi comparatiu de l’atenuació pragmàtica en la llengua catalana: converses col·loquials prototípiques-converses col·loquials perifèriques

<p>En aquest treball s'ha dut a terme una anàlisi comparativa de l'atenuació pragmàtica en un corpus de converses col·loquials en català amb l'objectiu d'estudiar de forma contrastiva l’atenuació, és a dir, d’observar el contrast d’ús d’aquesta estratègia pragmàtica entre converses col·loquials prototípiques i converses col·loquials perifèriques. És per això que s’ha cregut pertinent d’explicar prèviament el gènere conversacional i el registre col·loquial, és a dir, la conversa col·loquial i les seues característiques. Per a fer aquest estudi, s'ha recopilat un corpus propi (gravat i transcrit) format per cinc converses col·loquials: dos de prototípiques i tres de perifèriques. La metodologia seguida per a l'anàlisi de l'atenuació és la proposada per Albelda et alii (2014), en què es tenen en compte variables situacionals, sociolingüístiques i enunciatives, així com una llista de possibles procediments candidats a expressar les formes d'atenuació.</p>

Transgressions prosodiques et lénition lexicale

Pour éviter les tabous linguistiques le plus facile est de recourir à l’implicite de l’énonciation. Cela explique l’importance de la place de la prosodie signifiante (objet de la phonopragmatique) parmi les ressources de la multimodalité. Car les signes vocaux permettent de transmettre discrètement ce qui, dit avec des mots, constituerait une entorse aux conventions, codes sociétaux, règles de communication, ou même une rupture monolatérale d’interaction ; avec l’explicite on risque en effet de choquer, d’ennuyer, d’impressionner désagréablement ou de faire réagir vivement. La permissivité offerte par la couche vocale aide l’énonciateur à contourner sans danger – ou presque – des interdits de toutes sortes, comme : désir de diminuer l’impact de la violence infligée, plaisir d’être blessant, malveillant ou brutal sans en subir les conséquences, indifférence aux ressentis de l’interlocuteur, ou même volonté de faire croire au consensus. L’analyse (signifiant sonore et signifié pragmatique) d’exemples oraux (enregistrés ou non) de locuteurs de français langue première est suivie de celle de leurs effets prévus ou imprévus sur l’écouteur.

A relative verb for Spoken French: The missing link

The morphosyntactic status of the French ‘subject clitics’ has long generated controversy within many schools of linguistics. Absent from this discussion, however, has been the status of the interrogative/relative pronoun qui, ‘who.’ When the tokens of qui in a corpus of French conversation are subjected to a series of tests designed to assess the degree to which it is analytic or synthetic (Schwegler, 1990; see also Zwicky and Pullum, 1983), the results suggest that we have missed a key member of the paradigm. Not only have the ‘subject clitics’ become affixes, creating a prefixally inflected verb paradigm and ensuing pro-drop typology through subject cycle renewal, but the interrogative/relative pronoun qui seems to be following suite. It seems to have split into a free interrogative pronoun, Qui? ‘Who?,’ and a bound morpheme, i.e., an inflectional prefix, 3REL-V, that occurs on the finite verb in subject relative clauses, e.g., cleft constructions, where it distinguishes discourse-new from discourse-accessible information found in main clauses and bearing a different inflectional prefix, il-V. These findings have important implications for our understanding of how grammar, discourse, and prescriptive factors intersect, as learners reanalyze the input in abductively innovative ways influencing the course of language change (Andersen, 1973).

Native speaker–non-native speaker interaction: More than one type of accommodation

The leading studies in sociolinguistics and conversation analysis regarding the accommodation theory in an Italian context (Grassi, 2007; Orletti, 2000; Vietti, 2008) provided the source and inspiration for this study. Two fieldwork studies will be discussed, in which the two opposing purposes – convergence or divergence – that accommodation strategies can be used for, are clearly apparent. All data is in Italian and was taken from a corpus of spontaneous conversations between native speakers (NS) and non-native speakers (NNS), observed and recorded in the urban context of the city of Turin. This represents a typical case of interethnic communication, one where issues of multilingualism and multiculturalism come to the fore. Two contexts in particular will be examined: a formal one in which NS adults interact with NNS adults at a railway station information desk, and an informal one where NNS adolescents converse with both NNS adolescents and NS adults. In the latter, the accommodation strategies will primarily be related to phonetic segments with strong symbolic value; whereas in the former, multiple accommodation strategies are used, among which most frequently the management of the linguistic code and the use of prosodic aspects.