Segmenting spoken corpora in lesser-described languages: new perspectives for the structural analysis of speech - Plenary presentation PPT - HAL-SHS - Sciences de l'Homme et de la Société Accéder directement au contenu
Communication Dans Un Congrès Année : 2013

Segmenting spoken corpora in lesser-described languages: new perspectives for the structural analysis of speech - Plenary presentation PPT

Résumé

Amina Mettouchi SLE 2013 PLENARY Abstract - PPT in Annex Segmenting spoken corpora in lesser-described languages: new perspectives for the structural analysis of speech. As a result of the generalization of text-to-sound indexing in the gathering and constitution of spoken corpora in lesser-described languages, the issue of speech segmentation has acquired new significance. Pioneering studies exploring the nature of speech units were conducted by discourse and conversation analysts, such as Chafe, DuBois, Local and others. In a different framework, hypotheses regarding the phonology/prosody or the syntax/prosody mappings were elaborated by formal linguists, Pierrehumbert or Selkirk among others. All those studies and the subsequent rich body of publications have paved the way for systematic empirical investigations of the relationships between various levels of segmentation or constituency, which are now made easier thanks to the text-to-sound indexing of spoken corpora. Spoken corpora in lesser-described languages contain several features that make them potentially amenable to empirical analysis : (a) text-to-sound indexation, (b) the delimitation of several levels of annotation and (c) searchability. Text-to-sound indexation raises the issue of transcription: phonetic vs. morphophonological vs. orthographic. Annotation requires that forms be given labels; this in turn gives renewed urgency to the issue of the delimitation of linguistic forms. Finally, searchability entails systematicity of representation and annotation. Underlying those various issues is the basic question of segmentation: how do we chunk the spoken continuum of speech in the language we are working on? This question can be done away with by an arbitrary decision (using orthographic transcription, and chunking between pauses above 300 ms for instance). But it can also be considered, especially for lesser-described languages where linguists usually start from scratch, as a way to address the question of the language-internal nature of different types of constituents, and their interaction. Viewed in this perspective, it is relevant at several levels. At the level of transcription, the delimitation of words is a challenge. How this question is answered depends on the perspective which is chosen: grammar, prosody, phonology, lexicon, etc. Again at the level of transcription, the delimitation of larger units (phrases, clauses, paragraphs, intonation units...) is not straightforward, and depends on the criteria that are chosen: prosodic, syntactic, pragmatic.., etc. At the level of annotation, the delimitation of forms (morphemes and constructions) is perhaps the greatest challenge, and is addressed by investigation methods such as contrastive and complementary distribution, among others. In my talk, drawing on the work I have carried out in the area of sound-aligned spoken corpora in lesser-described languages (in particular within the CorpAfroAs project), I will show how systematic segmentation of speech into non-aprioristic, formally defined units can be relevant for the analysis of languages. In order to achieve that goal, I will first review the literature on the delimitation of units of speech, at various levels : words (prosodic, phonological, grammatical), syntactic phrases, clauses, sentences, prosodic phrases, intonation units, periods, paratones. I will then zoom in on the segmentation of Kabyle (Berber) textual data into intonation units, and its interaction with the linear ordering of morphosyntactic constituents. The work I have carried out in this area has made it possible to discover some properties of Kabyle that had not hitherto been identified or described, with respect to the coding of information structure and the domain of grammatical relations : (a) subject and object roles are computable only in some information structure constructions, themselves defined by prosodic boundaries, morphological marking, and linear ordering of constituents ; (b) only some types of prepositional phrases are separated from the verb by a prosodic boundary – those which belong to the semantic valency of the verb. Such empirical investigations of the way various levels of segmentation (prosodic, morphophonological, syntactic, ...) interact are important for the study of lesser-described languages as complex systems : beyond the inventory of forms and the identification of their functions, it is now easier, thanks to recent technical advances, to use systematically segmented and annotated sound-aligned texts in order to investigate in greater depth the syntactic organization of speech as well as the dynamics of discourse.
Mettouchi 2013 SLE Plenary.pdf (9.65 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

halshs-01513604 , version 1 (02-03-2023)

Identifiants

  • HAL Id : halshs-01513604 , version 1

Citer

Amina Mettouchi. Segmenting spoken corpora in lesser-described languages: new perspectives for the structural analysis of speech - Plenary presentation PPT. 46th Annual Meeting of the Societas Linguistica Europaea (SLE 2013), Societas Linguistica Europeae, Sep 2013, Split, Croatia. ⟨halshs-01513604⟩
10 Consultations
2 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More