Analysing Rhythm in Ritual Discourse in Yucatec Maya Using Automatic
Total Page:16
File Type:pdf, Size:1020Kb
Analysing rhythm in ritual discourse in Yucatec Maya using automatic speech alignment Valentina Vapnarsky, Claude Barras, Cédric Becquey, David Doukhan, Martine Adda-Decker, Lori Lamel To cite this version: Valentina Vapnarsky, Claude Barras, Cédric Becquey, David Doukhan, Martine Adda-Decker, et al.. Analysing rhythm in ritual discourse in Yucatec Maya using automatic speech alignment. Interspeech 2015 Speech beyond speech, Sep 2015, Dresden, Germany. halshs-01250490 HAL Id: halshs-01250490 https://halshs.archives-ouvertes.fr/halshs-01250490 Submitted on 4 Jan 2016 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. INTERSPEECH 2015 Analysing rhythm in ritual discourse in Yucatec Maya using automatic speech alignment Valentina Vapnarsky1, Claude Barras2, Cédric Becquey1, David Doukhan2 ^, Martine Adda-Decker2 & Lori Lamel2 1Centre EREA du LESC, CNRS & Université Paris Ouest, France 2LIMSI-CNRS, Univ. Paris-Sud, 91403, Orsay, France [email protected], [email protected] Abstract very small-scale manual analyses, whereas it is clear that more precise, systematic and semi-automatized analyses would be in- Over the years, research in ethno-linguistics contributed to dispensable to properly test the anthropological and linguistic gather corpora in a wide range of languages, cultures and topics. hypotheses. In the present work, we are investigating ritual speech in Yu- In the present work, we are investigating ritual speech in catec Maya. The ritual discourse tends to have a cyclic structure Yucatec Maya. The ritual discourse tends to have cyclic organi- with repetitive patterns and various types of parallelisms be- sations with repetitive patterns and various types of parallelisms tween speech sections. Previous studies have revealed an intri- between speech sections. Previous studies revealed an intricate cate connexion between a speech’s structure and vocal produc- connexion between a speech’s structure and vocal productions, tions, in particular through temporal aspects including rhythm, in particular through temporal aspects including rhythm, pauses pauses and durations of different speech sections. To further and duration of different speech sections. To further investigate investigate our findings by relying more strongly on the acous- our findings by relying more strongly on the acoustic record- tic recordings, automatic speech recognition tools may become ings, automatic speech recognition tools may become of great of great help, in particular to test various linguistic and ethno- help, in particular to test various hypotheses on the relation be- linguistic hypotheses. Unfortunately, Yucatec Maya, with less tween the vocal production, the text and the performance. Un- than one million native speakers, is an under-resourced lan- fortunately, Yucatec Maya, with less than one million native guage with respect to digital resources. As a total, 24 minutes speakers, is an under-resourced language with respect to digi- of ritual speech from three performances were manually tran- tal resources. As a total, 24 minutes of ritual speech from three scribed by expert linguists in Yucatec and a basic pronunciation performances were manually transcribed by expert linguists in dictionary for Yucatec was created accordingly. The transcribed Yucatec and a basic pronunciation dictionary was created ac- acoustic recordings were then automatically time-aligned on a cordingly. The transcribed recordings were then automatically phonetic and lexical basis. Automatic segmentations were used time-aligned on a phonetic and lexical basis. Automatic seg- to measure tempo changes, durations of breath units as well as mentations were used to measure tempo changes, durations of to examine their link with the structure of the ritual text. breath units as well as to examine their link with the structure Index Terms: ethnolinguistic, Yucatec Maya, ritual discourse, of the ritual text. automatic alignment, phonetic segmentation, tempo. In the following, we first present the Yucatec Maya lan- guage and the ritual speech and vocal production. We then 1. Introduction describe the corpus of three transcribed speeches. Section 5 Over the years, research in social sciences like anthropology, presents the automatic alignment methodology before an anal- linguistics and ethnomusicology contributed to gather corpora ysis of the results in Section 6 and the conclusions and perspec- in a wide range of languages, cultures and topics. In this context tives in the last section. of growing amount of sound archive databases, the DIADEMS1 research project aims at integrating automatic audio analysis 2. Yucatec Maya language tools in an indexing platform for ethnomusicological and ethno- linguistic archives and initiated a collaboration between several Yucatec Maya is one of some 30 languages of the Mayan fam- research laboratories in computer science and social science [5]. ily. It is spoken by about 800,000 speakers in the northern low- As part of this project, a special research is conducted on Mayan lands of the Maya area, throughout the Yucatan Peninsula. It ritual speech to test ethnolinguistic hypotheses2. Ritual dis- shows less dialectal variation than Highland Mayan languages, course has been a main topic of research, with its pragmatic but three main regional dialects have been distinguished [19]. and enunciative specificities, but previous studies emerged from Yucatec Maya is a mildly polysynthetic, initial predicate, head- marking language. It has a complex voice and aspectual system, This work was partly funded by the French National Agency for with split ergativity [2, 1, 24]. Most of its lexical roots are CVC, Research (ANR) under grant ANR-12-CORD-0022-05 (project DI- where the vowel can have four distinct values (see below) [3]. ADEMS). Most clauses end by a final-clause vowel clitic, from a five-term ^D. Doukhan is now with IRCAM. 1http://www.irit.fr/recherches/SAMOVA/ deictic paradigm [12]. DIADEMS/en/welcome/ Yucatec Maya has a typologically original phonological 2Other participants are M. Chosson and A. Monod Becquelin on system. Firstly, two series of consonants are contrasted by a Tseltal Maya. glottalization feature: ejective consonants ([k’], [p’], [t’],[ts’], Copyright © 2015 ISCA 344 September 6-10, 2015, Dresden, Germany [tS’]“ ) vs. voiceless stops and affricates ([k], [p], [t], [ts], [tS]“ ). 4. Corpus description Secondly, its five vowels ([a], [e], [i], [o], [u]) may be realized The three speeches analysed come from a village located about in four distinctive ways involving length (short vs. long), and 30km. south of Felipe Carrillo Puerto (Eastern region of the Yu- only for long vowels, tone (high vv´ vs. low vv` ) and glottal- catan Peninsula). They were audio-recorded in 1995 and 1996, ization for long high-tone vowels (modal vv´ vs. creaky v’v) by V. Vapnarsky during two rituals dedicated to guardian-spirits [10, 6]. Vowel distinctions have lexical and grammatical func- of the forest. The same speaker, a well-known ritual specialist tions [15]. Besides the length opposition, vowel lengthening is in the sub-region, was involved in each case. He knew about a commonly used, yet unstudied, prosodic means of expression, and agreed to be recorded. The recording method was intended with specificities depending on speech genres. to be the least intrusive possible, it was done in the context of long term fieldwork, making the performance as natural as pos- Table 1: Consonants and vowels in Yucatec Maya. In italic, sible. Although the content of the discourse is very similar, the phonemes found in loan words and/or rare. speaker shows some stylistic differences or variations between the two performances, the analysis of which is part of our ob- Labial Alveolar Palatal Velar Glott. jective. In particular, in the second ritual he was influenced by Stop p t d ts ch [tS]“ k g ‘ another ritual specialist he had heard a couple of weeks before, Ejective p’ t’ ts’ ch’ [tS’]“ k’ and this is reflected by the more emphatic tone and melodic Implosive b contours of his speech. Fricative f s x [S] h Speech 1 and 2 were performed in a first-fruit ceremony Nasal m n for the maize harvest (ho’olbesah nal). In Speech 1, the ritual Trill r specialist goes with his voice to different places of the terres- Approx. l y w trial and cosmological world to gather the spiritual entities and bring them to the altar; in Speech 2, he brings them back to Front Back their dwelling place, after the offerings. Speech 3 corresponds Close i, ìi, íi, i’i u, ùu, úu, u’u to the first phase (spirit gathering) of a ritual dedicated to the Mid e, èe, ée, e’e o, òo, óo, o’o same entities, which was performed for prophylactic purposes. Open a, àa, áa, a’a The three speeches have similar lengths (1/ 13:54 ; 2/ 11:25 ; 3/ 13:15) and their textual structure is very homogeneous. The general organisation of the text is the following: contextual an- 3. Yucatec Maya ritual speech and vocal choring of the ritual (space, time, sponsor, purpose); long invo- production cation of the spiritual entities with textual cycles (see extract in Table 2); credo (part of the offering); invocation of saints and Yucatec Maya has various elaborated ritual speech genres, per- other spiritual entities; closing. The speaker makes the sign of formed in agricultural, cynegetic and therapeutic ceremonies the cross at specific moments, kissing his hand (kissing-sound dedicated to, invoking guardian-spirits of the land and the for- audible in the recording). est, as well as in religious festivals with omnipresent prayers The recordings were manually segmented into vocal units and ritual dialogues [11, 13, 20, 22, 23].