ACOUSTIC ANALYSIS Acoustic methods applied to language Three main tools Praat¶ MATLAB Winpitch: F0, pitch tracking, pitch detection algorithms, on the fly alignment
Some PhDs on phonetics data collection / empirical - Adrien Méli: PhD on French pronunciation of vowels Maelle Amand: diphthongs in NECTE (co-supervision with Karen Corrigan and Aurélie Fischer) : analysing a Research project from the 1960’s with current data mining techniques : -Corpus analysis (vowel) Identification of variation / classification tasks - Aurélie Chlébowski : Acoustic analysis of nasal grunts -Esther Legrézause analysis of UM/HUM + misc. Analyses on data mining approaches, MA on accommodation: how you imitate natives as a learner (repetition task): Léa Burin
the kernel density estimate method Analysing the emergence of vowel categorisation in a longitudinal learner corpus: the kernel density estimate method Caen EPIP5 17 May 2017 Nicolas Ballier & Adrien Méli UFR études anglophones EA 3967, CLILLAC-ARP Univ Paris Diderot Sorbonne Paris Cité 3
Outline Research questions Corpus A Non-parametric approach: density estimation Unidimensional approach: Non-native and duration as cue to KIT/FLEECE Multidimensional approach: 3D representations of F1, F2 and duration The ‘window’ effect : the kernel method
Maptask from Anderson et al. 1991
Protocols and RQs Phonological awareness: Emphasis and read speech : ability to render italics (the ‘Landlady’) Stress patterns (Cameron’s speech) Missing: Perception tasks… Repetition tasks (ANGLISH) Reformulation tasks (LeaP)
Méli (PhD in progress)
DATA 20 female speakers / 5 male speakers Syllable features CVC / CV :
Speaker 20 Session 2: A preliminary analysis (Méli, in progress)
Unnormalized per-monopthong F1 and F2 values each dot represents the occurrence of one monophthong
Duration values for each monophthong (in seconds)
Duration for diphthongs
Per-diphthong distribution of the vectorial coefficients
DATA (duration of recordings)
MODELS Cosine transforms - filtering DATA lm models and repeated measures : mi random // word: fixed effects FREQUENCY OF LEARNERS: + FREQUENCY OF logistic regression -> parametric analysis Central limit theory >> to too many outliers. summarize the word: AVERAGE of word position PAIRWISE DIFFERENCES for time series CT ree ctree() C5.0 package rattle
Kernels (Paroissin 2015)
Density estimates Don’t use with this kind of data:
NO ASSUMPTIONS: PLOTTING THE DATA continuous functions : assumptions about the data : no discontinus DTAA - cross validation for the better - variance / bias (biased:one number // ) density estimates ONE D - bandwidth is a box: cross validation in boxes - bandwidth -> SAILING AND NORMALIZING THE DATA : means(=variance 1) SD divide by
Kde
3D representations of kde
Distribution per word
R packages NPDEN TESTING THE RELEVANCE NPUDIST Favorite mgcv PACKAGE library(splines)
DURATION > 0.03 s (aligners…)
Kde unidimensional density Bimodal distributions of duration ?
F2 x duration
Kernel effects (?)
Similar
Kitchen sink methods All learners All dimensions Pairwise comparisons for sessions
NEXT PLANS PhD viva R package with most of the coding Paper describing the syllable algorithm > Github Data paper with some subsamples of the data
REFERENCES Ballier, N., & Martin, P. (2013). Developing corpus interoperability for phonetic investigation of learner corpora. Automatic Treatment and Analysis of Learner Corpus Data, 59. Baayen, R. H. 2008. Analyzing linguistic data (Vol. 505). Cambridge, UK: Cambridge University Press.Best, C. T. 1995. A direct realist view of cross-language speech perception. In: Strange, W., (ed),Speech perception and linguistic experience: Theoretical and methodological issues. Baltimore: York Press, 171– 204. Bybee, J. 2007.Frequency of Use and the Organization of Language. Oxford: Oxford University Press. Bybee, J. 2010.Language, Usage and Cognition. Cambridge: Cambridge University Press Boersma, P. & Weenink, D. (2005). Praat: doing phonetics by computer (Version 5.3.71). Retrieved from http://www.praat.org. Bigi, B. (2012). Sppas: A tool for the phonetic segmentations of speech. In Proceedings of LREC 2012, pp. 1748–1755. De Cara B, & Goswami U. (2002). Similarity relations among spoken words: The special status of rimes in English. Behavior Research Methods, Instruments, & Computers, 34 (3), 416-423 R Development Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051- 07-0, URL http://www.R-project.org Gramacki, A., & Gramacki, J. (2015). FFT-Based Fast Computation of Multivariate Kernel Estimators with Unconstrained Bandwidth Matrices. arXiv preprint arXiv:1508.02766. http://arxiv.org/pdf/1508.02766.pdf Wand, M. and B. Ripley (2015). Functions for Kernel Smoothing Supporting Wand & Jones (1995). R package version 2.23-15. Wand, M. and M. Jones (1995). Kernel Smoothing. Chapman & Hall.
Some references Cauvin, E, 2013, « Intonational phrasing as a potential indicator for establishing prosodic learner profiles ». In S. Granger, G. Gilquin & F. Meunier (eds) (2013) Twenty Years of Learner Corpus Research: Looking back, Moving ahead. Corpora and Language in Use - Proceedings 1, Louvain-la-Neuve: Presses universitaires de Louvain, 75-88. Díaz Negrillo, A. 2007. A Fine-Grained Error Tagger for Learner Corpora. Ph.D. thesis, University of Jaén, Spain. Díaz-Negrillo, A., Ballier, N., & Thompson, P. (Eds.). 2013. Automatic Treatment and Analysis of Learner Corpus Data (Vol. 59). John Benjamins Publishing Company. Granger et. al., Sylviane. 2009. “The LONGDALE Project Longitudinal Database of Learner English”. http://cecl.fltr.ucl.ac.be/LONGDALE.html Gries, S. [2009] 2013. Statistics for Linguists: A Practical Introduction. Berlin, New York: Mouton de Gruyter Herment, S., Loukina, A., Tortel, A., Hirst, D., Bigi, B., 2013, AixOx, a multi-layered learners corpus: automatic annotation. In Díaz Pérez Javier & Díaz Negrillo Ana (eds.). Specialisation and variation in language corpora. Bern : Peter Lang. Méli, A., 2010, Aspects of a longitudinal corpus-based study of French learners of English. A preliminary investigation, Mémoire de Master 2 non-publié, sous la direction de N. Ballier, Université Paris Diderot. Méli, A. 2013, Phonological acquisition in the French-English interlanguage. Rising above the phoneme in Díaz-Negrillo, A., N. Ballier and P. Thompson (eds.), Automatic Treatment and Analysis of Learner Corpus Data, Amsterdam :Benjamins, 207–226. Milde, Jan-Torsten. & Gut, Ulrike (2002). “A Prosodic Corpus of Non-native Speech”. Speech prosody 2002. (10/09) http://aune.lpl.univ-aix.fr/sp2002/pdf/milde-gut.pdf Myssyk, A, 2011 Predicting and evaluating a speaker's level of English : a proposal for pronunciation criteria , Mémoire de Master 2 non-publié, sous la direction de N. Ballier, Université Paris Diderot. Tortel, A. & Hirst, D. 2008. ANGLISH. (10/09) http://crdo.fr/crdo000731 PRAAT : praat.org R : http://www.r-project.org/ SPPAS : http://aune.lpl.univ-aix.fr/~bigi/sppas/ WinPitch: winpitch.com
Gries, S. [2009] 2013. Statistics for Linguists: A Practical Introduction. Berlin, New York: Mouton de Gruyter Herment, S., Loukina, A., Tortel, A., Hirst, D., Bigi, B., 2013, AixOx, a multi-layered learners corpus: automatic annotation. In Díaz Pérez Javier & Díaz Negrillo Ana (eds.). Specialisation and variation in language corpora. Bern : Peter Lang. Méli, A., 2010, Aspects of a longitudinal corpus-based study of French learners of English. A preliminary investigation, Mémoire de Master 2 non-publié, sous la direction de N. Ballier, Université Paris Diderot. Méli, A. 2013, Phonological acquisition in the French-English interlanguage. Rising above the phoneme in Díaz-Negrillo, A., N. Ballier and P. Thompson (eds.), Automatic Treatment and Analysis of Learner Corpus Data, Amsterdam :Benjamins, 207–226. Milde, Jan-Torsten. & Gut, Ulrike (2002). “A Prosodic Corpus of Non-native Speech”. Speech prosody 2002. (10/09) http://aune.lpl.univ-aix.fr/sp2002/pdf/milde-gut.pdf http://bookzz.org/dl/2298946/b555a2
THANKS ! adrien.meli@gmail.com nballier@free.fr 37
ALTERNATES
Some Perspectives after Adrien Méli PhD Correlation with usage frequency ? Attractors (« lexical » magnets, modelling realisations for lexical sets (people) ? « Templatic effects » transfers of French syllable structures (CVC vs. CV)