Identification of regional accents in French: perception and categorization

Abstract

ABSTRACT This article is dedicated to the perceptual identification of French varieties by listeners from the surroundings of Paris and Marseilles. It is based on the geographical localization ofabout forty speakers from 6 Francophone regions: Normandy, Vendee, Romand Switzerland, Languedoc and the BasqueCountry. Contrary to the speech type (read or spontaneous speech) and the listeners’ region of origin, the speakers’degree of accentedness has a major effect and interacts with the speakers’ age. The origin of the oldest speakers (who have the strongest accent according to Paris listeners’ judgments) is better recognized than the origin of the youngest speakers. However confusions are frequent among the Southernvarieties. On the whole, three accents can be distinguished (by clustering and multidimensional scaling techniques): Northern French, Southern French and Romand Swiss. Index Terms : perceptual dialectology, French regionalaccents 1. INTRODUCTION In speech communication, considerable variability comes into play as a function of regional and social accents. Humans must cope with it, like machines for automatic processing. In order to account for various accents that may exist in French, a perceptual dialectology approach is developed, from speech corpora which computational techniques help analyzing.We now have at our disposal a large number of recordings collected in different places of the French-speaking area,within the framework of the PFC project (“Phonology ofContemporary French”) [1]. In the audio data, various accents are represented, as many deviations with respect to a norm. They may be detected by certain phonetic features which aresalient enough to be recognized and characterized. Are naive listeners capable of identifying these accents? How fine-grained is the distinction between, for example South-Westand South-East accents? How many accents are experts or non-specialists able to discern? What is the impact of the requested subjects’ geographical background? Thesequestions are not novel in dialectometry [2], even if we here distinguish between accents and traditional dialects. Clopper and Pisoni [3] recently demonstrated that, without prior training or feedback, American listeners, invited to listen to compatriots of various accents and to locate theirgeographical origin on a map of the USA, are able todistinguish three broad regions: New England, South,North/West. Studies were also devoted to British English,Dutch and Norwegian varieties [4, 5]. To our knowledge,results of a similar perceptual clustering task are not availablefor the French language, even though a North/Southdistinction sounds obvious to any French speaker. Toquantify to what extent listeners are able to determine the origin of a speaker, the key difficulty is that someparameters may be distinctive to certain listeners but not to others. A number of factors may affect the judgmentsandthe awareness of identifiable differences. Based solely on pronunciation, locals are expected to perform with afiner-grained perception than non-locals. In sociolinguistics,studies focus on the representations of specific varieties, stored in long-term memory, more or less stereotyped, and possibly different from behavioral reactions to actualspeech samples [6, 7]. However most studies aredescriptive, and their findings do not enable us to reliably predict the most salient properties that are associated to a given accent.The PFC project now allows systematic investigations. Oneof its objectives is to cover a vast territory throughoutinformants who are firmly rooted geographically — who were born and have spent most of their lives in the same place. Since several age categories and speech “styles”(reading and interview especially) are represented in the corpus, their influence on the listeners’ performance alsodeserves being quantified. We examined this issue in two steps, with a pre-test whose goal was a mere accent rating and two experiments strictly speaking of accent originidentification, on the basis of about forty speakers from sixFrancophone regions.It is indeed interesting to carry out perceptual tests in different locations, to compare theresults: the experiments we conducted involved listenersfrom the Paris region (pre-test and Experiment 1) and the Marseilles region (Experiment 2). The next section presentsthe corpus, the listeners’ tasks and a phonetic analysis of the speech material. Results are reported in section 3 and further analyzed in Section 4, before concluding remarks.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 62 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.