Speech variability: the nuances of phonetic reduction in conversational speech

We are pleased to announce the latest publication by Kübra Bodur (LPL) in Speech Communication, in collaboration with Corinne Fredouille (LIA), Stéphane Rauzy and Christine Meunier (LPL).

Reference: Bodur, K., Fredouille, C., Rauzy, S., & Meunier, C. (2025). Exploring the Nuances of Reduction in Conversational Speech: Lexicalised and Non-Lexicalised Reductions. Speech Communication, 173, 103268.

Full text article: https://www.sciencedirect.com/science/article/pii/S0167639325000834

 In addition, on 29 October, Kübra will defend her thesis, which aims to better understand the phenomenon of speech reduction, based on empirical studies in phonetics, prosody and developmental research.

Figure 1

Abstract:
Phonetic reduction, understood as the set of segmental modifications that result in a weakened, shortened, or simplified realization of an expected form, is a central phenomenon of spontaneous speech. This study aims to gain a better understanding of the mechanisms of phonetic reduction, whose influence may vary depending on the characteristics of linguistic units.
We propose to distinguish two types of reductions: lexicalized forms, which have become more or less stabilized in the language (such as chépa for je ne sais pas (Figure 1)), and non-lexicalized forms, which are more contextual and vary depending on speech rate or speaker style.
To test whether these two types of reduction are influenced by the same factors or not, we combined an analysis of frequent and perceptible sequences with an automatic corpus-based detection, focusing on temporally reduced sequences.
The results show that lexicalized reductions are mainly related to speech duration, while non-lexicalized reductions are more strongly influenced by articulation rate and by certain prosodic or morphosyntactic constraints. We also observe that some segments (Figure 2) and certain expressions, such as tu sais, are more often reduced depending on their discourse function.
This study thus reveals that reduction in speech is not a single phenomenon, but rather a set of distinct processes shaped by different factors. It opens the way to a better understanding of speech variability and its relation to the structure of language.

The maths of speech!

The LPL has been invited one more time at the Maths Fair, a popular science event organized by the association “Maths for everyone”, in partnership with the CIRM. The fair, which invites primary, secondary and high school students to discover mathematics from a different angle, was held on Thursday, May 11 in the town of Rognac (34 booths in total).

This event offers conferences, numerous stands and activities provided by provided by researchers, teachers and students aiming to promote mathematics.

The LPL booth was about “The maths of speech” and was animated by Alain Ghio and Audrey Fokeer, a master student in language sciences and currently an intern at the LPL. They presented to the public the functioning of speech, with demonstrations with Coke bottles to understand the phenomena of geometry and resonance. Examples of the sounds of Tanzania’s languages or of patients with voice disorders were also discussed. Finally, they offered a demonstration of an automatic online transcription system which – despite the noise – is working better and better.

As a reminder, in 2019, Alain Ghio and Gilles Pouchoulin had already participated in the same fair in Aix-en-Provence.

Credits: A. Ghio and A. Fokeer

 

Field mission in Tanzania: recording of the Iraqw and Hadza languages

Alain Ghio, research engineer at LPL, and Didier Demolin, professor at the Laboratory of Phonetics and Phonology (LPP) in Paris, carried out a joint mission from 13 to 26 February in the northeast region of Tanzania. The purpose of this mission is to record and study two languages ​​of this region: the language of the Iraqw (Kwermusl village) and the language of the Hadzabe (Mwangeza). The first language is distinguished by ejective consonants for which the larynx makes surprising movements. The second is a language spoken by a tribe of hunter-gatherers, which has 65 consonants, including a dozen clicks.

But what did they go to do more specifically?
Experimental field linguistics. The association of these different words is generally difficult to reconcile. And yet, that’s what they did. They went to meet Iraqw and Hadza speakers on the spot, in the bush, in Tanzania, and they recorded them with advanced laboratory techniques: aerophonometry, electroglottography, lab films, palatography ... which is a scientific feat almost new. Using these techniques in the laboratory in Amsterdam, Marseille or Los Angeles is common. Do field surveys with a portable pocket recorder. That is done. But doing aerophonometry in the middle of the savannah is a challenge. Didier Démolin is the specialist. Alain Ghio is co-inventor of the EVA2 aerophonometry device and masters the technical constraints of this type of experiment. [...]

For more information: Alain Ghio / Didier Demolin

See also the article published by the Laboratory of Phonetics and Phonology (LPP), partner of this project (in French):

Mission de collecte de données phonétiques en Tanzanie