Human interaction and the evolution of spoken accent

Project Information

InterAccent

Grant agreement ID: 742289

DOI

10.3030/742289

Project closed

EC signature date 24 August 2017

Start date 1 October 2017

End date 31 January 2023

Funded under

• EXCELLENT SCIENCE - European Research Council (ERC)

Total cost

€ 2 499 333,00

EU contribution

€ 2 499 333,00

Coordinated by

LUDWIG-MAXIMILIANS-UNIVERSITAET MUENCHEN
Germany

Periodic Reporting for period 4 - InterAccent (Human interaction and the evolution of spoken accent)

Reporting period: 2022-04-01 to 2023-01-31

Summary of the context and overall objectives of the project

The project's main concern was to develop a cognitive-computational model of sound change to predict the first stages of spoken accent development that come about when individuals in close contact and/or isolation interact with each other over a period of time. The high-risk aspect of the project derives from a combination of (a) tracking minute pronunciation changes to individuals over a timespan between several months to 4 years and then (b) testing whether these changes can be predicted using an agent-based computational model that is initialised with the spoken attributes of the speakers before they meet/interact with each other.

The project's contribution to society is at many levels. 1) The project is important historically by providing information about the origins of accents (e.g. why American English sounds different from British English). (2) The further social benefit lies in understanding how migration influences language change. 3) There are benefits to security because identifying the characteristics of a person's spoken accent by machine is important in forensic investigations of speech and language in e.g. assessing the likelihood that a person's voice corresponds to a recording made of the person. 4) There are social benefits to speech pathology since the acquisition of child speech data over a long period of time provides unique normative data against which pathological speech can be assessed. This is especially in communities investigated in this project such as Albania where such resources are still virtually nonexistent


The project's primary objective was to build a cognitive-computational model to explain how a spoken accent evolves. There were three associated secondary objectives. One was to explain how the interaction between individuals leads to innovations in spoken accent. Another was to explain how the accent of a community influences the spoken accent of an individual who belongs to, or who comes into contact with the community. Finally, another secondary objective was to understand how the characteristics of a spoken accent evolve from the ubiquitous phonetic variation in producing speech.

Overall, the project has shown through physiological and acoustical analyses of speech from children and adults across diverse dialects and languages that the direction in which spoken accent develops is influenced by spoken language contact and that such changes can be modelled and predicted with an agent-based model of sound change.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

1. Longitudinal acoustic and ultrasound recordings were obtained from primary schools in a rural area around 60 km from Munich. The results showed that Bavarian sounds in children were more like standard German than in adults but only if the children's Bavarian sounds were already skewed phonetically in the direction of the standard.

2. Similar recordings were obtained over four years from primary school children in a rural area of Albania and in the capital, Tirana. The results showed a shift of the rural variety towards the standard spoken in Tirana, more so in children than adults, but predominantly in vowels that do not function to distinguish between word meanings.

3. With regard to the development of a computational agent-based model of sound change in order to predict the direction of spoken accent development: the results of the simulations showed that the sounds of one group of dialect speakers that are skewed in the direction of another group of dialect speakers are likely to change through spoken interaction between speakers and listeners.

4. Recordings from real-time magnetic resonance imaging were made for the investigating of vowel nasalization in American and British English speakers. The results showed that the sound change by which vowels become increasingly nasalized over time is less advanced in British than American English.

5. Recordings from Antarctic 'winterers' were made before and during their long-term stay in Antarctica. The acoustic analysis provided evidence of the development of a common spoken accent whose direction could be predicted by the computational model in 1 above, using input data from the recordings before they went to Antarctica.

6. Summarising across 1-5 above, the project has shown that the direction of spoken accent change can be quite accurately predicted by a publicly-available, agent-based computational model of sound change. This issue has been addressed from numerous perspectives through analyses of varieties and dialects of Albanian, Chinese, English, French, German, and Spanish and disseminated in 26 international journal and conference proceedings publications as well as two awarded dissertations. The research project has resulted in over 20 reports in the media of which two were invited articles in a leading UK national newspaper. The longevity of the research is evident through collaborations with current scientists at other European universities who were former post-doctoral members of the project.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

1. The development of the agent-based computational model is beyond the state of the art because, in contrast to other models of sound change, it depends on exchanging real speech signals between actual speakers (whereas all other similar models use simulated data). This advancement provides the first ever computationally tractable association between imitation in human speech communication and the earliest stages of the development of sound change. It is also the only computational model of sound change that can model sound changes of mergers (e.g. in New Zealand English 'here' and 'hair' are both now pronounced as 'here') and splits (e.g. in American English a contrast is beginning to develop between oral and nasal vowels in some contexts).

2. The research project goes beyond the state of the art in developing techniques for modelling sound change by processing multiple, time-varying representations of the speech signal (such as the first two resonances of the vocal tract for understanding the type of 'here'/'hair' merger referred to above). Many empirically-based models of sound change derived from analyses of acoustic data are by contrast often based on processing single time points in the speech signal and very rarely on multiple time-varying parameters simultaneously.

3. The longitudinal analyses of speech, in which sound change is modelled by resampling from the same individuals over long time periods extends beyond the state of the art by combining evidence from multiple sources. These include: 1) the speech of an exceptionally large number of children for a study of this kind (e.g. 55 for Albanian) recorded at yearly intervals compared with adults also for the rarely studied language Albanian; 2) within the same individuals with respect to the effects on sound change over the lifespan caused by: i) contact with diverse speaking groups; ii) increasing age and memory loss; and 3) with regard to the relatively short-term sound changes in the speech of 'winterers' of the British Antarctic Survey before and after they spent time in Antarctica.

4. The comparison between two dialects that differ in the progression of sound change using high speed, real-time images of the vocal organs obtained from a magnetic resonance imaging scanner with 43 speakers extends beyond the state of the art given that physiological data from such a large sample of speakers has never before been applied to this type of problem.

interaccent.jpg