Distance from a model accent as a correlate of perceived "non-nativeness"

Jaroslaw Weckwerth

School of English, Adam Mickiewicz University, Poznan, Poland, Poland


This paper presents the results of one in a series of surveys exploring accent perceptions and attitudes in Polish learners of English. A methodology generally similar to that used by Gooskens and Heeringa (2004) was used to study the correlation between the learners' perceptions of "nativeness" and phonetic distance between the speech samples used and a model accent.

Reading passage samples of native Australian, New Zealand and South African English were played to a panel of Polish judges, along with samples of L2 English from speakers of Polish, Afrikaans, Thai and Mandarin Chinese. The judges -- students of "English philology" at a Polish university after an advanced two-year pronunciation course using a standard British English model -- assessed their "nativeness" and attractiveness. Importantly: (1) there were no samples of the model accent spoken by speakers who could be traditionally construed as "native"; (2) two of the non-native samples could be (impressionistically) considered close to the model; (3) the native-accent samples contained salient features differentiating them from the model; (4) the remaining non-native samples also contained salient non-standard features.

The resulting Likert-scale scores were subjected to a clustering analysis, and the mean scores were compared. The two non-native accents closest to the model scored highest both on the "nativeness" and attractiveness scales, and clustered together. The "strong" native accents occupied the "middle ground", also clustering together. The "strong" non-native ones received the lowest ratings.

A measure of phonetic distance from the model, based on the Levenshtein distance, as used in dialectometry research (cf. e.g. Heeringa 2004; Gooskens and Heeringa 2004), was computed for each sample. This was also subjected to a cluster analysis, and the clusterings proved generally similar to those obtained from the "nativeness" and attractiveness scales. For most of the speakers, there was a good correlation between the two sets of results.

It seems that the Levenshtein distance can be used as a measure of accent distance for non-native speakers of a language as well as native ones; and that the perception of "nativeness" in EFL learners is based on the distance of a speaker's accent from the model accent used in teaching.


Gooskens C. and W. Heeringa. 2004. Perceptive evaluation of Levenshtein dialect distance measurements using Norwegian dialect data. Language Variation and Change 16(3): 189–207.

Heeringa, W. 2004. Measuring dialect pronunciation differences using Levenshtein distance. PhD dissertation, University of Groningen.

(Word count: 404)

Session: Paper session
Attitude 4
Saturday, April 5, 2008, 09:00-10:30
room: 09