15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003


Objective Measurement of Intelligibility

Jared Bernstein

Ordinate Corporation, USA

Intelligibility is not often reported in language testing because measurement procedures have practical problems. Intelligibility is sensitive to the speaking skill of the candidate as well as to the predictability of the material and the intelligence and experience of the listeners. A given nonnative speaker will be more intelligible when saying predictable material to intelligent listeners who have experience with nonnative speech. Also, in a real language assessment procedure, after a listener has heard one candidate speak on a particular subject, that listener is no longer "naive" and will be better able to understand the next candidate's speech sample. An experiment conducted at Indiana University attempted to establish a stable intelligibility scale for nonnative speech. Utterances from 485 nonnative speakers of English were presented to 141 naive native listeners in such a way that no listener heard the same item from more than one speaker. This produced over 27,000 individual listener responses. Analysis of these data allows the nonnative speakers to be placed on an intelligibility scale that is reliable and clearly interpretable in terms of percent of words correctly heard for materials at a known level of predictability.

Full Paper

Bibliographic reference.  Bernstein, Jared (2003): "Objective measurement of intelligibility", In ICPhS-15, 1581-1584.