@article {117, title = {Mutual intelligibility of Dutch and German cognates by humans and machines}, year = {2010}, publisher = {Nederlandse Vereniging voor Fonetische Wetenschappen}, address = {Utrecht, The Netherlands}, abstract = {

Many languages are so closely related that they are mutually intelligible to a certain extent. Mutual intelligibility between language pairs is sometimes asymmetric. For example, Danes understand Swedish better than Swedes understand Danish (Delsing \& Lundin {\r A}kesson, 2005; Gooskens et al., in press) and Brazilians understand (Argentinean) Spanish better than Argentineans understand (Brazilian) Portuguese (Jensen, 1989). These results are usually explained by extra-linguistic factors such as asymmetric attitudes towards the (speakers of the) languages involved and unequal experience with the languages. Additionally, linguistic differences can also be asymmetric and can also account for asymmetric mutual intelligibility.

This project aims to investigate factors determining intelligibility between Dutch and German. Dutch listeners (score: 87\%) understand German better than Germans (score: 73\%) understand Dutch (H{\'a}z, 2005). The most obvious explanation for this asymmetry is the fact that Dutch children learn German at school while Dutch is not a part of the curriculum for German children. The first aim of our investigation is to rule out the influence of education by testing mutual intelligibility of Dutch and German with children of 11 to 12 years who have not yet learnt the neighbouring language at school. We selected ca. 750 highly frequent Dutch-German cognate nouns (Celex corpora). These cognates were read aloud by a perfect bilingual speaker of German and Dutch and presented to the subjects in a translation task. Prior tothe intelligibility test, the German and Dutch subjects answered questions on their attitudes towards and experience with the neighbouring language and its speakers.

The second aim of our investigation is to model the mutual Dutch-German intelligibility through automatic speech recognition. Dutch cognates used in the intelligibility test (described above) were presented to a German speech recognizer and the German words to a Dutch speech recognizer of the same make (Dragon NaturallySpeaking Standard version 10). The speech recognizers had been trained by our bilingual speaker in exactly the same way in both languages. The percentage of correctly recognized words per language is our measure of intelligibility. If the German-Dutch intelligibility is asymmetric it can be assumed that non-linguistic factors do not play a decisive role and that the asymmetry can be explained at least partly by linguistic factors.

The results of the human-based experiment will be compared to the results of the computer-based experiment.

}, author = {Vincent van Heuven and Charlotte Gooskens and Ren{\'e}e van Bezooijen} }