An identical intelligibility test was run in Grenoble with twenty French listeners and in Abidjan with twenty Ivorian listeners for whom French is a second language. The listeners had similar educational backgrounds, but differed in their linguistic competence with regards to the comprehension of French as spoken in France. Each subject had to manually transcribe 500 syntactically simple and semantically unpredictable sentences consisting of the most frequent French monosyllables. There were five stimulus series, one of spectrally distorted natural speech, and four of synthetic speech, produced by cross-varying two coding techniques and two prosodic models. The natural speech was obtained from the same French speaker whose voice served as reference in the text-to-speech synthesizers. The two automatic prosodic models were designed on the basis of French as spoken in France. Results were compared across the two groups of subjects. Overall intelligibility scores, measured as the number of correctly transcribed sentences, fully discriminate the performances of the Ivorian and the French listeners. Nevertheless, the perceived linguistic complexity is independent of the listener's origin. The linguistic complexity is here quantified in terms of the number of independent decision units each subject dealt with when transcribing responses. This number is calculated as the ratio r = log (Ps) / log (Pw) where Ps is the proportion of correct sentences and Pw the proportion of correct words. This index is strongly dependent on the linguistic redundancy contained in well-formed sequences. It is here used in order to estimate the ability with which listeners take advantage of syntactic information when decoding semantically anomalous but syntactically correct sentences.