Continuous speech segmentation is a cognitive process that humans engage in daily. Despite the lack of explicit word boundaries, speech comprehension seems effortless. A challenge may arise when encountering pairs of phonemically identical determiner phrases (DPs) in elision situations, like "l'affiche" (“the poster”) - "la fiche" (“the file”), both produced as /lafiʃ/. Accurate segmentation becomes crucial for comprehension in these cases. Previous research shows that French listeners surpass chance levels in discriminating and identifying such DPs. Acoustic analyses unveil differences in parameters like duration and fundamental frequency (F0). While these cues are not typically used in French speech segmentation, these findings suggest their potential exploitation for accurate segmentation.The first part of this dissertation includes two experiments examining the interplay of acoustic and semantic cues in processing French homophonic DPs, like l'affiche-la fiche or l'amie-la mie (/lami/). We aimed to ascertain the extent to which these cues are used when acoustic and contextual information is available. Behavioural measures may not directly elucidate speech segmentation, but they likely provide insights into post-segmentation processes. To address cue modulation during the perception of homophonic DPs, we conducted two EEG experiments to capture neural correlates by exploiting low-level (acoustic) and high-level (semantic) cues. A semi-passive sentence judgement task (Experiment 1) and a word identification task (Experiment 2) were administered to 51 French-speaking participants, manipulating attention towards sentence meaning (Experiment 1) or lexical form (Experiment 2). Acoustic-phonetic cues in disfavouring contexts were expected to yield P3 and N400 ERP components compared to favouring ones, reflecting unexpected acoustic detection and lexical retrieval difficulty. Larger P3 amplitudes were found in favouring contexts for homophonic DPs in both experiments. The reverse effect was found significant only in Experiment 2, with larger N400 amplitudes for disfavouring contexts. Our findings suggest that French speakers detect acoustic-phonetic information in homophonic DPs in elision. However, lexical access seems more challenging when attention was on lexical than sentential information. High-level information seems relevant during sentence processing, while low-level information is salient enough to be detected and processed.The second part explores how two populations navigate segmentation challenges in the absence of context: adults with Developmental Dyslexia (DD) and French learners. Experiment 3 tested 49 DD adults and 49 typical readers through a word repetition task. We assessed the role of two F0 parameters (slope and mean), exploring the impact of phonological deficits in DD on speech segmentation. Increasing the mean F0 of /a/ was expected to enhance vowel-initial item segmentation for typical readers, but individuals with DD were expected to show reduced sensitivity. Results showed increased vowel choices based on F0 properties, regardless of group. No interaction between conditions and groups suggests that both groups display similar segmentation strategies. Our findings underscore the crucial role of both F0 mean and slope in accurate French segmentation.Finally, Experiment 4 tested 31 native French speakers and 22 French learners (11 English, 11 Spanish speakers) through an AX discrimination task. We examined the sensitivity to acoustic-phonetic information of French homophonic DPs. English and Spanish speakers use different cues than French, relying on duration, F0, and intensity, and lack an elided form of the definite article. Results revealed that learners outperformed native listeners in discriminating homophonic DPs. Learners benefited from their L1 strategies in French segmentation, suggesting that cues embedded in elision guide both native and non-native listeners towards word segmentation.