Perceptual Similarity among Nasals Varying in Place of Articulation: a Multidimensional Scaling Analysis

To examine perceptual similarity in cross-language speech perception as a multidimensional metric, an AXB classification test using nasal consonants from Malayalam ([m], [n+ ], [n], [÷], [ø], [N]) was administered to three sets of listener groups with common coronal nasal inventories: dental-retroflex (Marathi, Punjabi), alveolar-retroflex (Tamil, Oriya), and alveolar (Bengali, American English). A two dimensional multidimensional scaling analysis of the results showed several interesting patterns. First, cross-linguistically, the stimuli frequently clustered into three groups, {[m]}, {[n+ ], [n], [÷]}, and {[ø], [N]}. Second, languagespecific differences were observed that were not predictable from the test groupsÕ nasal inventories. The dental-retroflex and alveolar groups showed intra-group differences in their clustering of stimuli and weighting of both perceptual dimensions, leading to language-specific perceptual spaces. Only the alveolarretroflex group spaces were similarly organized. The results demonstrate that descriptions of native categories must include language-specific weighting of acoustic cues.