Similarity Avoidance in East Bengali Fixed-Segment Echo Reduplication

Many languages employ reduplication processes in which one segment of the reduplicant is fixed. In East Bengali, /t/ is normally the initial consonant in reduplicants of this type (/ a i-ta i/ ‘cars, etc.’). However, when the base itself begins with /t/, speakers prefer alternate fixed segments, such as /f/ or /m/ (not */tæ a-tæ a/, but /tæ a-mæ a/ ‘cross-eyed, etc.’). Speakers also tend to avoid fixed segment /t/ when the base begins with a consonant that is similar to /t/ (not ??/t oŋ a toŋ a/, but /t oŋ a foŋ a/ ‘bags, etc.’). Results of an experiment indicate that bases beginning with consonants similar to /t/ (e.g. /t / or /d/) take reduplicants with fixed segment /t/ far less often than bases beginning with consonants dissimilar to /t/ (e.g. /p/ or /b /). If the avoidance of similarity between the initial consonants of base and reduplicant is indeed at work, on what basis is similarity being measured? To determine consonant similarity, speakers could be accessing language-specific measures of consonant similarity by observing the patterns in their lexicon, or they could be using a more universal measure of similarity, based on the phonological features of each consonant. To better understand the measurement of similarity, four metrics were tested against the experimental data. The current study explores a simple model in which similarity is determined by the weighted sum of shared features. One hypothesis regarding the source of these weights is considered here – a feature’s weight may reflect its ability to contrast the phonemes of the language.