Learning the Role of Intervocalic Constraints in Vowel-to-Vowel Dynamics

In a previous work (Coelho et al., 2002), we have shown that phonotactic constraints in the lexicon are robust enough to allow for the learning of the V-to-V relations of harmony and contour (the tendencies for vowels to share or avoid repetition of phonic properties). What rationalizes these results is that: (a) the environment is structured enough for a connectionist learning mechanism to extract the information needed (Elman et al., 1996), and (b) distributional biases in the lexicon code phonetic information (Albano, 2001). Here we report on experiments about the role of information carried by intervocalic segments in learning V-to-V relations.