The Complementary-Distribution Challenge for Phonological Learning: The Case of [h] and [ŋ] in English

The distribution of [h] and [ŋ] in English poses a well-known challenge for the structuralist assumption that segments in complementary distribution are allophones of the same phoneme. In this paper, we show that this distribution also poses a challenge within generative linguistics, a challenge of explanatory adequacy. The challenge is to explain why the English-learning child does not acquire a grammar in which [h] and [ŋ] are derived from the same underlying representation. The generative literature has not identified this as a problem, largely because of the assumption that [h] and [ŋ] are too phonetically dissimilar to participate in an alternation. However, we argue that phonetic accounts do not eliminate the challenge. We discuss simplicity-based theories of learning and show that under naive representational assumptions, simplicity-based learners prefer to derive [ŋ] from an underlying /h/, rather than a longer sequence of /ng/ (or an underlying /ŋ/), and therefore make wrong predictions about English speakers’ knowledge. The same learners, however, make correct predictions when combined with a representational framework that assumes underspecification and lexical redundancy rules. We therefore conclude that the acquisition problem can be addressed using simplicity-based learning, and that the complementary distribution of [h] and [ŋ] provides an argument for underspecification and lexical redundancy rules.