Learning Words amidst Phonemic Variability

Learning language is one of the most complex and intriguing feats of learning that humans achieve; but how we do it remains obscured behind many layers of questions we have yet to answer. Each small advance, however, moves us one step closer to understanding this incredible process. One aspect of this problem that has been explored substantially is word learning. In principle, understanding how a small number of words are learned in the lab could be scaled up to an entire language, giving scientists a more manageable task. If we can understand how different populations learn a small number of words under highly controlled conditions, understanding language acquisition in the real world becomes a task of exploring all of the conditions that may occur in everyday language use. In order to approach the problem of vocabulary acquisition, many researchers have classically treated linguistic elements (speech sounds, words) as discrete symbols. If each word is made up of specific, unchanging building blocks, then word learning can be broken down into learning these building blocks and then using that small number of building blocks to store representations of words. In this account, new words are learned when new combinations of symbols are encountered. In English, for example, the sounds that differentiate the word “cat” from the word “cut” are considered different phonemes. That is, exchanging the “a” for the “uh” sound is a clue to word learners that they are hearing a new word. These symbol-like sound categories, known as phonemes, indicate differences in meaning between words (Liberman et al., 1957). Once discrete phonemes are learned, it is argued, speakers can efficiently code learned words and identify when they are hearing a new word. Most language acquisition models describe phonemes as crystallizing into fixed categories by early childhood (Kuhl, 2000), making word learning as simple as identifying new words by their component speech sounds. However, other evidence suggests that phoneme representations are more flexible. Children can recognize words even if a constituent phoneme is violated (“fish” pronounced as “fesh”; Creel, 2012), and adults can learn labels for the same object differing minimally in vowel phonemes (e.g. ziv/zev) as readily as they learn a single label (Muench & Creel, 2013). Here we explore the role of the phoneme as the building block of word form identity, and seek to deliver a more