Variable wordforms, adaptable learners: evidence from real-time word comprehension and naturalistic corpora

When learning a language, typically-developing infants face the daunting task of learning both the sounds and the meanings of words. In this dissertation, we focus on a source of variability that complicates the one-to-one relationship between words and their meanings: wordform variability. In Chapter 1 we make a distinction between the micro timescale, where learning and acquisition can be observed during the comprehension of individual utterances, and the macro timescale, where a longitudinal lens can reveal intuitions about the speech that infants and toddlers can learn from. In Chapter 2, a corpus analysis confirms that infants hear more irregular verbs than irregular nouns. We then compare toddlers' phonological representations of irregular nouns and verbs to their regular counterparts in an eyetracking study. Toddlers demonstrate well-specified representations for verbs 8 months later than they do for nouns, but the sounds in irregular words are represented with high fidelity at the same time as regular words in both syntactic categories. In Chapter 3, an eyetracking study with adults explores word comprehension when coarticulation cues have been manipulated. Adults take available referents into account when they hear a word whose coarticulation cues match an unfamiliar word. We also find that participants vary in whether they map an unfamiliar word to an available unfamiliar object if the unfamiliar word is sufficiently similar to a relevant known word. In Chapter 4, we return to early childhood, using annotated data from a longitudinal corpus of naturalistic recordings from the lives of 6-17-month-old infants. In this corpus, we characterize and quantify wordform variability, and find that for high-frequency words, wordform variability may aid in word learning. Theoretical implications and next steps are discussed in Chapter 5. Taken together, this work suggests that throughout the lifespan, wordform variability is no problem for learners and can in some cases facilitate learning and comprehension.