DescriptionHow do children learn the meaning of words like “pretty” and “tall,” which are not only gradable and context dependent (Kennedy & McNally 2005), but encode speaker subjectivity? Despite their complex semantics (Stephenson 2007; Lasersohn 2009; Bylinina 2014), these and other adjectives like them, are some of the most frequently produced adjectives by children and their caregivers. How do children map the right meaning to these adjectives early in language acquisition? In this dissertation, I present the results of a corpus-based analysis of ambient language, and a word learning experiment using an adapted human simulation paradigm (Gillette et al. 1999) with scripted dialogues (Yuan & Fisher 2009; Arunachalam & Waxman 2010) demonstrating the influence of the syntactic environment in which these adjectives appear. Although previous literature has extensively explored syntactic bootstrapping in the verbal domain (Landau & Gleitman 1985; Gleitman 1990; among many others), few researchers have extended it to adjectives (Syrett & Lidz 2010; Becker 2015). Here I push this frame- work further within the adjectival domain, investigating it through the lens of subjective adjectives. I focus on five subclasses (TOUGH, SMART, PRETTY, TASTY, TALL). While they share overlapping properties, each is distinguished by a unique syntactic signature. Corpus results indicate distributional differences among adjectival subcategories, indicating that they pattern in predictable ways in the input. In the word learning experiment, participants were presented with a set of syntactic frames, with each set corresponding to one of the five classes, and asked to provide the meaning for a novel adjective along with their confidence level. Participants either responded incrementally after each frame or after all of the frames had been presented. They not only displayed an awareness of how syntactic frames narrowed the potential meaning for each adjective, but participants in the incremental conditions performed significantly better than those in the conditions in which they guessed after the frame set, and became more confident as they received more frames. Based on these results, I propose that adjectival (sub-)categories are assigned probabilities that shift as a function of data encountered that is either consistent or inconsistent with a particular categorization (Reiger & Gahl 2004; Yang 2002). Frames are weighted according to their relative informativity. As learners encounter more frames, they gradually update their hypothesis space, and ultimately assess the entire cluster of frames, consistent with previous proposals by (Naigles 1996; Mintz 2003) for verbs.