Abstract of work to appear at SSSR 2012, Montreal. Gwen Frishkoff (Georgia State University); Kevyn Collins-Thompson; Charles Perfetti; Scott Crossley - A test of incremental and adaptive word learning from context Purpose: Previous studies (Frishkoff, et al., 2011) have shown that MESA (Markov Estimation of Semantic Association) can be used to capture incremental gains in word knowledge over time. This study examined whether MESA is sufficiently robust to track these changes on a trial-by-trial basis. Method: English-speaking (L1) and Spanish-speaking (L2) participants were exposed to 45 very rare words in six different sentences. After each sentence, participants generated a synonym (or near-synonym) for the target word. MESA modeled response accuracy (distance from target meaning). Half of the participants received accuracy feedback ; the other half received the same amount of practice, without feedback. The degree of contextual support (high, low, or mixed-constraint) was systematically varied. Results: Replicating previous studies, the mean trajectories differed for words that were presented in high, low, or mixed-constraint contexts. Further, the MESA scores for individual words on individual trials generally increased over time and showed sensitivity to contextual support. Ongoing work is examining the effects of language proficiency and the role of feedback based on MESA scores, as well as delayed effects on familiarity and degree of semantic knowledge. Conclusions: Our findings suggest that MESA can be used to track changes in word-specific knowledge on a trial-by-trial basis. Such dynamic assessment is critical for effective instruction because robust word learning requires multiple encounters with a word in a variety of contexts. This approach should result in instruction that is both effective (i.e., robust learning) and efficient (i.e, tuned to support maximal gains on each exposure).