Probabilistic Modeling in Psycholinguistics: Linguistic Comprehension and Production

摘要: It must certainly be accounted a paradox that probabilistic modeling is simultaneously one of the oldest and newest areas in psycholinguistics. Much research linguistics psycholinguistics 1950s was statistical probabilistic. But this disappeared throughout 60’s, 70’s, 80’s. In highly unscientific survey (conducted by myself) six college textbooks handbooks published last 10 years, not single them mentions word ‘probability’ index. This omission astonishing when we consider input to language comprehension noisy, ambiguous, unsegmented. order deal with these problems, computational models speech processing, contrast, have had rely on for over 30 years. Computational techniques processing text, an medium which much less noisy than speech, just as heavily probability theory. Just pick arbitrary indicator, 77% papers year 2000 annual conference Association Linguistics relied or learning. Probability theory best normative model solving problems decisionmaking under uncertainty. perhaps it good model, but bad descriptive one. Despite fact originally invented cognitive human reasoning uncertainty, people do use tasks like production comprehension. Perhaps simply non-optimal, non-rational process? decade so, there emerging consensus cognition rational, relies processing. The seminal work Anderson (1990) gave Bayesian underpinnings memory, categorization, causation. Probabilistic cropped up many cognition; area are number recent categorization (Rehder 1999; Glymour Cheng 1998; Tenenbaum 2000; Griffiths 2001b; 2001a), also now finally being applied psycholinguistics, drawing from early Bayesian-esque precursers perception such Luce (1959) choice rule Massaro. What does mean claim probabilistic? has implications comprehension,