This book investigates the role that ambiguous evidence can play in the acquisition of syntax. To illustrate this, the book introduces a probabilistic learning model for syntactic parameters that learns a grammar of best fit to the learner's evidence. The model is then applied to a range of cross-linguistic case studies - in Swiss German, Korean, and English - involving child errors, grammatical variability, and implicit negative evidence. Building on earlier work on language modeling, this book is unique for its focus on ambiguous evidence and its careful attention to the effects of parameters interacting with each other. This allows for a novel and principled account of several acquisition puzzles. With its inter-disciplinary approach, this book will be of broad interest to syntacticians, language acquisitionists, and cognitive scientists of language.

Thesis: Ph. D.-- Linguistics, Massachusetts Institute of Technology, Department of Linguistics and Philosophy, 2015.

Preface --Introduction -- A model for language learning -- Core themes : a first look -- Revisiting the core themes : a look at some similar learning models -- The puzzle of ambiguous evidence -- Preliminary considerations -- The general case -- The subset case -- Ambiguous evidence and modeling learner errors and variability -- Errors -- Variability -- Ambiguity and development -- The learning model -- Introduction -- Overview of the model -- Introducing the model with two toy examples -- The learning procedure : a summary -- Ambiguous vs. unambiguous input -- Prior probabilities and the update procedure -- Generality of the model and learning results -- Comparison with other learning models -- Sakas and fodor (2001) : the structural triggers learner -- Gibson and wexler (1994) : the triggering learning algorithm -- Yang (2002) : the naive parameter learner -- Summary -- The acquisition of verb movement in swiss german : modeling child production errors and variability -- Introduction -- The core data of verb placement in Swiss German -- Adult grammar -- Child productions -- Some possible analyses -- Alternative #1 : overgeneralizing V2 in embedded clauses -- Alternative #2 : extraposition in embedded clauses -- Alternative #3 : overgeneralizing VR, VPR -- Schonenberger's analysis : verb movement in embedded clauses -- A learning model for the acquisition puzzle -- Analysis of the adult and child grammars -- Overview of the model -- Insight of the model -- Predictions for the model -- Results and discussion -- Priors and update procedure -- Results -- A closer look at the acquisition data : the distribution of subjects in embedded clauses -- Comparison with other learning models -- The broader German perspective -- The relation between input and learning -- Summary -- Head-finality and verb movement in Korean : modeling variability and non-variability across learners -- Introduction -- Modeling the effects of parameter interaction : the core example -- A schematic version of the model : learning in a 3-parameter hypothesis space -- Results of the 3-parameter model -- Making the model more general : a simplified Korean -- Expanding the hypothesis space -- Expanding the corpus -- Predictions for the model -- Han et al. (2007) and the current model : a deeper look at modeling end-state variability -- Review of Han et al. (2007) -- Comparison of Han et al. (2007) and the current model -- Toward a unification of Han et al. (2007) and the current model -- Results and discussion -- Results of the 5-parameter model -- Variability with a probabilistic learner : a broader perspective -- Comparison with other models -- Constraining the model : a first attempt -- Summary -- The case of zero derived causatives in English : learning from implicit negative evidence -- Introduction -- Pylkkanen (2008) and the learning challenge -- Review of pylkkanen (2008) -- The learning challenge -- Addressing the challenge -- Learning from implicit negative evidence : the case of zero-derived causatives -- Making the model more general -- Results -- Learning the grammar of the superset language -- Comparison with other models -- Summary -- Learning biases -- Introduction -- A problem for defaults -- Errors in Swedish -- Errors in English -- Toward accounting for the Swedish and English errors -- Constraining the model : a new proposal -- Summary -- Final summary -- References -- Swiss German input types and corresponding compatible grammars -- Additional evidence for a Root-selecting grammar in English? -- Index.