We use reinforcement learning to learn tree-structured neural networks for computing representations of natural language sentences. In contrast with prior work on tree-structured models, in which the trees are either provided as input or predicted using supervision from explicit treebank annotations, the tree structures in this work are optimized to improve performance on a downstream task. Exp...

Embodied semantics proposes that constructing the meaning of motion verb phrases relies on representations of motion in sensory cortex. However, the data reported by earlier studies as evidence for this claim are also explained by a symbolic-semantics view proposing interactions between dissociable systems. In the experiments reported here, participants were visually adapted to real and implied...

A new version of the MOSAIC model of syntax acquisition is presented. The modifications to the model aim to address two weaknesses in its earlier simulations of the Optional Infinitive phenomenon: an over-reliance on questions in the input as the source for Optional Infinitive errors, and the use of an utterance-final bias in learning (recency effect), without a corresponding utterance-initial ...

Research on early word learning reveals that verbs present a unique challenge. While English-acquiring 24-month-olds can learn novel verbs and extend them to new scenes, they perform better in rich linguistic contexts (when novel verbs appear with fully lexicalized noun phrases naming the event participants) than in sparser linguistic contexts (Arunachalam & Waxman, 2011; Waxman et al., 2009). ...

Sentiment analysis is an emerging research field. One of the major tasks of sentiment analysis is building sentiment lexicons and calculating their scores, which is an essential job that provides “material” for all sentiment analysis problems. In this paper, we propose a fuzzy language computation by taking linguistic context into account to provide an effective method for computing the sentime...