Semantic server

点击翻译按钮获取中文摘要

BEAGLE (Bound Encoding of the Aggregate Language Environment) is a computational model of learning; applied to language, it acquires the meaning of words by "reading" large samples of text (on the order of a decade's worth of a daily newspaper).  BEAGLE simulates classic semantic experiments, including typicality judgments, classification to semantic categories, semantic priming, and sentence-completion tasks.  It is currently state-of-the art.   I am requesting a laboratory computer that will allow me to continue work on BEAGLE by extending it in 3 directions:         Translation.   BEAGLE creates a vector (dimensionality ~3000) for each of about 90,000 words; the vector contains the history of that word's use in the text, including the order of words in the sentences.  A word-by-word similarity matrix derived from English should line up with the corresponding matrix from French.  I wish to extend BEAGLE by using its semantics as the basis for automatic translation.         Understanding Memory.   We have known for 50 years background knowledge determines performance in memory tasks.  Because semantics have, heretofore, been intractable, current models of memory try to finesse the issue.  BEAGLE's vectors make possible a knowledge-aware model of memory.  I wish to develop a knowledge-aware model of memory using BEAGLE's vectors to represent real-world knowledge.           Language and thought.   Wife is close to lover  in BEAGLE's English semantic space.  That relation is not necessarily true in all cultures:  In some, wife may be closer to chattel than lover.  Looking at such differences in the whole lexicon across two cultures, BEAGLE's semantic analysis may help us to understand the differences in thought patterns-perhaps even to understand how terrorists think.