Introducing Word Order Within the LSA Framework
Simon Dennis · 2015
The importance of word order in determining meaning is a hotly debated issue. The success of Latent Semantic Analysis (LSA) across a broad range of tasks that require semantic analysis is a testament to the importance of word choice as compared to word order. Nonetheless at the sentence level, where LSA has been less successful, word order is sometimes the only way in which the role or sense in which a term is being used can be ascertained. Furthermore, a correct interpretation of a sentence often depends on combining words productively into phrasal or clausal units and LSA provides no insight into how this is achieved. In this chapter, we summarize attempts to capture the effects of word order with the Syntagmatic Paradigmatic model (SP, Dennis, in press, 2004). The SP model is an instance-based model that conforms to LSA principles; relying on large corpora and simple mathematical operations rather than formally specified linguistic rules or ad hoc heuristics. Unlike LSA, however, the SP model addresses both the propositional and syntactic levels of analysis. The Importance of Word Order Latent Semantic Analysis (LSA) is a ”bag of words ” technique. A training corpus is represented as a word by document