Contradection - Detecting Contradictions in Text Open Access
Downloadable ContentDownload PDF
Language structures may come in a variety of forms, but are always received sequentially as meaning unfolds to the interpreter. Large bodies of text can create a myriad of relationships and concepts, often making it difficult to detect contradictions between elements that are far apart. This work explores the combination of deterministic and non-deterministic solutions for detecting relationships within syntactic structures, mapping those relationships, and determining if a negation in those relationships is 'decidable'. Using word vectors, one goal is to optimally map object-attribute relationships such that we may apply the Kullback-Leibler divergence to an attribute plane in the vector space. A proper divergence will allow give value to a cosine similarity function of object-attribute pairs whereby negative cosine will suggest a possible contradiction. Different parsing methods are expected to affect mapping and evaluation outcomes. The work aims to compare deterministic parsing, such as based on lexical and grammar structures, along with multinomial logistic regression in language modeling and stochastic semantic analysis for deriving relationship values; and to furthermore experiment using logical deductions to create predicate logic, through which the system may re-deduce to determine a logical coherence level. High coherence levels may be used to isolate confident relationships between objects and attributes to be compared with qualities inferred on those objects throughout the input text.
|Contradetcion - Rami Houssami (updated).pdf||2018-08-27||Open Access||