Faculty of Computer Science

Research Group Theoretical Computer Science

Oberseminar: Heterogene formale Methoden

Abstract:

Negation is both an operation in formal logic and in natural language by which a proposition is replaced by one stating the opposite, as by the addition of “not” or another negation cue. Treating negation in an adequate way is required for cognitive reasoning, which comprises commonsense reasoning and text comprehension. One task of cognitive reasoning is answering questions given by sentences in natural language. In our approach we want to combine techniques from logics and neural networks. Neural Networks however often suffer from the so-called “Clever Hans” problem, which make them infer information from involuntary cues. While there are several hilarious examples in image recognition, those problems apply also to text comprehension and commonsense reasoning. One of the most often used statistical cues is the presence of negation, or more precise, the word “not”, and it was shown that the language model BERT performs around the chance baseline if the data bias by “not” is erased. This issue on the one hand calls for explainable neural networks and on the other hand for a better negation handling. In general treating syntactical negation may cause problems, and this also holds for the cognitive reasoning system CoRg. We aim to erase problematic syntactical negation and replace the corresponding negatus, i.e., its negated event or poperty, with its antonym. This addresses the issue of a biased dataset and it enables our system CoRg to infer useful information. In this talk, we describe an effective procedure to determine the negatus in order to replace it with it inverse and our overall system for cognitive reasoning. We demonstrate the procedure with examples and evaluate it with several benchmarks.