About the Project
In recent years, there has been considerable interest in applying machine learning to tackle inferential problems in natural language. These include challenges such as, for example, recovering entailments from natural language text snippets, or constructing explanations and answering multiple-choice questions based on natural language resources. The emphasis here is on chaining together inferential steps in contexts featuring imprecision, uncertainty and ambiguity, perhaps with the aid of background knowledge. At the same time, there is a considerable body of theoretical knowledge concerning the computational complexity of solving logical problems involving specific fragments of natural language. For example, various grammatical constructions can be shown, on proof-theoretic, or even complexity-theoretic grounds, to generate entailments that cannot be retrieved using certain types of simple proof rules.
The aim of this project is to bring greater clarity to recent studies of entailment recognition using neural networks, by bringing to bear insights gained from the theoretical studies. The hypothesis under consideration is that neural networks are unable to replicate all but the simplest varieties of logical inference arising in natural language. We aim thereby to obtain more information on what, exactly, such techniques can and cannot do, and indeed, to what extent these limitations make a difference in respect of to practical tasks involving the processing of natural language.
Why not add a message here
Based on your current searches we recommend the following search filters.
Based on your current search criteria we thought you might be interested in these.