Pythia: Unsupervised Generation of Ambiguous Textual Claims from Relational Data

Abstract

Applications such as computational fact checking and data-to-text generation exploit the relationship between relational data and natural language text. Despite promising results in these areas, state of the art solutions simply fail in managing "data-ambiguity", i.e., the case when there are multiple interpretations of the relationship between the textual sentence and the relational data. To tackle this problem, we introduce Pythia, a system that, given a relational table 𝐷, generates textual sentences that contain factual ambiguities w.r.t. the data in 𝐷. Such sentences can then be used to train target applications in handling data-ambiguity.In this demonstration, we first show how our system generates data ambiguous sentences for a given table in an unsupervised fashion by data profiling and query generation. We then demonstrate how two existing applications benefit from Pythia's generated sentences, improving the state-of-the-art results. The audience will interact with Pythia by changing input parameters in an interactive fashion, including the upload of their own dataset to see what data ambiguous sentences are generated for it.

Generation of Training Examples for Tabular Natural Language Inference

“…While these works share some ideas with our approach, they cannot consume tables as input. One work focuses on the generation of ambiguous examples by profiling input relations for a new kind of metadata and in terms of example variety they only focus on look-up claims [50,51]. Semantic Parsing.…”