Scaling Semantic Parsers with On-the-Fly Ontology Matching

TLDR

Semantic parsing for large, open‑domain question answering, such as with Freebase, is hindered by the need to represent many phrases that do not map cleanly onto a fixed ontology, exemplified by terms like “daughter” or “number of people living in.” We propose a semantic parsing framework that learns to resolve these ontological mismatches. The system is trained on question–answer pairs, employs a probabilistic CCG to generate linguistically motivated logical forms, and incorporates an ontology‑matching module that adapts the logical forms to the target ontology. On two benchmark datasets, the approach achieves state‑of‑the‑art accuracy, including a nine‑point improvement on a recent Freebase QA corpus.

Abstract

We consider the challenge of learning semantic parsers that scale to large, open-domain problems, such as question answering with Freebase. In such settings, the sentences cover a wide variety of topics and include many phrases whose meaning is difficult to represent in a fixed target ontology. For example, even simple phrases such as ‘daughter’ and ‘number of people living in’ cannot be directly represented in Freebase, whose ontology instead encodes facts about gender, parenthood, and population. In this paper, we introduce a new semantic parsing approach that learns to resolve such ontological mismatches. The parser is learned from question-answer pairs, uses a probabilistic CCG to build linguistically motivated logicalform meaning representations, and includes an ontology matching model that adapts the output logical forms for each target ontology. Experiments demonstrate state-of-the-art performance on two benchmark semantic parsing datasets, including a nine point accuracy improvement on a recent Freebase QA corpus.