Building a Semantic Parser Overnight

Abstract

How do we build a semantic parser in a new domain starting with zero training examples? We introduce a new methodology for this setting: First, we use a simple grammar to generate logical forms paired with canonical utterances. The logical forms are meant to cover the desired set of compositional operators, and the canonical utterances are meant to capture the meaning of the logical forms (although clumsily). We then use crowdsourcing to paraphrase these canonical utterances into natural utterances. The resulting data is used to train the semantic parser. We further study the role of compositionality in the resulting paraphrases. Finally, we test our methodology on seven domains and show that we can build an adequate semantic parser in just a few hours.

Neural Symbolic Machines: Learning Semantic Parsers on Freebase with Weak Supervision

“…This is reminiscent of the floating parser (Wang et al, 2015;Pasupat and Liang, 2015), where a derivation tree that is not grounded in the input is incrementally constructed.…”

A Corpus and Semantic Parser for Multilingual Natural Language Querying of OpenStreetMap

“…So far, semi-automatic methods of constructing semantic parsers without bootstrapping from MRLs could not reach the accuracy of semantic parsers extracted from manually created MRLs (Wang et al, 2015). The semantic parser we extracted by simple monolingual machine translation (Andreas et al, 2013) achieves a F1 score of 77.3% for answer retrieval on our data.…”

“…Corpora that have been used for training and testing a number of semantic parsers are GEOQUERY (Zelle and Mooney, 1996 Kwiatkowski et al (2013) or Berant et al (2013) for FREE917. Newer research attempts to close the gap between lexical variability and structural complexity (Vlachos and Clark, 2014;Artzi et al, 2015;Pasupat and Liang, 2015), however, answer retrieval accuracy is low if semantic parsers cannot be bootstrapped from a corpus of queries and MRLs (Wang et al, 2015;Pasupat and Liang, 2015).…”

Controlled Natural Language

“…normative or medical texts that are rather controlled by nature but still infringe the boundaries of CNL or the target formalism [3]. It has also been demonstrated that CNL can serve as an efficient and user-friendly interface for Big Data end-point querying [4; 5], or for bootstrapping robust NL interfaces [6]. as well as for tailored multilingual natural language generation from the retrieved data [4].…”