Context Dependent Semantic Parsing: A Survey

Proceedings of the 28th International Conference on Computational Linguistics, pages 2509–2521 Barcelona, Spain (Online), December 8-13, 2020 2509 Context Dependent Semantic Parsing: A Survey Zhuang Li, Lizhen Qu, Gholamreza Haffari Faculty of Information Technology Monash University firstname.lastname@monash.edu Abstract Semantic parsing is the task of translating natural language utterances into machine-readable meaning representations. contextual information (e.g.Currently, most semantic parsing methods are not able to utilize dialogue and comments history), which has a great potential to boost semantic parsing performance. To address this issue, context dependent semantic parsing has recently drawn a lot of attention. In this survey, we investigate progress on the methods for the context dependent semantic parsing, together with the current datasets and tasks. We then point out open problems and challenges for future research in this area. The collected resources for this topic are available at: https://github.com/zhuang-li/ Contextual-Semantic-Parsing-Paper-List. 1 Introduction Semantic parsing is concerned with mapping natural language (NL) utterances into machine-readable structured meaning representations (MRs). These representations are in the formats of formal languages, e.g. Prolog, SQL, and Python. A formal language is typically defined by means of a formal grammar, which consists of a set of rules. Following the convention of the chosen formal language, MRs are also referred to as logical forms or programs. An MR is often executable in a (programming) environment to yield a result (e.g. results of SQL queries) enabling automated reasoning (Kamath and Das, 2018). Most research work on semantic parsing treats each NL utterance as an independent input, ignoring the text surrounding them (Kamath and Das, 2018), such as interaction histories in dialogues. The surrounding text varies significantly across different application scenarios. In a piece of free text, we refer to the surrounding text of a current utterance as its context. The context is different with respect to different utterances. In our sequel, we differentiate between context independent semantic parsing (CISP) and context dependent semantic parsing (CDSP) by whether a corresponding parser utilizes context information. A knowledge base or a database (on which a MR is executed for the purpose of question answering) can be considered as context as well (Krishnamurthy and Mitchell, 2012; Liang, 2016). This type of context does not change with respect to the utterances. In this survey, we only consider the former kind of context which does vary with different utterances. D Database about pets Q1 What are the different pet types? S1 SELECT DISTINCT pettype FROM pets Q2 For each of those, what is the maximum age? S2 SELECT max(pet age), pettype FROM pets GROUP BY pettype Q3 What about the average age? S3 SELECT avg(pet age), pettype FROM pets GROUP BY pettype Table 1: An example of CDSP from SPARC (Yu et al., 2019b), where each SQL query Si is the MR of the question Qi. The utilization of context in semantic parsing imposes both challenges and opportunities. As shown in Table 1, one challenge is to resolve references, such as those in “For each of those, what is the maximum This work is …

Content truncated to 500 words