Parsing Gapping Constructions Based on Grammatical and Semantic Roles

30670

Authors

nan

Abstract

AbstractA gapping construction consists of a coordinated structure where redundant elements are elided from all but one conjuncts. This paper proposes a method of parsing sentences with gapping to recover elided elements. The proposed method is based on constituent trees annotated with grammatical and semantic roles that are useful for identifying elided elements. Our method outperforms the previous method in terms of F-measure and recall.

Keywords

nan

Full Content

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 2747–2752, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 2747 Parsing Gapping Constructions Based on Grammatical and Semantic Roles Yoshihide Kato and Shigeki Matsubara Information & Communications, Nagoya University Furo-cho, Chikusa-ku, Nagoya, 464-8601 Japan yoshihide@icts.nagoya-u.ac.jp Abstract A gapping construction consists of a coordi- nated structure where redundant elements are elided from all but one conjuncts. This paper proposes a method of parsing sentences with gapping to recover elided elements. The pro- posed method is based on constituent trees an- notated with grammatical and semantic roles that are useful for identifying elided elements. Our method outperforms the previous method in terms of F-measure and recall. 1 Introduction A gapping construction consists of a coordinated structure where redundant elements are elided from all but one conjuncts. For example, we can elide the second redundant verb “ate” from the sentence “John ate bread, and Mary ate rice.” We need to recover elided elements to interpret sen- tences with gapping; however, little work has fo- cused on developing such methods. This paper proposes a method of parsing sen- tences with gapping. Our proposed method uses constituent trees annotated with grammatical and semantic tags and a special tag indicating gapped conjuncts. The method parses a sentence to obtain a tag-annotated constituent tree and analyzes gap- ping constructions using the resulting tree when it includes gapped conjuncts. The analysis is based on a sequence alignment algorithm using gram- matical and semantic tags. An experiment shows that our method outperforms the previous method in terms of F-measure and recall. 2 Gapping Construction This section first explains gapping constructions in the Penn Treebank (PTB, Marcus et al., 1993), on which our proposed method is based, and sum- marizes the previous work on analyzing sentences with gapping. 2.1 Gapping Constructions in the PTB A gapping construction consists of a coordinated structure where redundant elements are elided from all but one conjuncts. The constituents re- maining in a gapped conjunct are called remnants. The remnants have a corresponding constituent, called a correlate, in the ungapped conjunct.1 We can obtain the ungapped version of the conjunct by replacing each correlate with its corresponding remnant. In the PTB, the correspondences between rem- nants and correlates are annotated. Figure 1 shows an example of a PTB constituent tree, which in- cludes a gapping construction. The nodes marked with “-” hyphen indices are the correlates, while those marked with “=” equal indices are the rem- nants. A gapped conjunct is flattened, that is, all remnants are children of the conjunct node. The number assigned to a correlate and a remnant in- dicates a correspondence relation. For example, NP-SBJ-1 and NP-SBJ=1 in this tree are a corre- late and a remnant, respectively, and correspond to each other. We can obtain the constituent tree for “the six-month bills will still mature on May 3, 1990” by replacing NP-SBJ-1 with the tree whose root is NP-SBJ=1 and NP-TMP-2 with that whose root is PP-TMP=2. …

Content truncated to 500 words

Labeled Text

AbstractA gapping construction consists of a coordinated structure where redundant elements are elided from all but one conjuncts. This paper proposes a method of parsing sentences with gapping to recover elided elements. The proposed method is based on constituent trees annotated with grammatical and semantic roles that are useful for identifying elided elements. Our method outperforms the previous method in terms of F-measure and recall.

Publication Details

Paper ID:
30670
Year:
2020
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan