Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
Krishnamurthy, Jayant and Mitchell, Tom M.

Article Structure

Abstract

We present an approach to training a joint syntactic and semantic parser that combines syntactic training information from CCGbank with semantic training information from a knowledge base via distant supervision.

Introduction

Integrating syntactic parsing with semantics has long been a goal of natural language processing and is expected to improve both syntactic and semantic processing.

Prior Work

This paper combines two lines of prior work: broad coverage syntactic parsing with CCG and semantic parsing.

Parser Design

This section describes the Combinatory Categorial Grammar (CCG) parsing model used by ASP.

Parameter Estimation

This section describes the training procedure for ASP.

Experiments

The experiments below evaluate ASP’s syntactic and semantic parsing ability.

Discussion

We present an approach to training a joint syntactic and semantic parser.

Topics

logical forms

Appears in 65 sentences as: Logical Form (1) Logical form (3) logical form (27) Logical Forms (1) Logical forms (2) logical forms (42)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. The trained parser produces a full syntactic parse of any sentence, while simultaneously producing logical forms for portions of the sentence that have a semantic representation within the parser’s predicate vocabulary.
    Page 1, “Abstract”
  2. A semantic evaluation demonstrates that this parser produces logical forms better than both comparable prior work and a pipelined syntax-then-semantics approach.
    Page 1, “Abstract”
  3. Our parser produces a full syntactic parse of every sentence, and furthermore produces logical forms for portions of the sentence that have a semantic representation within the parser’s predicate vocabulary.
    Page 1, “Introduction”
  4. For example, given a phrase like “my favorite town in California,” our parser will assign a logical form like Ax.CITY(x) /\ LOCATEDIN(:E, CALIFORNIA) to the “town in California” portion.
    Page 1, “Introduction”
  5. ASP produces a full syntactic analysis of every sentence while simultaneously producing logical forms containing any of 61 category and 69 re-
    Page 1, “Introduction”
  6. This line of work has typically used a corpus of sentences with annotated logical forms to train the parser.
    Page 2, “Prior Work”
  7. The input to the parser is a part-of-speech tagged sentence, and the output is a syntactic CCG parse tree, along with zero or more logical forms representing the semantics of subspans of the sentence.
    Page 2, “Parser Design”
  8. These logical forms are constructed using category and relation predicates from a broad coverage knowledge base.
    Page 2, “Parser Design”
  9. The parser uses category and relation predicates from a broad coverage knowledge base both to construct logical forms and to parametrize the parsing model.
    Page 2, “Parser Design”
  10. Each lexicon entry maps a word to a syntactic category, semantic type, and logical form .
    Page 3, “Parser Design”
  11. In addition to the syntactic category, each lexicon entry has a semantic type and a logical form .
    Page 3, “Parser Design”

See all papers in Proc. ACL 2014 that mention logical forms.

See all papers in Proc. ACL that mention logical forms.

Back to top.

syntactic parsing

Appears in 24 sentences as: syntactic parse (7) syntactic parser (1) syntactic parses (1) syntactic parsing (16) syntactically parses (1)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. The trained parser produces a full syntactic parse of any sentence, while simultaneously producing logical forms for portions of the sentence that have a semantic representation within the parser’s predicate vocabulary.
    Page 1, “Abstract”
  2. Integrating syntactic parsing with semantics has long been a goal of natural language processing and is expected to improve both syntactic and semantic processing.
    Page 1, “Introduction”
  3. For example, semantics could help predict the differing prepositional phrase attachments in “I caught the butterfly with the net” and “I caught the butterfly with the spots A joint analysis could also avoid propagating syntactic parsing errors into semantic processing, thereby improving performance.
    Page 1, “Introduction”
  4. ideally improve the parser’s ability to solve difficult syntactic parsing problems, as in the examples above.
    Page 1, “Introduction”
  5. Our parser produces a full syntactic parse of every sentence, and furthermore produces logical forms for portions of the sentence that have a semantic representation within the parser’s predicate vocabulary.
    Page 1, “Introduction”
  6. Additionally, the parser uses predicate and entity type information during parsing to select a syntactic parse .
    Page 1, “Introduction”
  7. Our parser is trained by combining a syntactic parsing task with a distantly-supervised relation extraction task.
    Page 1, “Introduction”
  8. ASP’s syntactic parsing performance is within 2.5% of state-of-the-art; however, we also find that incorporating semantic information reduces syntactic parsing accuracy by ~ 0.5%.
    Page 2, “Introduction”
  9. This paper combines two lines of prior work: broad coverage syntactic parsing with CCG and semantic parsing.
    Page 2, “Prior Work”
  10. Broad coverage syntactic parsing with CCG has produced both resources and successful parsers.
    Page 2, “Prior Work”
  11. The parser presented in this paper can be viewed as a combination of both a broad coverage syntactic parser and a semantic parser trained using distant supervision.
    Page 2, “Prior Work”

See all papers in Proc. ACL 2014 that mention syntactic parsing.

See all papers in Proc. ACL that mention syntactic parsing.

Back to top.

knowledge base

Appears in 20 sentences as: Knowledge Base (1) knowledge base (17) knowledge bases (1) knowledge base’s (1)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. We present an approach to training a joint syntactic and semantic parser that combines syntactic training information from CCGbank with semantic training information from a knowledge base via distant supervision.
    Page 1, “Abstract”
  2. We suggest that a large populated knowledge base should play a key role in syntactic and semantic parsing: in training the parser, in resolving syntactic ambiguities when the trained parser is applied to new text, and in its output semantic representation.
    Page 1, “Introduction”
  3. Using semantic information from the knowledge base at training and test time will
    Page 1, “Introduction”
  4. A semantic representation tied to a knowledge base allows for powerful inference operations — such as identifying the possible entity referents of a noun phrase — that cannot be performed with shallower representations (e.g., frame semantics (Baker et al., 1998) or a direct conversion of syntax to logic (B08, 2005)).
    Page 1, “Introduction”
  5. This paper presents an approach to training a joint syntactic and semantic parser using a large background knowledge base .
    Page 1, “Introduction”
  6. Semantics are learned by training the parser to extract knowledge base relation instances from a corpus of unlabeled sentences, in a distantly-supervised training regime.
    Page 1, “Introduction”
  7. This approach uses the knowledge base to avoid expensive manual labeling of individual sentence semantics.
    Page 1, “Introduction”
  8. However, these approaches to semantics do not ground the text to beliefs in a knowledge base .
    Page 2, “Prior Work”
  9. Finally, some work has looked at applying semantic parsing to answer queries against large knowledge bases , such as YAGO (Yahya et al., 2012) and Freebase (Cai and Yates, 2013b; Cai and Yates, 2013a; Kwiatkowski et al., 2013; Be-rant et al., 2013).
    Page 2, “Prior Work”
  10. These logical forms are constructed using category and relation predicates from a broad coverage knowledge base .
    Page 2, “Parser Design”
  11. 3.1 Knowledge Base
    Page 2, “Parser Design”

See all papers in Proc. ACL 2014 that mention knowledge base.

See all papers in Proc. ACL that mention knowledge base.

Back to top.

semantic parser

Appears in 19 sentences as: semantic parser (9) semantic parsers (1) semantic parses (1) semantic parsing (9)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. We present an approach to training a joint syntactic and semantic parser that combines syntactic training information from CCGbank with semantic training information from a knowledge base via distant supervision.
    Page 1, “Abstract”
  2. We suggest that a large populated knowledge base should play a key role in syntactic and semantic parsing : in training the parser, in resolving syntactic ambiguities when the trained parser is applied to new text, and in its output semantic representation.
    Page 1, “Introduction”
  3. This paper presents an approach to training a joint syntactic and semantic parser using a large background knowledge base.
    Page 1, “Introduction”
  4. We demonstrate our approach by training a joint syntactic and semantic parser , which we call ASP.
    Page 1, “Introduction”
  5. Experiments with ASP demonstrate that jointly analyzing syntax and semantics improves semantic parsing performance over comparable prior work and a pipelined syntax-then-semantics approach.
    Page 2, “Introduction”
  6. This paper combines two lines of prior work: broad coverage syntactic parsing with CCG and semantic parsing .
    Page 2, “Prior Work”
  7. Meanwhile, work on semantic parsing has focused on producing semantic parsers for answering simple natural language questions (Zelle and Mooney, 1996; Ge and Mooney, 2005; Wong and Mooney, 2006; Wong and Mooney, 2007; Lu et al., 2008; Kate and Mooney, 2006; Zettlemoyer and Collins, 2005; Kwiatkowski et al., 2011).
    Page 2, “Prior Work”
  8. Finally, some work has looked at applying semantic parsing to answer queries against large knowledge bases, such as YAGO (Yahya et al., 2012) and Freebase (Cai and Yates, 2013b; Cai and Yates, 2013a; Kwiatkowski et al., 2013; Be-rant et al., 2013).
    Page 2, “Prior Work”
  9. The parser presented in this paper can be viewed as a combination of both a broad coverage syntactic parser and a semantic parser trained using distant supervision.
    Page 2, “Prior Work”
  10. The features are designed to share syntactic information about a word across its distinct semantic realizations in order to transfer syntactic information from CCGbank to semantic parsing .
    Page 4, “Parser Design”
  11. Given these resources, the algorithm described in this section produces parameters 6 for a semantic parser .
    Page 4, “Parameter Estimation”

See all papers in Proc. ACL 2014 that mention semantic parser.

See all papers in Proc. ACL that mention semantic parser.

Back to top.

CCG

Appears in 15 sentences as: CCG (16)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. Syntactic information is provided by CCGbank, a conversion of the Penn Treebank into the CCG formalism (Hockenmaier and Steedman, 2002a).
    Page 1, “Introduction”
  2. This paper combines two lines of prior work: broad coverage syntactic parsing with CCG and semantic parsing.
    Page 2, “Prior Work”
  3. Broad coverage syntactic parsing with CCG has produced both resources and successful parsers.
    Page 2, “Prior Work”
  4. These parsers are trained and evaluated using CCGbank (Hockenmaier and Steedman, 2002a), an automatic conversion of the Penn Treebank into the CCG formalism.
    Page 2, “Prior Work”
  5. Some work has also attempted to automatically derive logical meaning representations directly from syntactic CCG parses (Bos, 2005; Lewis and Steedman, 2013).
    Page 2, “Prior Work”
  6. This section describes the Combinatory Categorial Grammar ( CCG ) parsing model used by ASP.
    Page 2, “Parser Design”
  7. The input to the parser is a part-of-speech tagged sentence, and the output is a syntactic CCG parse tree, along with zero or more logical forms representing the semantics of subspans of the sentence.
    Page 2, “Parser Design”
  8. ASP uses a lexicalized and semantically-typed Combinatory Categorial Grammar ( CCG ) (Steedman, 1996).
    Page 2, “Parser Design”
  9. Most grammatical information in CCG is encoded in a lexicon A, containing entries such as:
    Page 2, “Parser Design”
  10. CCG has two kinds of syntactic categories: atomic and functional.
    Page 3, “Parser Design”
  11. Parsing in CCG combines adjacent categories using a small number of combinators, such as function application:
    Page 3, “Parser Design”

See all papers in Proc. ACL 2014 that mention CCG.

See all papers in Proc. ACL that mention CCG.

Back to top.

relation instances

Appears in 10 sentences as: relation instance (3) relation instances (7)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. Semantics are learned by training the parser to extract knowledge base relation instances from a corpus of unlabeled sentences, in a distantly-supervised training regime.
    Page 1, “Introduction”
  2. A knowledge base K (e.g., NELL), containing relation instances Mel, 62) E K.
    Page 4, “Parameter Estimation”
  3. Distant supervision is provided by the following constraint: every relation instance 7“(€1,€2) E K must be expressed by at least one sentence in 8031,62), the set of sentences that mention both 61 and 62 (Hoffmann et al., 2011).
    Page 5, “Parameter Estimation”
  4. \II is a deterministic OR constraint that checks whether each logical form entails the relation instance Mel, 62), deterministically setting yr 2 1 if any logical form entails the instance and yr 2 0 otherwise.
    Page 5, “Parameter Estimation”
  5. We then extract relation instances from each parse and apply the greedy inference algorithm from Hoffmann et al., (2011) to identify the best set of parses that satisfy the distant supervision constraint.
    Page 6, “Parameter Estimation”
  6. Using Freebase relation instances produces cleaner training data than NELL’s automatically-extracted instances.
    Page 6, “Experiments”
  7. Using the relation instances and Wikipedia sentences, we constructed a data set for distantly-supervised relation extraction.
    Page 6, “Experiments”
  8. Each system was run on this data set, extracting all logical forms from each sentence that entailed at least one category or relation instance .
    Page 8, “Experiments”
  9. Logical forms were marked correct if all category and relation instances entailed by the logical form were expressed by the sentence.
    Page 8, “Experiments”
  10. The extraction metric assigns partial credit by computing the precision and recall of the category and relation instances entailed by the predicted logical form, using those entailed by the annotated logical form as the gold standard.
    Page 9, “Experiments”

See all papers in Proc. ACL 2014 that mention relation instances.

See all papers in Proc. ACL that mention relation instances.

Back to top.

distant supervision

Appears in 7 sentences as: Distant supervision (1) distant supervision (6)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. We present an approach to training a joint syntactic and semantic parser that combines syntactic training information from CCGbank with semantic training information from a knowledge base via distant supervision .
    Page 1, “Abstract”
  2. The parser presented in this paper can be viewed as a combination of both a broad coverage syntactic parser and a semantic parser trained using distant supervision .
    Page 2, “Prior Work”
  3. Distant supervision is provided by the following constraint: every relation instance 7“(€1,€2) E K must be expressed by at least one sentence in 8031,62), the set of sentences that mention both 61 and 62 (Hoffmann et al., 2011).
    Page 5, “Parameter Estimation”
  4. tures of the best set of parses that satisfy the distant supervision constraint.
    Page 6, “Parameter Estimation”
  5. This maximization is intractable due to the coupling between logical forms in E caused by enforcing the distant supervision constraint.
    Page 6, “Parameter Estimation”
  6. We then extract relation instances from each parse and apply the greedy inference algorithm from Hoffmann et al., (2011) to identify the best set of parses that satisfy the distant supervision constraint.
    Page 6, “Parameter Estimation”
  7. The procedure skips any examples with sentences that cannot be parsed (due to beam search failures) or where the distant supervision constraint cannot be satisfied.
    Page 6, “Parameter Estimation”

See all papers in Proc. ACL 2014 that mention distant supervision.

See all papers in Proc. ACL that mention distant supervision.

Back to top.

relation extraction

Appears in 7 sentences as: relation extraction (7)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. Our parser is trained by combining a syntactic parsing task with a distantly-supervised relation extraction task.
    Page 1, “Introduction”
  2. Training is performed by minimizing a joint objective function combining a syntactic parsing task and a distantly-supervised relation extraction task.
    Page 4, “Parameter Estimation”
  3. The syntactic component Osyn is a standard syntactic parsing objective constructed using the syntactic resource L. The semantic component Osem is a distantly-supervised relation extraction task based on the semantic constraint from Krishnamurthy and Mitchell (2012).
    Page 4, “Parameter Estimation”
  4. The semantic objective corresponds to a distantly-supervised relation extraction task that constrains the logical forms produced by the semantic parser.
    Page 5, “Parameter Estimation”
  5. Using the relation instances and Wikipedia sentences, we constructed a data set for distantly-supervised relation extraction .
    Page 6, “Experiments”
  6. Comparing against this parser lets us measure the effect of the relation extraction task on syntactic parsing.
    Page 7, “Experiments”
  7. The parser is trained by jointly optimizing performance on a syntactic parsing task and a distantly-supervised relation extraction task.
    Page 9, “Discussion”

See all papers in Proc. ACL 2014 that mention relation extraction.

See all papers in Proc. ACL that mention relation extraction.

Back to top.

semantic representation

Appears in 7 sentences as: semantic representation (7)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. The trained parser produces a full syntactic parse of any sentence, while simultaneously producing logical forms for portions of the sentence that have a semantic representation within the parser’s predicate vocabulary.
    Page 1, “Abstract”
  2. We demonstrate our approach by training a parser whose semantic representation contains 130 predicates from the NELL ontology.
    Page 1, “Abstract”
  3. We suggest that a large populated knowledge base should play a key role in syntactic and semantic parsing: in training the parser, in resolving syntactic ambiguities when the trained parser is applied to new text, and in its output semantic representation .
    Page 1, “Introduction”
  4. A semantic representation tied to a knowledge base allows for powerful inference operations — such as identifying the possible entity referents of a noun phrase — that cannot be performed with shallower representations (e.g., frame semantics (Baker et al., 1998) or a direct conversion of syntax to logic (B08, 2005)).
    Page 1, “Introduction”
  5. Our parser produces a full syntactic parse of every sentence, and furthermore produces logical forms for portions of the sentence that have a semantic representation within the parser’s predicate vocabulary.
    Page 1, “Introduction”
  6. This synergy gives our parser a richer semantic representation than previous work, while simultaneously enabling broad coverage.
    Page 2, “Prior Work”
  7. Our parser ASP produces a full syntactic parse of any sentence, while simultaneously producing logical forms for sentence spans that have a semantic representation within its predicate vocabulary.
    Page 9, “Discussion”

See all papers in Proc. ACL 2014 that mention semantic representation.

See all papers in Proc. ACL that mention semantic representation.

Back to top.

dependency parse

Appears in 4 sentences as: dependency parse (2) dependency parses (2)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. The grammar for ASP contains the annotated lexicon entries and grammar rules in Sections 02-21 of CCGbank, and additional semantic entries produced using a set of dependency parse heuristics.
    Page 6, “Experiments”
  2. These entries are instantiated using a set of dependency parse patterns, listed in an online appendix.2 These patterns are applied to the training corpus, heuristically identifying verbs, prepositions, and possessives that express relations, and nouns that express categories.
    Page 7, “Experiments”
  3. This approach trains a semantic parser by combining distant semantic supervision with syntactic supervision from dependency parses .
    Page 8, “Experiments”
  4. The best performing variant of this system also uses dependency parses at test time to constrain the interpretation of test sentences — hence, this system also uses a pipelined syntax-then-semantics approach.
    Page 8, “Experiments”

See all papers in Proc. ACL 2014 that mention dependency parse.

See all papers in Proc. ACL that mention dependency parse.

Back to top.

entity mention

Appears in 4 sentences as: entity mention (3) Entity mentions (1) entity mentions (1)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. Entity mentions appear in logical forms via a special mention predicate, M, instead of as database constants.
    Page 4, “Parser Design”
  2. This performance loss appears to be largely due to poor entity mention detection, as we found that not using entity mention lexicon entries at test time improves ASP’s labeled and unlabeled F-scores by 0.3% on Section 00.
    Page 8, “Experiments”
  3. Approximately 50% of errors are caused by marking common nouns as entity mentions (e.g., marking “coin” as a COMPANY).
    Page 8, “Experiments”
  4. Poor entity mention detection is a major source of error in both cases, suggesting that future work should consider integrating entity linking with joint syntactic and semantic parsing.
    Page 9, “Discussion”

See all papers in Proc. ACL 2014 that mention entity mention.

See all papers in Proc. ACL that mention entity mention.

Back to top.

parsing model

Appears in 4 sentences as: parsing model (4)
In Joint Syntactic and Semantic Parsing with Combinatory Categorial Grammar
  1. The parsing model in this paper is loosely based on C&C (Clark and Curran, 2007b; Clark and Curran, 2007a), a discriminative log-linear model for statistical parsing.
    Page 2, “Prior Work”
  2. This section describes the Combinatory Categorial Grammar (CCG) parsing model used by ASP.
    Page 2, “Parser Design”
  3. The parser uses category and relation predicates from a broad coverage knowledge base both to construct logical forms and to parametrize the parsing model .
    Page 2, “Parser Design”
  4. prove comparability, we reimplemented this approach using our parsing model , which has richer features than were used in their paper.
    Page 8, “Experiments”

See all papers in Proc. ACL 2014 that mention parsing model.

See all papers in Proc. ACL that mention parsing model.

Back to top.

PDF available from www.aclweb.org.
Send us an email