Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
Kozareva, Zornitsa and Hovy, Eduard

Article Structure

Abstract

A challenging problem in open information extraction and text mining is the learning of the selectional restrictions of semantic relations.

Introduction

Building and maintaining knowledge-rich resources is of great importance to information extraction, question answering, and textual entailment.

Related Work

A substantial body of work has been done in attempts to harvest bits of semantic information, including: semantic lexicons (Riloff and Shepherd, 1997), concept lists (Lin and Pantel, 2002), isa relations (Hearst, 1992; Etzioni et al., 2005; Pasca, 2004; Kozareva et al., 2008), part-of relations (Girju et al., 2003), and others.

Recursive Patterns

A singly-anchored pattern contains one example of the seed term (the anchor) and one open position for the term to be learned.

Bootstrapping Recursive Patterns

4.1 Problem Formulation

Semantic Relations

So far, we have described the mechanism that learns from one seed and a recursive pattern the selectional restrictions of any semantic relation.

Results

In this section, we evaluate the results of our knowledge harvesting algorithm.

Conclusion

We propose a minimally supervised algorithm that uses only one seed example and a recursive lexico-syntactic pattern to learn in bootstrapping fashion the selectional restrictions of a large class of semantic relations.

Topics

semantic relations

Appears in 31 sentences as: semantic relation (12) semantic relations (20)
In Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
  1. A challenging problem in open information extraction and text mining is the learning of the selectional restrictions of semantic relations .
    Page 1, “Abstract”
  2. We propose a minimally supervised bootstrapping algorithm that uses a single seed and a recursive lexico-syntactic pattern to learn the arguments and the supertypes of a diverse set of semantic relations from the Web.
    Page 1, “Abstract”
  3. We evaluate the performance of our algorithm on multiple semantic relations expressed using “verb”, “noun”, and “verb prep” lexico-syntactic patterns.
    Page 1, “Abstract”
  4. (Pennacchiotti and Pantel, 2006) proposed an algorithm for automatically ontologizing semantic relations into WordNet.
    Page 1, “Introduction”
  5. Given these considerations, we address in this paper the following question: How can the selec-ti0nal restrictions of semantic relations be learned automatically from the Web with minimal eflort using lexico-syntactic recursive patterns?
    Page 2, “Introduction”
  6. 0 A novel representation of semantic relations using recursive lexico-syntactic patterns.
    Page 2, “Introduction”
  7. 0 An automatic procedure to learn the selectional restrictions (arguments and supertypes) of semantic relations from Web data.
    Page 2, “Introduction”
  8. Section 3 addresses the representation of semantic relations using recursive patterns.
    Page 2, “Introduction”
  9. The middle string denotes some (unspecified) semantic relation while the first and third denote the learned arguments of this relation.
    Page 2, “Related Work”
  10. But TextRunner does not seek specific semantic relations , and does not reuse the patterns it harvests with different arguments in order to extend their yields.
    Page 2, “Related Work”
  11. Clearly, it is important to be able to specify both the actual semantic relation sought and use its textual expression(s) in a controlled manner for maximal benefit.
    Page 2, “Related Work”

See all papers in Proc. ACL 2010 that mention semantic relations.

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

Back to top.

recursive

Appears in 25 sentences as: Recursive (1) recursive (24) recursivity (1)
In Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
  1. We propose a minimally supervised bootstrapping algorithm that uses a single seed and a recursive lexico-syntactic pattern to learn the arguments and the supertypes of a diverse set of semantic relations from the Web.
    Page 1, “Abstract”
  2. Given these considerations, we address in this paper the following question: How can the selec-ti0nal restrictions of semantic relations be learned automatically from the Web with minimal eflort using lexico-syntactic recursive patterns?
    Page 2, “Introduction”
  3. 0 A novel representation of semantic relations using recursive lexico-syntactic patterns.
    Page 2, “Introduction”
  4. Section 3 addresses the representation of semantic relations using recursive patterns.
    Page 2, “Introduction”
  5. Learned terms can then be replaced into the seed position automatically, creating a recursive procedure that is reportedly much more accurate and has much higher final yield.
    Page 3, “Recursive Patterns”
  6. No other study has described the use or effect of recursive patterns for different semantic relations.
    Page 3, “Recursive Patterns”
  7. Therefore, going beyond (Kozareva et al., 2008; Hovy et al., 2009), we here introduce recursive patterns other than DAP that use only one seed to harvest the arguments and supertypes of a wide variety of relations.
    Page 3, “Recursive Patterns”
  8. Practically, we can turn any of these patterns into recursive form by giving as input only one of the arguments and leaving the other one as an open slot, allowing the learned arguments to replace the initial seed argument directly.
    Page 3, “Recursive Patterns”
  9. For example, for the relation “fly to”, the following recursive patterns can be built: “* and (seed) fly to *”, “(seed) and >“fly t0 *”, “*fly to (seed) and *”, “*fly to * and (seed)”, “(seed) fly to *” or “*fly to (seed)”, where (seed) is an example like John or Ryanair, and (*) indicates the position on which the arguments are learned.
    Page 3, “Recursive Patterns”
  10. Potentially, one can explore all recursive pattern variations when learning a relation and compare their yield, however this study is beyond the scope of this paper.
    Page 3, “Recursive Patterns”
  11. We are particularly interested in the usage of recursive patterns for the learning of semantic relations not only because it is a novel method, but also because recursive patterns of the DAP fashion are known to: (1) learn concepts with high precision compared to singly-anchored patterns (Kozareva et al., 2008), (2) use only one seed instance for the discovery of new previously unknown terms, and (3) harvest knowledge with minimal supervision.
    Page 3, “Recursive Patterns”

See all papers in Proc. ACL 2010 that mention recursive.

See all papers in Proc. ACL that mention recursive.

Back to top.

WordNet

Appears in 7 sentences as: WordNet (8) WordNet’s (1)
In Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
  1. (Pennacchiotti and Pantel, 2006) proposed an algorithm for automatically ontologizing semantic relations into WordNet .
    Page 1, “Introduction”
  2. However, despite its high precision entries, WordNet’s limited coverage makes it impossible for relations whose arguments are not present in WordNet to be incorporated.
    Page 1, “Introduction”
  3. They mapped each argument of the relation into WordNet and identified the senses for which the relation holds.
    Page 3, “Related Work”
  4. Unfortunately, despite its very high precision entries, WordNet is known to have limited coverage, which makes it impossible for algorithms to map the content of a relation whose arguments are not present in WordNet .
    Page 3, “Related Work”
  5. To surmount this limitation, we do not use WordNet , but employ a different method of obtaining superclasses of a filler term: the inverse doubly-anchored patterns DAP‘1 (Hovy et al., 2009), which, given two arguments, harvests its supertypes from the source corpus.
    Page 3, “Related Work”
  6. (Hovy et al., 2009) show that DAP‘1 is reliable and it enriches WordNet with additional hyponyms and hypernyms.
    Page 3, “Related Work”
  7. Since our problem definition differs from available related work, and WordNet does not contain all harvested arguments as shown in (Hovy et al., 2009), it is not possible to make a direct comparison.
    Page 8, “Results”

See all papers in Proc. ACL 2010 that mention WordNet.

See all papers in Proc. ACL that mention WordNet.

Back to top.

knowledge base

Appears in 4 sentences as: knowledge base (2) knowledge bases (2)
In Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
  1. We also compare our results with existing knowledge base to outline the similarities and differences of the granularity and diversity of the harvested knowledge.
    Page 1, “Abstract”
  2. 0 A comparison of the results with some large existing knowledge bases .
    Page 2, “Introduction”
  3. Initially, we decided to conduct an automatic evaluation comparing our results to knowledge bases that have been extracted in a similar way (i.e., through pattern application over unstructured text).
    Page 6, “Results”
  4. In this section, we compare the performance of our approach with the semantic knowledge base Yago4 that contains 2 million entitiess, 95% of which were manually confirmed to be correct.
    Page 7, “Results”

See all papers in Proc. ACL 2010 that mention knowledge base.

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

Back to top.

part-of-speech

Appears in 4 sentences as: part-of-speech (4)
In Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
  1. We noticed that despite the specific lexico-syntactic structure of the patterns, erroneous information can be acquired due to part-of-speech tagging errors or flawed facts on the Web.
    Page 4, “Bootstrapping Recursive Patterns”
  2. In total, we collected 30GB raw data which was part-of-speech tagged and used for the argument and supertype extraction.
    Page 5, “Semantic Relations”
  3. wrong part-of-speech tag none of the above
    Page 6, “Results”
  4. The majority of the occurred errors are due to part-of-speech tagging.
    Page 6, “Results”

See all papers in Proc. ACL 2010 that mention part-of-speech.

See all papers in Proc. ACL that mention part-of-speech.

Back to top.

part-of-speech tagging

Appears in 4 sentences as: part-of-speech tag (1) part-of-speech tagged (1) part-of-speech tagging (2)
In Learning Arguments and Supertypes of Semantic Relations Using Recursive Patterns
  1. We noticed that despite the specific lexico-syntactic structure of the patterns, erroneous information can be acquired due to part-of-speech tagging errors or flawed facts on the Web.
    Page 4, “Bootstrapping Recursive Patterns”
  2. In total, we collected 30GB raw data which was part-of-speech tagged and used for the argument and supertype extraction.
    Page 5, “Semantic Relations”
  3. wrong part-of-speech tag none of the above
    Page 6, “Results”
  4. The majority of the occurred errors are due to part-of-speech tagging .
    Page 6, “Results”

See all papers in Proc. ACL 2010 that mention part-of-speech tagging.

See all papers in Proc. ACL that mention part-of-speech tagging.

Back to top.