29.3264, Review: Cognitive Science; Computational Linguistics: Veale, Shutova, Klebanov (2017)
The LINGUIST List
linguist at listserv.linguistlist.org
Fri Aug 24 18:14:34 UTC 2018
LINGUIST List: Vol-29-3264. Fri Aug 24 2018. ISSN: 1069 - 4875.
Subject: 29.3264, Review: Cognitive Science; Computational Linguistics: Veale, Shutova, Klebanov (2017)
Moderator: linguist at linguistlist.org (Malgorzata E. Cavar)
Reviews: reviews at linguistlist.org (Helen Aristar-Dry, Robert Coté)
Homepage: https://linguistlist.org
Please support the LL editors and operation with a donation at:
https://funddrive.linguistlist.org/donate/
Editor for this issue: Jeremy Coburn <jecoburn at linguistlist.org>
================================================================
Date: Fri, 24 Aug 2018 14:14:06
From: Pia Sommerauer [pia.sommerauer at live.com]
Subject: Metaphor
Discuss this message:
http://linguistlist.org/pubs/reviews/get-review.cfm?subid=36366859
Book announced at http://linguistlist.org/issues/28/28-4137.html
AUTHOR: Tony Veale
AUTHOR: Ekaterina Shutova
AUTHOR: Beata Beigman Klebanov
TITLE: Metaphor
SUBTITLE: A Computational Perspective
SERIES TITLE: Synthesis Lectures on Human Language Technologies edited by Graeme Hirst
PUBLISHER: Morgan & Claypool Publishers
YEAR: 2017
REVIEWER: Pia Sommerauer, Vrije Universiteit Amsterdam
SUMMARY
Despite their central importance for many domains of language, metaphor and
other kinds of figurative language have, until rather recently, received
comparatively little attention from the computational world. The book ‘A
computational perspective’ summarizes recent developments in Natural Language
Processing (NLP) as well as theoretical accounts from long-standing traditions
in fields like philosophy, psychology and linguistics in order to introduce
metaphor to NLP researchers as well as interested scholars from other fields.
The first chapter, ‘Introducing metaphor’ first and foremost provides the
reader with a feeling for metaphor on the level of language, but also on a
deeper, conceptual level. The idea of conceptual metaphors as mapping between
a source and a target domain is illustrated by well-chosen examples. It is
shown that metaphor and other instances of figurative language are by no means
easy to model in all their aspects and variations and pose highly complex
problems to computational approaches. The chapter ends with situating the
problems within philosophy, psychology and cognitive sciences on the one hand,
where they have received considerable theoretical attention, and computational
linguistics, in particular NLP, on the other hand, which has shied away from
its complexities for a long time. It is argued that despite the challenges
involved, the area is particularly suitable to designing cognitive models that
are “practical and efficient and cognitively plausible” (p. 3).
Chapter 2, ‘Computational approaches to metaphor: Theoretical Foundations’,
aims at providing an overview of the most important theoretical approaches to
metaphor and other phenomena of figurative language. It opens with raising the
questions of how metaphor can be defined, why it is used and what its
underlying mechanisms are. In the course of the chapter, aspects of metaphor
interpretation (the ‘meaning’ of metaphor) are discussed from different
semantic perspectives, before examining closely related phenomena of
figurative speech such as simile and analogy and their relation to metaphor.
One central idea introduced about metaphorical language is that it ideally
functions as a conceptual pact between the participants of a conversation
(Brennan & Clark, 1996), relying on a variety of cultural factors. After
explaining the fundamental assumptions of Conceptual Metaphor Theory (Lakoff &
Johnson 1980) and Conceptual Integration Theory (Fauconnier & Turner, 2008),
the concepts introduced in the chapter are integrated in the final section,
which argues that rather than constituting different phenomena with different
underlying mechanisms, they can be placed on a “continuum of integration” (p.
30) ranging from loose comparisons invited by similes to tightly integrated
conceptual metaphors and blends.
Following these theoretical foundations, Chapter 3, ‘Artificial Intelligence
and Metaphor’, outlines three major ways of approaching metaphor from a
computational modelling perspective. The first approach frames the problem of
metaphor detection and interpretation as a correction-problem in which
non-literal instances should be recognized as ‘violations’ and translated into
literal equivalents. These systems mainly operate with hard-coded semantic
knowledge in the form of selectional preferences or semantic features.
Alternatively, the problem can be approached from the perspective of mapping
structures to find analogies between source and target concepts. Such
analogical approaches crucially depend on theoretical assumptions and design
choices of knowledge representation. The third approach introduced in this
chapter operates on the basis of schemas representing general metaphorical
mappings. These mappings are used to detect instances of deeper, conceptual
metaphors. This third approach can be seen as the closest implementation of
Conceptual Metaphor Theory and, in contrast to the other approaches, makes the
encodings of metaphors explicit. The chapter closes with a reflection on the
aspects shared by all three approaches and potential future work in the
direction of computational models of conceptual integration.
The main purpose of Chapter 4, ‘Metaphor Annotation’, is to give a
comprehensive overview of ways of approaching metaphor annotation and
available annotated data sets. A particular focus is placed on the Metaphor
Identification Procedure (MIP) suggested by the Pragglejaz Group (2007) and
its variants, in particular the MIPVU (Steen et al. 2010) developed in the
course of annotating the VUAmsterdam corpus. This procedure stands out as it
is the first general approach to metaphor identification independent of genre
or text type. Several difficult but crucial aspects of annotation guidelines
that can have significant impact on the result of the annotation procedures
are discussed, such as what extent annotators should rely on external
resources or their own intuition. The chapter closes with a comprehensive
overview and comparison of available data sets.
While Chapter 3 outlines how computational models of metaphor can employ
knowledge to detect and interpret metaphors, Chapter 5, ‘Knowledge Acquisition
and Metaphor’, provides an account of how this knowledge can be acquired.
Structured knowledge bases, as well as the vast amount of naturally occurring
tests found on the Web and complementary approaches exploiting the strengths
and weaknesses of each type of resources, are introduced. Particular attention
is given to ways in which pattern searches inspired by Hearst patterns (Hearst
1992) can be combined to extract knowledge that is rarely expressed explicitly
in a straightforward way but constitutes an important component in figurative
expressions, such as common stereotypes.
Statistical methods, in particular models based on the distribution words have
in large corpora have recently found a wide application in many NLP tasks, one
of which is the analysis of metaphor. Chapter 6, ‘Statistical approaches to
metaphor’, reviews a number of such statistical methods, which have the
advantage of being largely independent of lexical resources, leading to
potentially higher coverage and better portability. Statistical information
can for instance be used to determine selectional preferences but faces
similar challenges when faced with conventionalized metaphors. Several
approaches based on clustering are introduced, in which a space of words based
on linguistic features is partitioned into groups with similar words. The
resulting structure can be exploited to find, for instance, words from
different domains which are metaphorically associated with the same source
domain. The chapter also outlines approaches that build on the idea of topical
structures that are ‘interrupted’ by words from a metaphorically used source
domain and briefly outlines approaches based on vector space models, whose
information about lexico-syntactic contexts is exploited for paraphrasing
approaches and source-domain assignment. The final section summarizes
approaches exploiting information about concreteness by, for instance, finding
pairs of abstract and concrete expressions.
Following these outlines of major computational approaches and methodologies,
Chapter 7, ‘Applications of Metaphor Processing’, introduces several
interesting directions in which automatic tools can benefit from metaphor
processing. First and foremost, it is stressed that metaphors can yield
insights into the way different linguistic communities conceptualize the
world. Furthermore, the ability to use and comprehend metaphors can provide
indications about second language proficiency. These areas of application can
benefit substantially from automatic metaphor processing tools. Another
direction is applying metaphor processing to tools supporting creative tasks.
In particular, metaphor generation tools could provide support for writing
tasks. An example of how automatically retrieved mappings could be used to
create an entire narrative structure is provided.
The concluding chapter highlights the many ways in which metaphor can be
approached and interpreted, depending on a variety of contextual factors. Most
importantly, however, it underlines the importance of metaphorical phenomena
for NLP, as they affect virtually any task. Hence, robust and available and
accessible tools offering metaphor processing is what the field should aim
for.
EVALUATION
The main goal of this book is “to be a comprehensive guide to the major
landmarks in the computational treatment of metaphor” (p. 4) for an audience
ranging from beginning to experienced researchers in the NLP community.
Without doubt, this goal is achieved as the book treats a multitude of recent
approaches from all major perspectives on NLP and artificial intelligence and
contains an introduction to major theoretical assumptions and the role of
metaphor processing and generation in applications. Furthermore, the book
makes a convincing case for the central role metaphor plays in many tasks that
involve semantic interpretations, making clear why metaphor should no longer
play a marginal role in NLP. Beyond this, the book has several other
strengths.
First, the book provides a vast number of truly illustrative examples that not
only exemplify linguistic phenomena and help the reader grasp subtle
differences, but also invite the reader to to think a step further and ‘play’
with metaphorical mappings. The introductory chapter in particular outlines
the most important phenomena in a seemingly effortless way by means of a
number of very clear examples. The subsequent chapters continue to supply a
range of such illustrations (often taken from well-known media discourse),
which might be of particular help for readers less familiar with linguistic
theories.
A second strong point is the comprehensive overview of data sets annotated for
figurative language use provided in Chapter 4. In particular, the table
presented at the end offers a highly useful and well-organized overview or
resources encompassing, among others, information about the language, text
type of the source data, size, inter-annotator agreement and type of
annotations. This is of great help for researchers in the field as well as
related fields looking for resources.
Third, the book employs a number of effective figures that are easy to
understand and help to clarify complex theoretical ideas, such as Conceptual
Integration or or rather complex computational approaches, such as statistical
methods to metaphor processing.
An aspect that could cause potential difficulties for the reader is that in
some instances, the relation to other parts of the book is not entirely clear
or arises rather implicitly. Up to a certain extent, this is not the fault of
the authors, but due to the nature of the field, which has developed its
theoretical assumptions in philosophy, psychology and linguistics, whereas the
major computational models and implementations have been developed within the
rather engineering-driven field of NLP. For rather inexperienced readers, this
might render it somewhat difficult to connect and compare the different
approaches. Starting researchers may find this challenging. Unfortunately, the
rather dense layout does not always support the reader in finding their way
through the book and sectioning is not employed consistently.
In conclusion, the book constitutes a highly useful summary of approaches,
with chapters on theoretical assumptions, artificial intelligence
perspectives, annotation and data sets, knowledge retrieval, statistical
approaches and applications that also lend themselves well to be read
individually and yield valuable explanations as well as practically useful
overviews of data sets and approaches. It makes a convincing case for the
central role of metaphor for a wide variety of semantic phenomena and in
particular, for NLP tasks and encourages further research in a newly
established field.
REFERENCES
Brennan, S.E. and Clark, H.H. 1996. Conceptual pacts and lexical choice in
conversation. Journal of Experimental Psychology: Learning, Memory, and
Cognition 22(6), p. 1482.
Fauconnier, G. and Turner, M. 2008. The way we think: Conceptual blending and
the mind’s hidden complexities. Basic Books.
Group, P. 2007. MIP: A method for identifying metaphorically used words in
discourse. Metaphor and symbol 22(1), pp. 1–39.
Hearst, M.A. 1992. Automatic acquisition of hyponyms from large text corpora.
In: Proceedings of the 14th conference on Computational linguistics-Volume 2.
Proceedings of the 14th conference on Computational linguistics-Volume 2.
Association for Computational Linguistics, pp. 539–545.
Lakoff, G. and Johnson, M. 1980. Metaphors we live by Chicago. Chicago
University
Steen, G. 2010. A Method for Linguistic Metaphor Identification : From MIP to
MIPVU. John Benjamins Publishing Co.
ABOUT THE REVIEWER
Pia Sommerauer is currently a PhD student at the Computational Lexicology and
Terminology Lab Amsterdam at Vrije Universiteit Amsterdam and the Metaphor Lab
Amsterdam. She completed her Master's degree in linguistics with a
specialization in Human Language Technology at Vrije Universiteit Amsterdam.
Her research focuses on the way ambiguity involved in the semantic phenomena
sense and reference is represented in distributional semantic models.
------------------------------------------------------------------------------
***************** LINGUIST List Support *****************
Please support the LL editors and operation with a donation at:
The IU Foundation Crowd Funding site:
https://iufoundation.fundly.com/the-linguist-list
The LINGUIST List FundDrive Page:
https://funddrive.linguistlist.org/donate/
----------------------------------------------------------
LINGUIST List: Vol-29-3264
----------------------------------------------------------
More information about the LINGUIST
mailing list