=Paper=
{{Paper
|id=Vol-2377/paper8
|storemode=property
|title=Can Knowledge Graphs and Deep Learning Approaches help in Representing, Detecting and Interpreting Metaphors?
|pdfUrl=https://ceur-ws.org/Vol-2377/paper_8.pdf
|volume=Vol-2377
|authors=Mehwish Alam
|dblpUrl=https://dblp.org/rec/conf/esws/Alam19
}}
==Can Knowledge Graphs and Deep Learning Approaches help in Representing, Detecting and Interpreting Metaphors?==
https://ceur-ws.org/Vol-2377/paper_8.pdf
Can Knowledge Graphs and Deep Learning
Approaches help in Representing, Detecting and
Interpreting Metaphors?
Mehwish Alam
FIZ Karlsruhe - Leibniz Institute for Information Infrastructure, AIFB Institute,
KIT, Karlsruhe, Germany
Abstract. This paper gives an introduction to Conceptual Metaphor
Theory (CMT) as introduced by George Lakoff and discusses the possible
research problems that can open in the context of Knowledge Graphs and
Deep Learning Methods and Metaphors in different mediums.
1 Proposal
Typically when a human mind thinks of a metaphor, the mind tries to map
one concept to the another concept based on their properties or functionality
etc. In Conceptual Metaphor Theory (CMT) [9], George Lakoff discusses that
in the presence of a metaphor there are cross-domain mappings, i.e., a mapping
between a source domain and a target domain. For example, in
Corruption is inf ecting our society.
, the source domain is infection (i.e., a Disease) which is mapped to the target
domain Corruption (i.e., a Criminal Activity).
A published resource is available on-line called MetaNet [1], which defines
a list of such metaphors and each metaphor consists of a source and a tar-
get domain. In case of the above example, it evokes the metaphor Crime is a
disease. Each of these domains are represented as a linguistic frame called as
source frame and target frame respectively. These frames resembles the frames as
introduced in FrameNet [2], however, there are only few exact matches between
the frames in both the resources, meaning that MetaNet contains its own specific
frames. For the running example, the source frame is a Disease and the tar-
get frame is a Criminal Activity, where each of the roles of source frame i.e.,
disease and patient map to the roles in the target frame criminal activtiy
and victim respectively.
Can Knowledge Graphs Capture such kind of Semantics. While thinking
in terms of Knowledge Graphs, can this information about cross-domain map-
ping be represented in the form of a Knowledge Graph? Amnestic Forgery [5,
6] is one of the attempts to integrate the metaphors from MetaNet to the ex-
isting linguistic linked data cloud based on Frame Semantics, Framester [4]. In
72 Mehwish Alam
this resource each metaphor is represented following the theory of Description
& Situation (D&S) [7]. According to this, a metaphor is a description and its
occurrence in the text is a situation. One of the drawbacks of this resource is
that it keeps very general metaphors. There is a need to find a middle ground
between the cross-domain mappings as represented by frames and mappings oc-
curring in the text. In order to find such kind of mappings we need to process
the textual resources rich in metaphors such as poems or corpora specifically
created for metaphors.
One of the solutions is to use previously designed deep learning methods [8]
for distinguishing between metaphoric and literal expressions. Then finally learn-
ing from these metaphoric expressions their specific domains and enrich the
Knowledge Graph with this kind of information. Another solution would be to
create such kind of mappings in the existing Knowledge Graphs such as DBpedia
which contain those domains and are represented based on their literal meanings
but are not connected to the other domains based on their possible metaphoric
relation. This can help in better Identification/interpretation of metaphors or
generation of new metaphors.
Metaphors in Different Mediums Metaphors not only occur in language
but they also occur in different mediums such as visual metaphors (occurring
in images which can be related to political comics, advertisement or art work).
A metaphor can also be expressed in multiple mediums such as text with im-
age or gestures which can be found in videos. The last kind of metaphors are
referred to as multi-modal metaphors [3]. Tensors can help in dealing with multi-
dimensionality in such kind of metaphors. Following these lines many other
tasks come into play such as: (i) Metaphor identification along with their in-
terpretation by combining the information present in different mediums, (ii)
Capturing/Modeling cultural biases, meaning that the metaphor is interpreted
differently based on cultural background.
References
1. E. Dodge, J. Hong, and E. Stickles. Metanet: Deep semantic automatic metaphor
analysis. In Proceedings of the Third Workshop on Metaphor in NLP. Association
for Computational Linguistics, 2015.
2. C. J. Fillmore. Frame semantics and the nature of language. Annals of the New
York Academy of Sciences, 280(1), 1976.
3. C. J. Forceville and Eduardo Urios-Aparisi. Multimodal Metaphor. De Gruyter
Mouton, Berlin, Boston, 2009.
4. A. Gangemi, M. Alam, L. Asprino, V. Presutti, and D. R. Recupero. Framester: A
wide coverage linguistic linked data hub. In EKAW, 2016.
5. A. Gangemi, M. Alam, and V. Presutti. Amnestic forgery: An ontology of conceptual
metaphors. In FOIS, 2018.
6. A. Gangemi, M. Alam, and V. Presutti. Linked metaphors. In In: ISWC 2018
Posters & Demonstrations, Industry and Blue Sky Ideas Tracks with (ISWC 2018),
2018.
Title Suppressed Due to Excessive Length 73
7. A. Gangemi and P. Mika. Understanding the semantic web through descriptions and
situations. In On The Move to Meaningful Internet Systems 2003: CoopIS, DOA,
and ODBASE - OTM Confederated International Conferences, CoopIS, DOA, and
ODBASE, 2003.
8. G. Gao, E. Choi, Y. Choi, and L. Zettlemoyer. Neural metaphor detection in context.
In EMNLP, 2018.
9. G. Lakoff and M. Johnson. Metaphors we Live by. University of Chicago Press,
Chicago, 1980.