=Paper=
{{Paper
|id=Vol-1341/paper11
|storemode=property
|title=An Informatics Perspective on Argumentation Mining
|pdfUrl=https://ceur-ws.org/Vol-1341/paper11.pdf
|volume=Vol-1341
|dblpUrl=https://dblp.org/rec/conf/argnlp/Schneider14
}}
==An Informatics Perspective on Argumentation Mining==
https://ceur-ws.org/Vol-1341/paper11.pdf
An Informatics Perspective on Argumentation Mining
Jodi Schneider⇤
INRIA Sophia Antipolis France
jschneider@pobox.com
Abstract An informatics perspective (i.e. concerned with
supporting human activity) could help us under-
It is time to develop a community research
standing how we will apply argumentation min-
agenda in argumentation mining. I suggest
ing; this should sharpen the definition of the argu-
some questions to drive a joint commu-
mentation mining task(s). Given such an opera-
nity research agenda and then explain how
tionalization, we can then use the standard natural
my research in argumentation, on support
language processing approach: define a corpus of
tools and knowledge representations, ad-
interest, make a gold standard annotation, test al-
vances argumentation mining.
gorithms, iterate...
1 Time for a community research agenda For instance, to operationalize the definition of
argumentation mining (Q1), we need to know:
This year, argumentation mining is receiving sig-
nificant attention. Five different events from April Q1a How do we plan to use the results of argu-
to July 2014 focus on topics such as arguing on the mentation mining?
Web, argumentation theory and natural language Q1b What domain(s) and human tasks are to be
processing, and argumentation mining. A coor- supported?
dinated research agenda could help advance this Q1c What is the appropriate level of granularity
work in a systematic way. of argument structures in a given context?
We have not yet agreed on the most fundamen- Which models of argumentation are most ap-
tal issues: propriate?
Q1 What counts as ‘argumentation’, in the con- This can be challenging because argumentation
text of the argumentation mining task? has a variety of meanings and uses, in fields from
philosophy to rhetoric to law; some of the pur-
Q2 How do we measure the success of an argu- poses for using argumentation are shown in Fig-
mentation mining task? (e.g. corpora & gold ureUnderstanding
1. how we will use the results of
standards) argumentation mining can help address important
questions related to Q2, such as measuring the suc-
“Argumentation mining, is a relatively cess of algorithms and support tools for identify-
new challenge in corpus-based discourse ing arguments. In particular:
analysis that involves automatically iden-
Q2a How accurate does argumentation mining
tifying argumentative structures within a
need to be?
document, e.g., the premises, conclusion,
Q2b In which applications are algorithms for auto-
and argumentation scheme of each argu-
matically extracting argumentation most ap-
ment, as well as argument-subargument
propriate?
and argument-counterargument relation-
ships between pairs of arguments in the
Q2c In which applications are support tools for
document.”1 (Green et al., 2014)
semi-automatically extracting argumentation
more appropriate?
⇤
This work was carried out during the tenure of an In my work I have tried to bring applications of
ERCIM “Alain Bensoussan” Fellowship Programme. The re- argumentation mining to the forefront. My work
search leading to these results has received funding from the
European Union Seventh Framework Programme (FP7/2007- falls into three main areas: supporting human
2013) under grant agreement no 246016. argumentation with computer tools (CSCW), rep-
Figure 1: Argumentation can be used for many purposes.
Download an editable version of this figure from FigShare DOI http://dx.doi.org/10.6084/m9.figshare.1149925
resenting argumentation in ontologies (knowledge In this approach, argumentation mining sup-
representation), and mining arguments from so- ports scalability, by providing automatic or semi-
cial media (information extraction using argumen- automatic identification of the relevant arguments.
tation theory). I have applied this methodology to Wikipedia
information quality debates, which are used to de-
2 Computer-Supported Collaborative
termine whether to delete a given topic from the
Work encyclopedia (Schneider, 2014b). We tested two
Arguing appears throughout human activity, to argumentation models: Walton’s argumentation
support reasoning and decision-making. The ap- schemes (Schneider et al., 2013) and the theory of
plication area determines the particular genres factors/dimensions (Schneider et al., 2012c), and
and subgenres of language that should be inves- our annotated data is available online.2 Whereas
tigated (Q1b). The appropriate level of granular- Walton’s argumentation schemes could have pro-
ity (Lawrence et al., 2014) depends on whether vided support for writing arguments, we instead
we are in a literary work or a law case or a so- chose to use domain-specific decision factors to
cial media discussion (Q1c). The acceptable error filter the overall debate in the prototype support
rate (Q2a) follows from human tolerances, which tool we built. One difference is that Walton’s
we expect to depend on the area; this in turn de- argumentation schemes are at the micro-level—
termines whether we completely automate argu- structuring the premises and conclusions of a
mentation mining (Q2b) or merely provide semi- given argument—whereas decision factors are at
automatic support (Q2c). This is why I emphasize the macro-level, identifying the topics important
looking at application areas to determine which to discuss; this distinction may be relevant for ar-
problems to focus our attention on, for argument gumentation mining (Schneider, 2014a).
mining.
My thesis described a general, informatics ap- 3 Knowledge Representation
proach to supporting argumentation in collabora- Argumentation mining assumes a way to pack-
tive online decision-making (Schneider, 2014b): age arguments so that they can be exchanged and
1. Analyze requirements for argumentation sup- shared. Structured representations of arguments
port in a given situation, context, or commu- allow “evaluating, comparing and identifying the
nity. relationships between arguments” (Rahwan et al.,
2. Consider which argumentation models to 2011). And the knowledge representations most
use; test their suitability, using features such commonly used for the Web are ontologies.
as the appropriate level of granularity and the To investigate the existing ontologies for struc-
tasks to be supported. turing arguments on the social web, we wrote “A
3. Build a prototype support tool, using a model Review of Argumentation for the Social Semantic
of argumentation structures. Web” (Schneider et al., 2012b).
4. Evaluate and iterate. 2
http://purl.org/jsphd
The review compares: 4 Mining from Social Media
• 13 theoretical models for capturing argument
The third strand of our research is in mining argu-
structure (Toulmin, IBIS, Walton, Dung,
ments from social media.
Value-based Arg. Frameworks, Speech
Act Theory, Language/Action Perspective, 4.1 Characteristics of social media
Pragma-dialectic, Metadiscourse, RST, Co-
To identify arguments in social media, we need
herence, and Cognitive Coherence Rela-
to know where to look. The intention of the au-
tions).
thor might be relevant, for instance we can ex-
• Applications of these theoretical models. pect different types of argument in messages, de-
• Ontologies incorporating argumentation (in- pending on whether they are recreation, infor-
cluding AIF, LKIF, IBIS and many others). mation, instruction, discussion, and recommenda-
• 37 collaborative Web-based tools with ar- tion (Schneider et al., 2014b). In (Schneider et
gumentative discussion components (drawn al., 2012a), we suggested that relevant features
from Social Web practice as well as from aca- for argumentation in social media may include the
demic researchers). genre, metadata, properties of users, goals of a
Thus the argumentation community can choose particular dialogue, context and certainty, infor-
from a number of existing approaches for struc- mal and indirect speech, implicit information, sen-
turing argumentation on the Web. timent and subjectivity.
Still, new approaches continue to be suggested.
Peldszus and Stede have suggested a promising 4.2 Information extraction based on
proposal for annotating arguments using Free- argumentation schemes
man’s argumentation macrostructure (Peldszus In a corpus of camera reviews, we examine the
and Stede, 2013). And for biomedical communi- argument that consumers give in reviews, focus-
cations, Clark et al have proposed a micropublica- ing on rationales about camera properties and con-
tions ontology based on Toulmin’s model for pay- sumer values.
as-you-go construction of claim-argument net- In collaboration with Liverpool researchers in-
works from scientific papers (Clark et al., 2014). cluding Adam Wyner (Wyner et al., 2012), we
We are using this ontology—the micropublica- describe the argumentation mining task in con-
tions ontology3 —to model evidence about phar- sumer reviews as an information extraction task,
macokinetic drug interactions (Schneider et al., where we fill slots in a predetermined argumenta-
2014a) in a joint project organized by Richard tion scheme, such as:
Boyce. Consumer Argumentation Scheme:
We have also developed two ontologies related Premise: Camera X has property P.
to argumentation. First, WD, the Wiki Discussion Premise:Property P promotes value V for agent A.
ontology4 (Schneider, 2014b) was alluded to in Conclusion: Agent A should Action1 camera X.
Section 2: WD is used for argumentation support Further details of the information extraction are
for decision-making discussions in ad-hoc online given in (Schneider and Wyner, 2012). In par-
collaboration, applying factors/dimensions theory. ticular, we developed gazetteers for the camera
Second, ORCA is an Ontology of Reasoning, Cer- domain and user domain, and selected appropri-
tainty and Attribution5 (de Waard and Schneider, ate discourse indicators and sentiment terminol-
2012). Based on a taxonomy by de Waard, ORCA ogy. These form part of an NLP pipeline in the
is motivated by scientific argument. ORCA al- General Architecture for Text Engineering frame-
lows distinguishing completely verified facts from work. Resulting annotations can be viewed on a
hypotheses: it records the certainty of knowledge document or searched with a corpus indexing and
(lack of knowledge; hypothetical; dubitative; dox- querying tool, informing an argument analyst who
astic) as well as its basis (reasoning, data, uniden- wishes to construct instances of the consumer ar-
tified) and source (author or other, explicitly or im- gumentation scheme.
plicitly; or none). We have also presented additional argumenta-
3
http://purl.org/mp/
tion schemes that model evaluative expressions in
4
http://purl.org/wd/ reviews, focusing in (Wyner and Schneider, 2012)
5
http://vocab.deri.ie/orca on user models within a context of hotel reviews.
5 Conclusions Jodi Schneider, Brian Davis, and Adam Wyner.
2012a. Dimensions of argumentation in social me-
We have described our work related to argumen- dia. Knowledge Engineering and Knowledge Man-
tation mining, which uses CSCW, knowledge rep- agement, pages 21–25.
resentation, argumentation theory and information Jodi Schneider, Tudor Groza, and Alexandre Passant.
2012b. A review of argumentation for the Social
extraction. As we noted, different approaches Semantic Web. Semantic Web-Interoperability, Us-
are appropriate for identifying and modeling ar- ability, Applicability, 4(2):159–218.
guments in online debates (Schneider, 2014b) ver- Jodi Schneider, Alexandre Passant, and Stefan Decker.
sus scientific papers (Schneider et al., 2014a), so 2012c. Deletion discussions in Wikipedia: Decision
different application areas need to be considered. factors and outcomes. In Proceedings of the Interna-
tional Symposium on Wikis and Open Collaboration,
We hope that our questions about argumentation WikiSym 2012, pages 17:1–17:10.
mining—starting with What counts as ‘argumen- Jodi Schneider, Krystian Samp, Alexandre Passant, and
tation’, in the context of the argumentation mining Stefan Decker. 2013. Arguments about deletion:
task? and How do we measure the success of an How experience improves the acceptability of argu-
argumentation mining task?—drive the commu- ments in ad-hoc online task groups. In Proceedings
of the ACM conference on Computer Supported Co-
nity towards establishing shared tasks. Shared cor- operative Work, CSCW 2013, pages 1069–1080.
pora and well-defined tasks are needed to propel Jodi Schneider, Carol Collins, Lisa Hines, John R
argumentation mining beyond a highly discussed Horn, and Richard Boyce. 2014a. Modeling ar-
area into an agreed upon research challenge. guments in scientific papers. In The 12th ArgDiaP
Conference: From Real Data to Argument Mining.
Jodi Schneider, Serena Villata, and Elena Cabrio.
2014b. Why did they post that argument? Com-
References municative intentions of Web 2.0 arguments. In Ar-
Tim Clark, Paolo N. Ciccarese, and Carole A. Goble. guing on the Web 2.0 at the 8th ISSA Conference on
2014. Micropublications: a semantic model for Argumentation.
claims, evidence, arguments and annotations in Jodi Schneider. 2014a. Automated argumentation
biomedical communications. Journal of Biomedical mining to the rescue? Envisioning argumentation
Semantics, 5(27), July. and decision-making support for debates in open on-
line collaboration communities. In Proceedings of
Anita de Waard and Jodi Schneider. 2012. Formalising
the First Workshop on Argumentation Mining, pages
uncertainty: An Ontology of Reasoning, Certainty
59–63, Baltimore, Maryland, June. Association for
and Attribution (ORCA). In Joint Workshop on Se-
Computational Linguistics.
mantic Technologies Applied to Biomedical Infor-
matics and Individualized Medicine (SATBI+SWIM Jodi Schneider. 2014b. Identifying, Annotating, and
2012) at International Semantic Web Conference Filtering Arguments and Opinions in Open Collab-
(ISWC 2012). oration Systems. Ph.D. dissertation, Digital Enter-
prise Research Institute, National University of Ire-
Nancy Green, Kevin Ashley, Diane Litman, Chris land, Galway. Corpus and supplementary mate-
Reed, and Vern Walker. 2014. Workshop descrip- rial also available online at http://purl.org/
tion, first workshop on argumentation mining at the jsphd.
association for computational linguistics.
Adam Wyner and Jodi Schneider. 2012. Arguing from
John Lawrence, Chris Reed, Colin Allen, Simon McAl- a point of view. In First International Conference
ister, and Andrew Ravenscroft. 2014. Mining argu- on Agreement Technologies.
ments from 19th century philosophical texts using
Adam Wyner, Jodi Schneider, Katie Atkinson, and
topic based modelling. In Proceedings of the First
Trevor Bench-Capon. 2012. Semi-automated ar-
Workshop on Argumentation Mining, pages 79–87.
gumentative analysis of online product reviews. In
Association for Computational Linguistics.
Computational models of argument: Proceedings of
Andreas Peldszus and Manfred Stede. 2013. From ar- COMMA 2012.
gument diagrams to argumentation mining in texts:
a survey. International Journal of Cognitive Infor-
matics and Natural Intelligence (IJCINI), 7(1):1–31.
Iyad Rahwan, Bita Banihashemi, Chris Reed, Douglas
Walton, and Sherief Abdallah. 2011. Represent-
ing and classifying arguments on the Semantic Web.
The Knowledge Engineering Review, 26(04):487–
511.
Jodi Schneider and Adam Wyner. 2012. Identifying
consumers’ arguments in text. In SWAIE 2012: Se-
mantic Web and Information Extraction. In conjunc-
tion with EKAW 2012.