=Paper=
{{Paper
|id=Vol-1341/paper10
|storemode=property
|title=Encompassing Uncertainty in Argumentation Schemes
|pdfUrl=https://ceur-ws.org/Vol-1341/paper10.pdf
|volume=Vol-1341
|dblpUrl=https://dblp.org/rec/conf/argnlp/BaroniGLT14
}}
==Encompassing Uncertainty in Argumentation Schemes==
https://ceur-ws.org/Vol-1341/paper10.pdf
Encompassing uncertainty in argumentation schemes
Pietro Baroni and Massimiliano Giacomin
Università degli Studi di Brescia
{pietro.baroni, massimiliano.giacomin}@ing.unibs.it
Beishui Liao∗ Leon van der Torre
Zhejiang University University of Luxembourg
baiseliao@zju.edu.cn leon.vandertorre@uni.lu
Abstract When dealing with natural language sources,
one of the challenging problems is to handle the
In existing literature, little attention has
uncertainty of arguments. In fact, natural language
been paid to the problems of how the un-
statements typically include several kinds of un-
certainty reflected by natural language text
certainty. This calls for the need to encompass
(e.g. verbal and linguistic uncertainty) can
uncertainty in the formalisms which are meant to
be explicitly formulated in argumentation
provide a representation of natural arguments, first
schemes, and how argumentation schemes
of all in argumentation schemes, in order to avoid
enriched with various types of uncertainty
that some useful information carried by the text
can be exploited to support argumentation
source is lost in the first modelling step.
mining and evaluation. In this paper, we
To illustrate this problem, let us consider a sim-
focus on the first problem, and introduce
ple example concerning two conflicting natural
some preliminary ideas about how to clas-
language excerpts E1 and E2 , possibly taken from
sify and encompass uncertainty in argu-
some medical publications:
mentation schemes.
E1 : According to [Smith 98], drug X often
1 Introduction causes the side effect Y.
Mining and evaluating arguments from natural E2 : According to recent experimental trials, it
language text (Green et al., 2014) is a relatively is highly likely that drug X does not increase the
new research direction with applications in sev- probability of the side effect Y.
eral areas ranging from legal reasoning (Palau and In order to identify argument structures in these
Moens, 2011) to product evaluation (Wyner et al., texts, one may resort to specific argumentation
2012). Argumentation schemes (Walton et al., schemes. Referring to the classification proposed
2008) are commonly adopted in this context as a in (Walton et al., 2008), E1 can be represented by
first modeling tool: it is assumed that natural ar- an argument A1 which is an instance of the scheme
guments adhere to a set of paradigmatic schemes, Argument from Expert Opinion, while E2 by an
so that these schemes can be used both to drive argument A2 which is an instance of the scheme
the identification of the arguments present in the Argument From Falsification.
text and, after that, to support their formal rep- After A1 and A2 are identified, it may be noted
resentation. As a further step, the assessment of that (though expressed with different linguistic nu-
argument justification status requires to identify ances) their conclusions are in conflict: briefly, A1
the relations among them and to apply a formal leads to the claim that X causes Y, while A2 to the
method, called argumentation semantics to derive claim that X does not cause Y. As a consequence,
the status from these relations. For instance, the a mutual attack relation between A1 and A2 can
well known1 Dung’s theory of abstract argumenta- be identified. Then, the arguments and their at-
tion (Dung, 1995) focuses on the relation of attack tacks can be formalized as an abstract argumen-
between arguments and provides a rich variety of tation framework AF = ({A1 , A2 }, {(A1 , A2 ),
alternative semantics (Baroni et al., 2011) for ar- (A2 , A1 )}) and the status of arguments in AF can
gument evaluation on this basis. be evaluated according to a given argumentation
∗ semantics. For instance, under grounded seman-
Corresponding author
1
Due to space limitations, we assume knowledge of tics, both A1 and A2 are not accepted. It must
Dung’s theory in the following. be noted however that such a modelling approach
(and the relevant outcome in terms of argument 2012; Hunter, 2013a; Hunter, 2014) while, to the
evaluation) overlooks some information which is best of our knowledge, lesser work has been de-
(implicitly or explicitly) carried by the text and voted to encompassing uncertainty in argumenta-
that may lead, in particular, to have one of the ar- tion schemes.
guments prevailing over the other. For instance, as This long-term research goal involves several
considered in (Bex et al., 2013), one may have a basic questions including:
preference relation over argument schemes so that, 1) How the uncertainty reflected by natural lan-
for instance, the scheme Argument From Falsifi- guage text can be explicitly formulated in argu-
cation is preferred to the scheme Argument from mentation schemes?
Expert Opinion. Accordingly, A2 would be pre- 2) How argumentation schemes enriched with
ferred to A1 , and the attack relation would not be various types of uncertainty can be exploited to
mutual, due to the inability of A1 to attack A2 (see support argument mining and evaluation?
the notion of preference-dependent attack in (Bex 3) Which is (are) the most appropriate abstract
et al., 2013)). In this case, we would get a differ- formalism(s) for the evaluation of arguments with
ent argumentation framework AF 0 = ({A1 , A2 }, uncertainty?
{(A2 , A1 )}). Then, under grounded semantics,
A1 is rejected, while A2 is accepted. Texts with Semi-formal Formal
uncertainty argumentation argumentation
However, a static preference relation on the (e.g. verbal / with with
adopted scheme appears too rigid: in most cases linguistic uncertainty uncertainty
uncertainty) B
the preference for an argument over another one
A
@ B A
is not simply based on their structure but, rather, R
@ BN AU
Argument Argument Argument
on their content. To exemplify, in this case, one Mining Formalization Evaluation
may have different opinions on the reliability of
the source [Smith 98], mentioned in E1 , and of
Argumentation
the experimental trials mentioned in E2 . More- schemes with
over, the two excerpts include several terms ex- uncertainty
pressing vagueness and/or uncertainty, like of-
ten, highly likely, the probability of, that may Figure 1: From natural language to argument eval-
be taken into account in the preference ranking uation: a schematic process
of arguments. However, this is not possible in
the approach sketched above, since the argument By focusing on the first question, this paper
schemes adopted in the formalization do not en- presents some preliminary ideas for encompassing
compass these forms of uncertainty and the rele- uncertainty in argumentation schemes.
vant information carried by the text is lost in the The paper is organised as follows. We review
first modelling step. some examples of uncertainty classifications in
Given the pervasiveness of vagueness and un- natural language texts in Section 2 and analyze
certainty in natural language this appears to be the non-uniformity of uncertainty representation
a severe limitation for the use of argumentation in existing argumentation schemes in Section 3.
schemes in argument mining from texts. To over- Then, in Section 4 we exemplify and discuss a pre-
come this problem we envisage the study of ar- liminary approach for encompassing uncertainty
gumentation schemes extended with uncertainty in argumentation schemes. Finally, Section 5 con-
in the context of the process sketched in Figure cludes the paper.
1. Here argumentation schemes with uncertainty
2 Classifying uncertainty types in
are used to extract arguments from texts, keeping
natural language texts
explicit the relevant uncertainties that can then be
used in the step of argument evaluation using suit- In natural language texts different types of uncer-
able abstract formalisms and semantics with un- tainty can be identified. To give a brief account of
certainty. As to the latter step, the study of ex- the richness and complexity of this topic and of the
tensions of Dung’s framework with explicit un- research activities that are being carried out in this
certainty representation is receiving increasing at- area, we quickly recall some examples of uncer-
tention in recent years (Li et al., 2011a; Thimm, tainty classifications considered in the literature.
In the context of scientific discourse, de Waard tinct uncertainty type. On the other hand, linguis-
and Maat (2012) distinguish knowledge evaluation tic uncertainty may be regarded as a generic type
(also called epistemic modality) from knowledge of uncertainty, of which other more specific forms
attribution (also called evidentiality). The former of uncertainty are subtypes. This generic type can
basically concerns the degree of commitment with be associated to those natural language statements
respect to a given statement, while the latter con- to which a more specific uncertainty type can not
cerns the attribution of a piece of knowledge to be applied. For the sake of the preliminary analy-
a source. Accordingly, different kinds of uncer- sis carried out in this paper, we will adopt the latter
tainty can be identified. view.
For instance, according to de Waard and Maat Regan et al. (2002) distinguish between epis-
(2012), sources of knowledge may be distin- temic uncertainty (uncertainty in determinate
guished into the following categories: facts) and linguistic uncertainty (uncertainty in
1) Explicit source of knowledge: the knowledge language) and claims that the latter has received
evaluation can be explicitly owned by the author by far less attention in uncertainty classifications
(‘We therefore conclude that . . . ’) or by a named in the fields of ecology and biology. Linguistic
referent (‘Vijh et al. [28] demonstrated that . . . ’). uncertainty is in turn classified into five distinct
2) Implicit source of knowledge: if there is no types: vagueness, context dependence, ambigu-
explicit source named, knowledge can implicitly ity, indeterminacy of theoretical terms, and un-
still be attributed to the author (‘ these results sug- derspecificity, with vagueness being claimed to be
gest . . . ’) or an external source (‘It is generally the most important for practical purposes. In fact,
believed that . . . ’). all of them refer in some way to the problem that
3) No source of knowledge: the source of some natural language expressions admit alterna-
knowledge can be absent entirely, e.g. in factual tive interpretations. Hence this classification is fo-
statements, such as ‘transcription factors are the cused on a specific form of uncertainty and the use
final common pathway driving differentiation’. of the term linguistic uncertainty here is rather re-
stricted with respect to other works.
Since different sources may have different de-
grees of credibility, this leads to identify a first Taking into account the discussion above, in this
type of uncertainty, namely the (possibly implicit) paper we consider, as a starting point, three uncer-
source uncertainty. tainty types:
As to knowledge evaluation, de Waard and Maat 1) Source uncertainty, denoted in the following
(2012), following Wilbur et al. (2006), distinguish as U1 , concerning the fact that to evaluate the cred-
four levels of certainty in the degree of commit- ibility of different statements one may take into
ment of a subject to a statement: 1) Doxastic (firm account the credibility of their sources;
belief in truth), 2) Dubitative (some doubt about 2) Uncertainty about a statement, denoted as
the truth exists), 3) Hypothetical (the truth value is U2 , arising in situations where a subject making
only proposed), and 4) Lack of knowledge. a statement expresses a partial degree of commit-
This kind of evaluation, called uncertainty ment to the statement itself;
about statements in the following, is typically ex- 3)Linguistic uncertainty or uncertainty inside
pressed through suitable linguistic modifiers. a statement, denoted as U3 , namely uncertainty
Actually linguistic modifiers have a quite generically present in natural language statements,
generic nature and have been the subject of spe- with no further more precise meaning specified.
cific studies by themselves: Clark (1990) pro- For instance in the sentences “According to
vides an extensive review of experimental stud- [Smith 98], Drug X causes headache” and “Ac-
ies concerning the use of linguistic uncertainty ex- cording to recent experimental trials, Drug X
pressions, such as possible, probable, likely, very causes headache”, one may identify U1 since they
likely, highly likely, etc., and their numerical rep- refer the statement “Drug X causes headache” to a
resentation. Linguistic uncertainty is pervasive source (a paper and clinical trials, respectively).
in natural language communication. On the one On the other hand, the sentence “It is likely that
hand, it can be regarded as a form of uncertainty Drug X causes headache” provides an example of
expression (alternative to, e.g., numerical or im- U2 since the statement “Drug X causes headache”
plicit uncertainty expressions) rather than as a dis- is not regarded as certain.
Finally, a sentence like “Drug X sometimes true (false)?
causes severe headache” provides an example of CQ2: Is a an honest (trustworthy, reliable)
U3 . source?
For a more articulated example including sev- CQ3: Did a assert that A is true?
eral uncertainty types, let us consider the follow- In this scheme, no explicit uncertainty is in-
ing text, taken from (Swenson, 2014): “. . . , the cluded, but the critical questions correspond to
Mg inhibition of the actin-activated ATPase activ- several forms of uncertainty that may affect it.
ity observed in class II myosins is likely the re- The second scheme, called Argument from
sult of Mg-dependent alterations in actin binding. Cause to Effect (ACE), is defined as follows:
Overall, our results suggest that Mg reduces the
ADP release rate constant and rate of attachment Major Premise: Generally, if A occurs, then B
to actin in both high and low duty ratio myosins. ” will (might) occur.
Here, some expressions (likely and suggest that) Minor Premise: In this case, A occurs (might
indicate a partial commitment of authors to the occur).
corresponding statements (U2 ), and the knowl- Conclusion: Therefore, in this case, B will
edge source is made explicit by the citation of (might) occur.
(Swenson, 2014) (U1 ). Further, the vague terms CQ1: How strong is the causal generalization?
(high and low) correspond to a form of generic CQ2: Is the evidence cited (if there is any)
linguistic uncertainty inside the relevant statement strong enough to warrant the casual
(U3 ). generalization?
CQ3: Are there other causal factors that could
3 Non-uniformity of uncertainty interfere with the production of the effect in
representation in existing schemes the given case?
In this case, in addition to the implicit uncer-
Given that uncertainty pervades natural language
tainty corresponding to critical questions, explicit
texts and argumentation schemes appear as suit-
expressions of uncertainty are included, namely
able formal tool for argumentation mining from
the modifier Generally and the might specifica-
texts, the question of how to capture uncertainty in
tions in the parentheses.
argumentation schemes naturally arises. This ap-
Clearly the representation of uncertainty in the
pears to be an open research question, as the state-
two schemes is not uniform (since the second
of-the-art formulation of argumentation schemes
scheme encompasses explicit uncertainty in the
(Walton et al., 2008) does not consider uncertainty
premises and the conclusion, while the first does
explicitly, and, more critically, does not seem to
not) but it is not clear whether this non-uniformity
deal with uncertainty in a systematic way, though
is based on some underlying difference between
somehow recognizing its presence. To exemplify
the schemes or is just accidental in the natural
this problem let us compare two argumentation
language formulation of the schemes. Indeed, it
schemes2 from (Walton et al., 2008).
seems possible to reformulate these schemes in a
The first scheme we consider, called Argument
dual manner (adding explicit uncertainty mentions
from Position to Know (APK), is defined as fol-
to the first one, removing them from the second
lows:
one) while not affecting their meaning, as follows:
Major Premise: Source a is in a position to APK with explicit uncertainty:
know about things in a certain subject Major Premise: Source a is (possibly) in a
domain S containing proposition A. position to know about things in a certain
Minor Premise: a asserts that A (in domain S) subject domain S containing proposition A.
is true (false). Minor Premise: a asserts that A (in domain S)
Conclusion: A is true (false). is (might be) true (false).
CQ1: Is a in a position to know whether A is Conclusion: A is (might be) true (false).
2
Recall that an argument scheme basically consists of a ACE without explicit uncertainty:
set of premises, a conclusion defeasibly derivable from the Major Premise: If A occurs, then B will occur.
premises according to the scheme, and a set of critical ques-
tions (CQs) that can be used to challenge arguments built on Minor Premise: In this case, A occurs.
the basis of the scheme. Conclusion: Therefore, in this case, B will
occur. vide here some preliminary examples of point 2,
The above-mentioned non-uniformity suggests using for point 1 the simple classification intro-
that a more systematic treatment of uncertainty in duced in Section 2. In particular we suggest that
argument schemes is needed in order to face the the scheme specification should be accompanied
challenges posed by the representation of natural by an explicit account of the types of uncertainty
language arguments. it may involve, while the use of linguistic un-
certainty expressions in the scheme (like in ACE
Indeed, a recent work (Tang et al., 2013) ad-
above) should be avoided within the natural lan-
dresses the relationships between uncertainty and
guage description of the scheme itself. This ap-
argument schemes in a related but complemen-
proach prevents the non-uniformities pointed out
tary research direction. While the work described
in Section 3 and enforces the adoption of clear
in the present paper aims at enriching argumenta-
modelling choices about uncertainty at the mo-
tion schemes proposed in the literature with ex-
ment of definition of the scheme. In particular, as
plicit uncertainty representation in a systematic
evidenced below, it may point out some ambigui-
way, Tang et al. (2013) introduce several novel ar-
ties in the definition of the scheme itself.
gument schemes concerning reasoning about un-
certainty. This is done using Dempster-Shafer the- In the following examples, we explicitly asso-
ory of evidence in the context of a formalism for ciate uncertainty types with the premises of the
the representation of evidence arguments. Differ- considered schemes (that may affected by them)
ent schemes basically differ in the choice of the and with the critical questions (that point out
rule for (numerical) evidence combination among the potential uncertainty affecting the premises).
the many alternative combination rules available Analysing the uncertainty possibly affecting the
in the literature, and the critical questions in each scheme itself or its applicability (that may also
scheme refer to the applicability conditions of the be expressed by some critical questions) is left to
relevant rule (e.g. Is each piece of evidence in- future work (and requires a richer classification
dependent?). Investigating the possible reuse of of uncertainty types), while, according to point 4
some of the specific ideas presented by Tang et al. above, the uncertainty about the conclusion is re-
(2013) in the context of our broader modelling ap- garded as a derived notion and, for the sake of the
proach is an interesting direction of future work. present analysis, is considered as derived uncer-
tainty, denoted as DU. The syntax we use to as-
4 Encompassing uncertainty in sociate uncertainty types with parts of argument
argumentation schemes schemes is as follows: {. . .}[Ux , . . .], where the
part of the scheme (possibly) affected by uncer-
Devising a systematic approach to encompass
tainty is enclosed in braces and is followed by the
natural language uncertainty in argumentation
relevant uncertainty type(s) enclosed in brackets.
schemes is a long term research goal, posing
many conceptual and technical questions and chal- First, let us consider the APK scheme. Here,
lenges, partly evidenced in the previous sections. the major premise explicitly refers to a source a,
We suggest that such an approach should include so it can be associated with U1 (as evidenced by
the following “ingredients”: the critical questions CQ1 and CQ2). Further one
may consider that the inclusion of proposition A in
1) a classification of uncertainty types;
domain S and the proposition A itself can be spec-
2) a characterization of the uncertainty types
ified with some linguistic uncertainty (U3 ). As to
relevant to each argumentation scheme;
the minor premise, since it refers explicitly to a
3) a formalism for the representation of uncer-
given assertion, it can be associated with uncer-
tainty evaluations (of various types) in actual ar-
tainty about assertions (U2 ). Actually, the critical
guments, i.e. in instances of argument schemes;
question CQ3 refers to the minor premise and its
4) a mechanism to derive an uncertainty evalu- statement “Did a assert that A is true?” is, in fact,
ation for the conclusion of an argument from the ambiguous as far as the type of uncertainty is con-
evaluations concerning the premises and the ap- cerned. On the one hand it might raise a doubt
plied scheme. about the fact that a did actually make any asser-
While each of the items listed above is, by it- tion about A, on the other hand it might raise a
self, a large and open research question, we pro- doubt about the contents of the assertion made by
a. For instance, a might have made a weaker as- ally occur?”, which would turn out very similar
sertion, like “A is probably true”, or a completely in nature to CQ3 in the APK scheme. The fact
different assertion like “A is false”. The three that a question like CQ+ is not considered in this
alternatives mentioned above are rather different scheme, points out a further non-uniformity in the
and involve different uncertainty types. The pos- formulation of argument schemes: one may won-
sibility that a made a weaker assertion is a case der why a sort of explicit confirmation of the mi-
of U2 , while if a made a completely different as- nor premise is required by a critical question in the
sertion (or no assertion at all) about A, the entire APK scheme, while the same kind of confirma-
minor premise is challenged, and this amounts to tion is not required in the ACE scheme. While one
be uncertain about the credibility of the (implicit) might answer that similar questions may have a
source from which we learned that “a asserted that different importance in different schemes, we sug-
A is true”, hence a case of U1 . As this ambiguity gest that a further analysis is needed to address
is present in the current formulation of the scheme, these issues in a systematic way and that a clas-
we leave it unresolved and indicate both types of sification of uncertainty types can be very useful
uncertainty for the minor premise and CQ3. in this respect. To point out this, we add CQ+
This leads to reformulate APK as follows: in the revised version of the ACE scheme, with
Major Premise: {Source a is in a position to the relevant uncertainty type U1 associated with
know about things in a certain subject the minor premise. Finally, CQ3 raises the ques-
domain S}[U1 ] {containing proposition tion about possible other factors interfering with
A}[U3 ]. the causal relation between A and B, i.e. suggests
the presence of possible exceptions in the applica-
Minor Premise: {a asserts that A (in domain S)
tion of the scheme. This kind of uncertainty is not
is true (false)}[U1 , U2 ].
encompassed in our simplistic preliminary classi-
Conclusion: {A is true (false)}[DU].
fication, hence we let it unspecified (denoted as
CQ1: {Is a in a position to know whether A is
[??]), as a pointer to future developments. This
true (false)?}[U1 ]
leads to reformulate ACE as follows:
CQ2: {Is a an honest (trustworthy, reliable)
source?}[U1 ] Major Premise:{If A occurs, then B will occur}
CQ3: {Did a assert that A is true?}[U2 , U1 ]. [U1 , U3 ].
Minor Premise: {In this case, A occurs}
Let us now consider the ACE scheme. Its first
[U1 , U3 ].
premise is a causal generalization, which, as sug-
Conclusion:{Therefore, in this case, B will
gested by the use of (might) in its original formu-
occur} [DU ].
lation, is not always valid. In our simple classifi-
CQ1: {How strong is the causal
cation this can be regarded as a form of linguistic
generalization?}[U3 ]
uncertainty inside the statement (U3 ). This kind
CQ2: {Is the evidence cited (if there is
of uncertainty may also affect the actual formu-
any) strong enough to warrant the casual
lation of the statements A and B in the instantia-
generalization?}[U1 ]
tions of the scheme. The major premise is chal-
CQ+: {Does A actually occur?}[U1 ]
lenged by CQ1 and CQ2. While their interpreta-
CQ3: {Are there other causal factors that
tion allows some overlap, CQ1 seems to concerns
could interfere with the production of the
the “strength” of the causal generalization as it is
effect in the given case?}[??]
formulated, while CQ2 refers to the implicit evi-
dential source of knowledge supporting the causal
5 Conclusions
generalization. Accordingly, CQ1 may be referred
to U3 , while CQ2 to U1 . In recent years, the issue of combining ex-
The minor premise concerns the observation of plicit uncertainty representation and argumenta-
a fact (the occurrence of A), that might involve lin- tion has received increasing attention, with sev-
guistic uncertainty U3 . Indeed, also the observa- eral works dealing in particular with probabilistic
tion of the occurrence of A might have a source, argumentation (Dung and Thang, 2010; Hunter,
so that, in principle, the second premise might 2012; Hunter, 2013b; Li et al., 2011b). These
be affected by U1 , and one might have an addi- works are based on formal argumentation the-
tional critical question CQ+ like “Does A actu- ories like Dung’s abstract argumentation frame-
works (Dung, 1995) or logic-based argumenta- P. M. Dung. 1995. On the acceptability of arguments
tion (Hunter, 2013b). This paper suggests that and its fundamental role in nonmonotonic reasoning,
logic programming, and n-person games. Artificial
these investigations on the formal side should be
Intelligence, 77(2):321–357.
complemented by efforts on the conceptual and
semi-formal side, with particular reference to the N. Green, K. Ashley, D. Litman, C. Reed, and
argumentation schemes model. Argumentation V. Walker, editors. 2014. ACL 2014 Proceedings of
the First Workshop on Argumentation Mining, Balti-
schemes provide a very intuitive semi-formal rep- more, Maryland, USA.
resentation approach for natural arguments and are
indeed adopted in several works as a first level A. Hunter. 2012. Some foundations for probabilistic
abstract argumentation. In Proc. of COMMA 2012,
modelling tool to identify and extract arguments pages 117–128.
from natural language texts. However, as evi-
denced in this paper, argumentation schemes need A. Hunter. 2013a. A probabilistic approach to mod-
elling uncertain logical arguments. International
to be enriched and extended in order to capture
Journal of Approximate Reasoning, 54(1):47–81.
the various kinds of uncertainty typically present
in natural language arguments. The present work A. Hunter. 2013b. A probabilistic approach to mod-
provides a preliminary contribution to this re- elling uncertain logical arguments. International
Journal of Approximate Reasoning, 54(1):47–81.
search line, by pointing out some problems and
providing some simple examples of how they A. Hunter. 2014. Probabilistic qualification of attack
might be tackled. Future work directions are huge in abstract argumentation. International Journal of
Approximate Reasoning, 55(2):607 – 638.
and include an extensive review of the uncertainty
types considered in the literature, with special at- H. Li, N. Oren, and T. J. Norman. 2011a. Probabilistic
tention to works in the area of argumentation min- argumentation frameworks. In Proc. of TAFA 2011,
pages 1–16.
ing, and a systematic analysis of the various ways
argument schemes may be affected by different H. Li, N. Oren, and T. J. Norman. 2011b. Probabilistic
uncertainty types. argumentation frameworks. In Proc. of TAFA 2011,
pages 1–16.
Acknowledgment R. Mochales Palau and M.-F. Moens. 2011. Argumen-
tation mining. Artif. Intell. Law, 19(1):1–22.
The research reported in this paper was partially
supported by the National Science Foundation H. M. Regan, M. Colyvan, and M. A. Burgman. 2002.
of China (No.61175058), and Zhejiang Provin- A taxonomy and treatment of uncertainty for ecol-
ogy and conservation biology. Ecological Applica-
cial Natural Science Foundation of China (No. tions, 12(2):618–628.
LY14F030014).
A. M. Swenson. 2014. Magnesium modulates actin
binding and adp release in myosin motors. J. Biol.
References Chem.
P. Baroni, M. Caminada, and M. Giacomin. 2011. An Y. Tang, N. Oren, S. Parsons, and K. Sycara. 2013.
introduction to argumentation semantics. Knowl- Dempster-shafer argument schemes. In Proc. of
edge Eng. Review, 26(4):365–410. ArgMAS 2013.
M. Thimm. 2012. A probabilistic semantics for ab-
F. Bex, S. Modgil, H. Prakken, and C. Reed. 2013. On
stract argumentation. In Proc. of ECAI 2012, pages
logical specifications of the argument interchange
750–755.
format. J. Log. Comput., 23(5):951–989.
D. Walton, C. Reed, and F. Macagno. 2008. Argumen-
D. A. Clark. 1990. Verbal uncertainty expressions: A tation Schemes. Cambridge University Press.
critical review of two decades of research. Current
Psychology: Research & Reviews, 9(3):203–235. W. J. Wilbur, A. Rzhetsky, and H. Shatkay. 2006. New
directions in biomedical text annotation: definitions,
A. de Waard and H. P. Maat. 2012. Epistemic modal- guidelines and corpus construction. BMC Bioinfor-
ity and knowledge attribution in scientific discourse: matics, 7:356.
A taxonomy of types and overview of features. In
Proc. of ACL 2012, pages 47–55. A. Wyner, J. Schneider, K. Atkinson, and T. J. M.
Bench-Capon. 2012. Semi-automated argumenta-
P. M. Dung and P. M. Thang. 2010. Towards (prob- tive analysis of online product reviews. In Proc. of
abilistic) argumentation for jury-based dispute reso- COMMA 2012, pages 43–50.
lution. In Proc. of COMMA 2010, pages 171–182.