Dealing with arguments in a natural debate can profit from formal representation techniques - in order to facilitate the inspection of their role and interrelations and even reasoning support to determine the state of sets of arguments. An important issue in building such representations is the intended and accurate attachment of newly raised arguments in the context of the previous debate. In this paper, we propose a method for determining likely attachment points for newly raised arguments, based on discourse concepts, such as given and new information in the natural language formulation of arguments. In the long run, the approach is likely to make the incremental building of formal representations easier and it may even lead to an increase of the accuracy of the formal representation in some cases.
Building semi-formal and formal graphical representations of a
natural debate can be supported by tools, such as ARAUCARIA
[
In this paper, we propose a method for determining plausible attachment points for newly raised arguments, based on discourse concepts, such as given and new information in the natural language (NL) formulation of arguments, but also interpreting some typical argumentative roles expressed in NL. The content of such arguments is partially interpreted and maintained in context which yields evidence for the relations among the arguments as well as for their argumentative role - such as indicating whether they can be a support or an attack. We consider this approach a first step towards building a semantic representation of NL arguments, focused on their argumentative role, which aims at more rigour in the representation of arguments so that in the long run more formal reasoning services are enabled. This paper is organized as follows. We first motivate the need for supporting the incremental update of formal representations of a natural debate. Then we outline a method that aims at finding plausible attachment points for a new argument in an argument graph that represents the current state of a debate. We illustrate this idea by a walk-through of a moderately complex example. Finally, we discuss the state of affairs and future prospects. 2
In argumentation frameworks, the proper content of an argument which may have been made originally in NL is not accessible because its content is abstracted away in some atomic proposition p, or even in an argument identifier. Hence, given a new argument p in a debate, the task is to refer it to the appropriate pi in the given state of an argument graph, according to the intention of the person who raised the argument (p → pi for a support or p → ¬pi for an attack). This may be associated with some cognitive load, in particular for argument graphs of increasing size. However, because the content of the arguments is completely abstracted away, there is no formal support possible from the side of a system.
However, if arguments are available in NL form, these
formulations typically contain a number of linguistic clues, even
without world knowledge, on the basis of which relations
between arguments appear plausible. This is because arguments,
though expressed in a concise and abbreviated form [
When a new argument is raised, this is typically done in a
form where old (given) and new information is combined in the
NL formulation. Several linguistic theories (e.g., [
MainC The main claim of an argument, without restrictions MainE The main entity of an argument SubE A subordinate entity of an argument MainA The main assertion of an argument, without main entity Author The witness or expert who made the argument Content The content of the argument made by someone else
Evalu+/- The argument expresses a good or bad assessment Change↑↓ The argument expresses positive or negative change
One argument (portion) is a paraphrase of another one
One argument (portion) is inferable from another one
related to in the present debate. We understand an argument as a
statement made by a participant in a debate, which may have the
usual form of a support or an attack, but may also expand on the
description of an argument already made; it may elaborate the
content details of another argument, or add an explicit
assessment. As shown in [
Some arguments may have a special form which expresses
polarity, such as evaluatives "<x> is good/bad", and assertions
about changes "<y> is in- or decreasing". Asserting evaluatives,
also in combination with assertions about changes, is
consistent with only one role of an argument. For example, if <x> is
claimed to be positive or desirable, and <x> is claimed to be true
or increasing, then an argument "<y> leads to or causes <x>" can
only function as a support for <x>, but never as an attack on it.
While these pieces of information can readily be applied to
suggesting or restricting updates to the underlying argument
graph, actually obtaining this information is rarely easy.
Reference to given information may be indirect, implicit or
encapsulated in some paraphrase, so that formally recognizing relations
may be difficult in some cases. Moreover, work on discourse
parsers, such as [
In order to support the recognition of cases as described above, we define substructures of NL arguments in an abstract form, to be used by an interpretation procedure. In Table 1, the functions needed to access components of arguments, to check Procedure Propose-Attachment-Points (ArgTree,NewArg) Attachments ← empty forall Arg ∈ ArgTree do if Author (NewArg) then Common ← Compare (Arg,Content(NewArg)) else Common ← Compare (Arg,NewArg) endif if Common
then <Add the pair (Arg,Common) to Attachments> endif endfor return <sort Attachments according to size of Common parts>
Argcommon ← empty if Para(New(Arg1), Arg2) then return Arg2 endif forall ArgPart ∈ {MainC(New(Arg1)),MainE(MainC(New(Arg1))),
MainA(MainC(New(Arg1))), SubE(MainC(New(Arg1)))} do If Para(ArgPart,MainC(Arg2))
then <add MainC(Arg2) to Argcommon> endif If Para(ArgPart,MainE(MainC(Arg2)))
then <add MainE(MainC(Arg2)) to Argcommon> endif If Para(ArgPart,MainA(MainC(Arg2)))
then <add MainA(MainC(Arg2)) to Argcommon> endif If Para(ArgPart,SubE(MainC(Arg2)))
then <add SubE(MainC(Arg2)) to Argcommon> endif return Argcommon endfor
Given(NewArg) ← <Reference in Newarg>
New(NewArg) ← <Complement to Reference in NewArg>
If New(NewArg) <is only an assessment>
then <Propose-for-Conflation(Attachment,NewArg)> endif
their substructure and to link them to components of other
arguments are listed. An argument is conceived as a proposition,
typically a state or an event, with entities involved, possibly
with restrictions; it may be within the scope of a mental attitude,
such as an expert opinion. Propositions are built with semantic
case role fillers, abstracting away, for instance, from passive
voice and function verb constructs. To start with, there are two
complementing discourse functions, Given and New. The
distinction between Given and New information supports the
identification of potential links between arguments, including evidence
for where some piece of information has first been introduced.
Both Given and New are structured propositions corresponding
to the entire argument, with annotations indicating the active
parts, in a complementary way. In Given, the portions of an
argument whose content is covered by some previous argument
are marked, that is, this part serves as a reference to express
linking between arguments. New is then what is not covered by
Given. Then some functions are defined which provide access to
an argument's substructure that are potential candidates for
identifying relations between arguments via their components.
Components of an argument include its main claim (MainC),
disregarding restrictions which may be expressed in subordinate
clauses. MainC can be decomposed into its main entity (MainE),
typically an agent, and the assertion ascribed to it (MainA), that
is (MainC without MainE) and other subordinate entities (SubE).
If an argument is embedded in a propositional attitude, that is,
there is one external person referred to to which the argument
content is attributed, the functions Author and Content are used
to pull out the reference to that person and to the embedded
argument. Moreover, arguments may take the specific form of "<x>
is positive/negative" or "<y> increases/decreases". This may
have consequences on the role of arguments related to such
arguments. Functions Evalu+, Evalu-, Change↑ and Change↓ check
for these forms. Finally, there are two functions which are
intended to bridge variations in wording and reference to some
piece of information: Para is a two-place function which yields
true if two assertions or entities used as parameters are
paraphrases of one another, that is, they are semantically equivalent;
this may be verified, for instance, by systems checking for
logical entailment, such as [
The procedure searching for proposed attachment points is
given in pseudo-code in Figure 1. It consists of the main
procedure Propose-Attachment-Points, with a subprocedure Compare.
This yields an ordered list of hypotheses, from which the person
who produced the argument can choose. Once an attachment
point is confirmed, the book keeping procedure Update is called.
As an initialization, the point of debate, which constitutes the
argument tree at the beginning, is completely marked as New.
Then the procedure Propose-Attachment-Ponits is invoked for
each newly raised argument NewArg. NewArg is compared with
the genuine part (New) of all previously raised arguments in
ArgTree, including the point of debate. The restriction to the
New part of previous arguments to be compared is motivated by
the preference to arguments where content has been introduced
(is New) over those where it merely refers to (Given). A
distinction is made as to whether the NewArg expressed a propositional
attitude (Author) or not, to select the proper content for the
comparison. The procedure Compare carries out the comparison, for
the whole argument and for its components MainC, MainE,
MainA, and its SubEs, checking whether any of these parts are
semantically equivalent in some combination. Successful
comparisons are collected and at the end sorted by the degree of
commonality. There may be one, several or no candidates. Once the
intended attachment point is picked by the person who raised the
argument (which may be one of the attachment points proposed
or another one), the procedure Update inserts NewArg in the
argument tree and assigns its components the states of Given or New.
In addition, Update proposes the conflation of two arguments if
the new argument expresses only an assessment. No attempt is
made yet to check consistency of argumentative roles.
1. Every householder should pay tax for the garbage which the
householder throws away.
2. No householder should pay tax for the garbage which the
householder throws away.
3. Paying tax for garbage increases recycling.
4. Recycling more is good.
5. Paying tax for garbage is unfair.
6. Every householder should be charged equally.
7. Every householder who takes benefits does not recycle.
8. Every householder who does not take benefits pays for every
householder who does take benefits.
9. Professor Resicke says that recycling reduces the need for
new garbage dumps.
10. A reduction of the need for new garbage dumps is good.
11. Professor Resicke is not objective.
12. Professor Resicke owns a recycling company.
13. A person who owns a recycling company earns money from
recycling.
14. Supermarkets create garbage.
15. Supermarkets should pay tax.
16. Supermarkets pass the taxes for the garbage to the consumer.
In this section, we introduce the running example used by Wyner
and his co-authors [
When a user raises a new argument, he also specifies the argument to which the new one is related and the category of that relation. Since humans generally tend to be sloppy in their formulations, express pieces of information in limited degrees of explicitness, especially in inference-rich discourse, and may find it hard to precisely identify semantic relations in a given context, we can expect a number of problems associated with user specifications of this kind.
Later, a transformation method has been proposed [
WALKING THROUGH THE EXAMPLE In this section, we illustrate the envisioned effects of our method, exemplified by Wyner's running example. We sketch the incremental building of a new argument graph, geared by the proposals for attachment points at every newly raised argument. The first two assertions, the points of the debate, we treat as a union since one is the negation of the other. The subsequently raised arguments are dealt with as follows:
In this argument, its main entity (Paying tax for garbage) is considered a paraphrase of the claim's main assertion (Should pay tax for the garbage, disregarding modality). Increases recycling then becomes the New part of this argument.
The main entity of this argument (Recycling more) is assessed as a paraphrase of the New part of the previous argument (increases recycling), the only match. Since the argument conforms to the Evalu+ pattern, its embedding in the previous argument is proposed to yield a support of the positive variant of the point of dabate. This conflation of arguments corresponds to the version of the argument graph in Figure 4.
The main entity of this argument (Paying tax for garbage) is a paraphrase of the main assertion of the point of debate, yielding is unfair as the New part. Note that a reference to 3/4 9a 10 9 11 12 5/6 8 13 7
15 14 16a argument (3) is not validated, since Paying tax for garbage is marked as Given in this argument.
The best match for this argument is between its main entity and the main entity of the point of debate. The contrast between unfair and should be charged equally as a reason for a link would require too much inference capabilities.
For this argument, there are two possible attachments: (1) Every householder with the main entity of the debate, and (2) the main assertion does not recycle with the New part of argument (3), as a weakly related paraphrase, at least. Recognizing some sense of the apparently intended rule, "it is unfair, because some householders do not recycle" is out of reach.
This argument has a stronger relation to the previous one (7) via Every householder who does (not) take benefits than to the main point of debate, which is merely via Every householder. Hence, a supporting relation between arguments (7) and (8) is suggested as in Figure 4 rather than these arguments being sister nodes as in Figure 3.
In the content of this embedded argument, the main entity, recycling, gives rise to two potential attachment points the main entity of argument (3) and, to a less direct extent, does not recycle in argument (7). 10. A reduction of the need for new garbage dumps is good.
In relation to the previous argument, this argument is structurally almost identical to argument (4) in connection with argument (3): it matches the New part of argument (9), only adding an evaluation. Similarly as with the previous pair of arguments, their conflation in the representation is proposed, which is even more compact than the version in Figure 4. 11. Professor Resicke is not objective.
Through reference to a previously introduced person, the proposed attachment point prominenty stands out here. ence. In addition, precisely elaborating the access functions to The main entity in this argument, Professor Resicke can components of an argument, such as main entity, are to be done. refer to its previous references in arguments (9) and (11); the The success of the method will largely depend on how the stronger connection between is not objective and owns a predicates Para in Infer can be fleshed out so that a reasonable recycling company is not recognizible. share of references can be established with acceptable effort. 13. A person who owns a recycling company earns money from In the future, we intend to address these issues of formalirecycling. zation, as well as to investigate the application to larger corpora
For this argument, the main entity matches the New part of
of argumentative texts. A useful extension of the method is a
the previous argument. It is therefore proposed to expand on
more fine-grained elaboration
of preferences on attachment
it (Figure 4), rather than to become its sister node (Figure 3)
points, geared by the focus of attention, similar to models of
ordinary discourse. Moreover, distinctions among
modalities
For this assertion, only the embedded entity garbage matches
may yield evidence for preferred structures among arguments.
with components of other arguments; but only in its role of
Finally, knowledge about argumentation structures [
This argument can be attached to the previous one since they share the main entity, Supermarkets. In addition, the point of debate also qualifies as an attachment point, because it shares the main assertion should pay tax with the new argument. The preference among these two, including the relation between arguments (14) and (15) is hard to assess. None of the approaches addressing the running example proposed criteria for adjoining argument (15) in between argument (14) and the point of debate, as the human did. 16. Supermarkets pass the taxes for the garbage to the consumer.
This argument has a stronger relation to the previous argument via its main entity Supermarkets, than to the point of debate by its embedded entity the taxes for the garbage. Since argument (15) is previously established in a place so that argument (14) depends on it, Supermarkets is only accessible for reference in argument (15), where it is New. 6
In this paper, we have described a method that aims at identifying and proposing attachment points for newly raised NL arguments in a given state of an argument graph. The method makes use of some linguistic concepts to select components in the NL formulation of arguments and to interpret their argumentative role, prominently Given and New information. As it has been demonstrated by a small example, the choices made are mostly reasonable, but not always as (possibly) intended by the human. In addition, some relations between arguments have been found to be superior to those chosen by the human in some cases. We believe that our method is not only useful to support navigation for the incremental building of argument graphs; it can also support the categorization and placement of newly raised arguments, and, in the long run, enable logical reasoning services.
Admittedly, the degree of elaboration is still on some kind of anecdotal level so far. The interpretation of some of the predicates used in the formalization need to be fleshed out more precisely. In particular, this comprises the Para predicate, that determines whether two assertions are paraphrases of one another or somehow close to it to justify an argumentative refer