Argumentation Mark-Up: A Proposal 
Jean-Franqois Delannoy 
RES International and University of Ottawa, Ottawa, Canada 
Abstract 
This is a proposal for a an XML mark-up of argumentation. The annotation can be used to help the reader (e.g. by means of 
selective highlighting or diagramming), and for further processing (summarization, critique, use in information retrieval). 
The article proposes a set of markers derived from manual corpus annotation, exemplifies their use, describes a way to 
assign them using surface cues and limited syntax for scoping, and suggests further directions, including an acquisition tool, 
the application of machine learning, and a collaborative DTD definition task. 
1. Introduction 
Computational linguistics has had a long interest in 
discourse representation, but generally on the structure of 
monologue or dialogue rather than on content analysis, 
rhetoric, or argumentation proper. There is an increasing 
amount of work on mark-up, that is, indicating locally in 
the text the linguistic information - rather than in some 
knowledge-representation formalism. Mark-up can be 
designed for the reader, for a browser (in charge of 
displaying the information according to some template), or 
for further processing (a summarizer, a semantic analyzer, 
a reasoning system) - or a combination of these. 
1.1. Reading aids 
Active reading involves a task of text mapping, selecting, 
anchoring and short-term memorization, all the more 
taxing when reading a scarcely-formatted text. Speed 
reading techniques (Judson & Schaill 73) address in fact 
not only speed per se, but also efficient information 
extraction and critiquing. (Note: I am much in favour of 
giving speed reading more recognition -- and I even 
practice it live in a live radio program on literature, to 
quickly select passages.). More people have to read more 
text, and often this cannot be done in integral reading 
mode (cf. the idea, in Carver 90, of "gear shifts" between 
six reading modes). Highlighting helps reading, as seen in 
common experience (also see numerous references in 
Ostler 98). With a rich and appropriate annotation, the 
reader has a better idea of what is important in some given 
respect, and retention is better. Then, it may well be useful 
to 1) develop a powerful reader-oriented annotation 
system, and 2) implement automatic highlighting or 
interactive assisted annotation. 
Those heavy readers can be helped more appropriately by 
summaries, highlighting, hypertext, etc. All try to 
facilitate access and use, by helping selective perception 
and interpretation. One should allow at least multiple 
views of a text. There is a qualitative difference of the 
finesse of understanding if one can manipulate text, or 
access essential information (Cremmins 82, Jewinsky 90), 
possibly an order of magnitude faster. Those points were 
addressed in a May 1999 colloquium organized by this 
author (Delannoy 99). 
1.2. Processing, modularity, and 
interchange 
Corpus annotation has largely addressed text structure and 
categorial tagging, but until very recently has been leaving 
out most linguistic annotation: lexical (word-sense 
disambiguation), syntactic, sentence semantic, passage 
semantics. A mark-up standard would facilitate modular 
processing and interchange among teams working on 
argumentation. 
1.3 Other work on argumentation 
There does not seem to be focused work on argumentation 
annotation, as distinguished from dialogue structures. 
Moulin 95 uses speech acts, but does not discuss 
argumentation structure and validity. 
The Global Document Annotation initiative (GDA), a 
small-team Japanese-German effort to develop a wide set 
of tags; it treats argumentation marginally, with half a 
dozen causality relations: cause, reason, motivation, 
purpose, concession, contrast .... This is rather similar, in 
fact, to Fillmore-inspired verbal cases, as those used in the 
clause-level analysis part of Barker et al. 98b. 
The Text Encoding Initiative (TEI) will soon be 
broadening its scope, so far mostly concerned with text 
structure, but has not circulated concrete proposals yet on 
discourse. 
The encoding schemes DR/and Mate do not consider the 
question. 
18 
2. Argumentation analysis 
I focus on argumentation, which is, in a way, the 
articulation of discourse and content analyses. 
Argumentation analysis is particularly represented in 
critical reading (of news, editorials, political 
communiques, other people's analyses on politics or 
economics, and legal interpretation). 
Argumentation is often not to be treated in a narrow 
predicate-logic framework. It is better account by Toulmin 
et al. 84 (using a combination of warrants and evidence to 
reach a conclusion) rather than, for example, Groarke et al. 
97 (with classical predicate logic, and treating induction as 
a marginal case). In fact, a lot of everyday argumentation 
is based on induction from models left implicit, rather than 
deduction from clearly shared models. One tag, 
<FRAMEOFREF> addresses analogical or newly created 
models used in a belief/construction. 
In sum, the purpose of a semi-automatic tracking and 
analysis of argumentation includes the following: 
synthesizing claims 
clarifying claims in association with the relevant 
evidence and warrants 
listing the actors' position on an issue (pro, contra) 
pinpointing terminological conflicts 
visualizing underlying models of the domain 
singling out non-demonstrative rhetorical effects 
- coherence checking 
comparing alternatives (some argument can be 
shown to be better than others, in absolute or for a 
given knowledge base) 
a critique of argumentative rhetorical devices, 
including fallacies. 
3. The current tag set 
3.1. Introduction 
The current system of tags is inspired by RST and works on reasoning like Toulmin et al. 84 or Walton 96. and from 
manual experiment. 
The four classes of tags are: 
1. Reasoning and rhetoric 
2. Modelling: description of the world; actors 
3. Textual and narrative 
4. Evaluative 
Classes 1, 2 and 4 are the most characteristic of our focus on argument; class 3 serves to accommodate text flow, inasmuch 
as it is pertinent to the development of an argument. 
Class 4 tags are assigned by the user, or, in part, by a semantic analyzer + reasoning system if available. 
Remarks: 
<ID> and <REF> can be added to any tag to handle inter-segment reference. 
<ACTOR> is used to tag an entity, and so plays a directly semantic role, besides its interplay in the discourse analysis. 
Text type: political analysis. Description, evaluation, and scenario. 
3.2. The tag set 
TYPE TAG ELEMENTS DESCRIPTION 
|. REASONING AND RHETORIC 
Reasoning and <ASSN> assertion 
argumentation 
<JUSTIF> type={ fact, goal, analogy, justification 
declaration } 
<EVID> evidence 
<BACKGROUND> background 
<CONCESSION> concession 
<CONTRAST> contrast 
<EX> reality: {present, past, virtual, example 
19 
m 
m 
m TYPE TAG ELEMENTS DESCRIPTION 
projected} 
<C-EX> 
<HYP> 
<REL> 
<CONS> 
<CONDITION> 
<ATT-CONS> 
<CONTRA JUST> 
<PRO> 
<CONTRA> 
<ANN-ASSN> 
<WHATIF> 
<METAPHOR> 
<INTERP> 
Rhetorical 
<REAS_TECHNIQUE> 
<TROPE> 
<RHETQUES> 
<FOCUS> 
Tone <POL> 
2. MODELLING 
Description of a <FRAME OF REF> 
state of affairs 
<SITUATION> 
<BACKGROUND> 
<SCENARIO> 
Actors and 
assertions; 
polemical stance 
<ALTERNATIVE> 
<RECOM> 
<ACTOR> 
<OPP> 
3. TEXTUAL AND NARRATIVE 
Text function <INTRO> 
<DEVT> 
<REPHRASE> 
<CONCL> 
<SUMMARY> 
<QUOTE> 
<REF> 
<LINK> 
<COMMENT> 
Narrative <EVENT> 
structure 
Speech acts, actor 
interaction 
<LATEREVENT> 
<QUES> 
<ANSW> 
<REQUEST> 
type = {modusPonens, 
modusTollens, induction ... } 
type: { stresslmportance, 
stressUrgency } 
tone = { ironic, derisive, afflicted, 
aggressive} 
ref={ nature, etc.} 
mod= { certain,likely,unlikely,imposs 
ible} 
mod= { certain,likely,unlikely,imposs 
ible} 
name, title 
expose = { lie, inconsequence} 
strategy = {counter-facts, bad- 
consequences, bad-outweighs-good, 
untenable-principles, 
badPrecedents } 
BY (element allowed in any tag) 
adeq = {adequate, beside the point} 
counter-example 
hypothesis 
related to 
enounce consequences (of state of 
affairs) 
premiss or condition 
attitude as a natural consequence 
justification for opposition 
position-pro, position-contra 
contradict-without-justification 
announce assertion 
thought experiment 
metaphor 
interpretation of a state of affairs, 
a stance, or a declaration 
reasoning technique 
tropes as commonplaces 
(mere) rhetorical question 
use of focus (e.g. importance of 
an element in the reasoning) 
varieties of polemic tone 
model (description of one's view 
of a state of affairs); its frame of 
reference 
description of the current situation 
background on the situation 
hypothesis on the future 
alternative 
recommendation 
an actor: a person or institution 
saying or doing something 
opposition 
introduction 
development 
rephrasing 
conclusion 
summary point 
quoting an actor 
reference 
(hyper)link 
author's comment 
event with actor, time, type etc. 
later or subsequent event 
author of an utterance, believer of 
an opinion 
question 
anSWfff 
requesting information 
m 
m 
m 
m 
m 
m 
m 
m 
m 
m 
u 
m 
m 
m 
mm 
20 
TYPE TAG ELEMENTS DESCRIPTION 
Reference tags 
Other 
<GIVE-INFO> 
<REBUTTAL> 
<APOLOGY> 
<FLATTERY> 
<SUGGESTION> 
<ID> 
<REF> 
<REFDERE> 
<SOURCE> 
4. EVALUATIVE 
Logical evaluation, <GOOD ARG> 
including fallacies 
<FALLACY> 
Reader's position 
and comments 
<CONTRAD> 
<CONTRAD EXT> 
<COMMONPLACE> 
<AGREE> 
<DISAG> 
<SEARCH FOR EVID > 
<KW> 
<REM> 
to = 113 of referent 
to = ID of referent 
supplying information 
rebuttal of an actor's point by 
another actor 
apology 
flattery 
type = { beggingTheQuestion, 
redHerring, ignoratioElenehi, 
simplisticModel, strawMan, 
AdMajoritatem, adHominem, 
overGeneralization, etc. } 
ref. "de dicto": to another, 
quotable, segment 
ref. "de re": to content of another 
segment 
source of an information or 
argument: a book, etc. 
good (formally, or 
valid) argument 
fallacy 
plausibly 
self-contradiction 
contradiction/opposition with 
another source 
commonplace, logically valid or 
not 
pro 
contz~ 
should search for evidence. For 
the user's own use. 
key word or key expression 
reader's comment 
4. Examples 
4.1, Example: "Chavez rules" The Economis  May 1999, p 34 
This except is the evaluation pan of the article, between the biography and a bibliography. 
Text type: political analysis. Description, evaluation, and scenario. 
Original text 
If they ever doubted it, Venezuela's political elite now know who's boss: President Hugo Chavez. He promised during his 
election campaign to end their hold on power. He has now taken two giant steps toward doing so - and replacing it with his 
own. First, the president cowed Congress into granting him nearly all the powers that he had demanded to enact economic 
and financial legislation by decree. Then, on April 24th, a referendum gave him a huge majority for the creation of a new 
assembly to rewrite the constitution. 
The former paratroop colonel, dismissesd in the early days of his challenge for power as a political neophyte, has now put 
the sophisticates rudely in their place. He has also shown up the weakness of the two big traditional parties. Crushed in the 
presidential election, they are still groping for a strategy to counter Mr Chavez and his alleged dictatorial tendencies, and 
who no sign so far of finding one. 
Still more so \[i.e. a victory\], in its likely effects, was the backing Mr Chavez won for the creation of a new constituent 
assembly. The turnout, admittedly, was only 39%. But, of those who did bother to vote, over 90% backed his proposal. The 
21 
country's 1961 constitution, he claims, has only perpetuated the rule of the elite in a sham democracy. The new assembly is 
to be elected - on non-party lines - in late June, and is meant to have a constitution ready by early January for approval by 
referendum later that month. 
The assembly will no doubt shake up the structures of government- not least, probably, in allowing a president the two 
consecutive terms that Mr Chavez seeks. Whether he can earn and win re-election it is far too early to say. But, for now, bar 
some seriously bad luck or bad bungling, he is firmly in the saddle, and strongly popular. 
Annotated text 
<SITUATION>If they ever doubted it, Venezuela's political elite now know who's boss: President <ACTOR 
NAME=Chavez ID= l>Hugo Chavez. </ACTOR> 
He promised during his election campaign to end their hold on power. 
He has now taken two giant steps toward doing so - 
and replacing it with his own. 
<DETAILS>First, the president cowed Congress into granting him nearly all the powers that he had demanded to enact 
economic and financial legislation by decree. <LATER_EVENT> Then, on April 24th, a <KW>referendum</KW> gave 
him a huge majority 
for the creation of a new assembly to rewrite the constitution. 
</DETAILS> 
</SITUATION> 
<ACTOR REF=Chavez TITLE=former__paratroop_colonel>The former paratroop colonel, </ACTOR> 
<CONTRAST BY=unspecified DOMAIN=opinion on abilities>dismissesd in the early days of his challenge for power as 
a political neophyte, </CONTRAST> 
has now put the sophisticates rudely in their place. 
He has also shown up the weakness of the <ACTOR ID 1> two big traditional parties. 
<CAUSE> Crushed in the presidential election, </CAUSE> 
<REF = "the two big traditional parties"> they are still groping for a strategy to counter Mr Chavez and his alleged 
dictatorial tendencies, 
and who show no sign so far of finding one. 
<INTERP> Still more so \[i.e. a victory\], in its likely effects, was the backing Mr Chavez won for the creation of a new 
constituent assembly. 
<EVID> <CONCESSION> The turnout, admittedly, was only 39%. </CONCESSION> 
<But, of those who did bother to vote, over 90% backed his proposal.</EVID> 
<DECLA BY=Chavez>The country's 1961 constitution, he claims, has only perpetuated the rule oft he elite in a sham 
democracy. </DECLA> 
<SCENARIO> The new assembly is to be elected - on non-party lines - in late June, 
and is meant to have a constitution ready by early January 
for approval by referendum later that month. 
</SCENARIO> 
< SCENARIO > 
The assembly will no doubt shake up the structures of government- 
not least, probably, in allowing a president the two consecutive terms that Mr Chavez seeks. 
Whether he can earn and win re-election it is far too early to say. 
</SCENARIO> 
<SITUATION>Bug for now, <CONDITION MOD =unlikely>bar some seriously bad luck or bad bungling 
</CONDITION>, he is firmly in the saddle, 
and strongly popular. </SITUATION> 
22 
4.2. Example 2: Encyclopaedia Britannica on Herbert Marcuse 
Text type: encyclopedic (with ironic distance) 
Original text 
A Hegelian-Freudian-Marxist, Marcuse was wedded to the ideas of radicalization, vociferous dissent, and 
"resistance to the point of subversion." He believed that Western society was unfree and repressive, that its 
technology had bought the complacency of the masses with material goods, and that it had kept them intellectually 
and spiritually captive. However, although a frank exponent of resistance to the established order, Marcuse did not 
applaud the campus demonstrations. "I still consider the American University an oasis of free speech and real 
critical thinking in the society," he said. "Any student movement should try to protect this citadel... \[but\] try to 
radicalize the departments inside the university." 
Annotated text 
A <POL tone=irony> Hegelian-Freudian-Marxist</POL>, <ACTOR NAME=Marcuse>Marcuse was </ACTOR> 
<POL tone=irony>wedded to the ideas </POL> of radicalization, <POL TONE =irony>vociferous</POL> dissent, 
and <QUO type=Excerpt> author=Marcuse> "resistance to the point of subversion."</QUO> 
He <POL TONE =irony>believed</POL> that Western society was unfree and repressive, that its technology had 
bought the <POL TONE=irony>complacency</POL> of the masses with material goods, and that it had kept them 
intellectually and spiritually captive. 
<CONTRAST>However, 
<CONCESSION> although a frank exponent of resistance to the established order 
</CONCESSION>, 
Marcuse did not applaud the campus demonstrations. 
<EVIDENCE> 
<QUOTE BY=Marcuse>"I still consider the American University an oasis of free speech and real critical thinking 
in the society," he said. 
<RECOM>"Any student movement should try to protect this citadel \[but\] try to radicalize the departments inside 
the university ." 
</RECOM> 
</QUOTE> 
</EVIDENCE> 
</CONTRAST> 
5. Processing 
5.1. Method 
The annotator is to be associated to a summarizer 
(Delannoy et al. 1998). The summarizer is a non- 
knowledge-based selector of passages which combines 
indices of different aspects: keywords assigned 
dynamically, position, word distribution, coherence via 
lexical chains (using WordNet or Roget: Barker et al. 98a) 
and markers, and thesaurus relations. The initial 
granularity is that of sentences, although we are trying to 
refine it with a grammar. In the compilation of the final 
summary, sentences are selected one by one, with a bias to 
sentences containing keywords underrepresented so far. As 
a partly implemented feature, the input is classified in one 
of various text types (neutral, news story, political 
analysis, narrative, etc.), and the user selects one of several 
output types (events, changes, utterances without 
treatment, or details of argumentation) 
The processing for argumentation mark-up (in progress) 
uses a chart parser for segmenting, rather than the shallow 
segmenting in Marcu 97. It targets more than "clause-like" 
units: non-clause circumstantials (adverbials or adjectivals) 
are recognized and kept separate. This is to allows for 
summarization by elimination of minor constituents ("text 
reduction": Greffenstette 98), and also for the treatment of 
eircumstantials in narrow/broad scope as required ( as in: 
"villagers, who holds guns, will be considered outlaws" vs. 
"villagers who hold guns will be considered outlaws"). 
Non-binary branching (3 and more) is allowed, as are 
cross-level relations, e.g. between a clause and a whole 
paragraph, in particular for cases of propositional reference 
("Those points will be discussed etc."), or meta-textual 
reference ("the concept outlined in the previous paragraph 
is now applied to etc."), that is, de dicto as opposed to de 
re. 
The set of cues is inspired by the extensive list in Marcu 
97. There are rules of scoping for the cues into relations 
and their scope over text segments. The construction of the 
tree of rhetorical relations is done by applying rules. 
23 
The user can intervene in the process to rectify a decision 
of the system, or to insert free-style comments or links. 
5.2. Output 
The output can be: 
• the input text, enriched with marked up of 
argumentative structure in the discourse XML (if 
desired, highlighted via a browser) 
• a graph of the argumentation components (e.g. for a 
critical thinking class) 
• a point-form summary of claims made by the author 
and, if applicable, of the actors reported on 
• a highlight of fallacies: bad arguments (with flaws of 
evidence, relevance, cohesion, etc.) 
• answers to standard questions in general or on the type 
of text, as: timeline, list of persons and charges 
mentioned, declarations, predictions, etc. 
• a chart of the positions of various actors pro and 
contra the main points at issue - and of their 
oppositions (e.g. on an article of political analys!s). 
At the time of writing, this is not operational. As intended, 
the selection of the output type is done by an HTML 
interface lauching a CGI which parses the XML and 
translates it to HTML, selectively to reflect the choice of 
viewing made by the user. 
6. Discussion 
6.1. Suitability and scope 
- In our manual experiments, XML-style annotation 
appears to cover all argument structures, in terms of 
adequacy and non-ambiguity. Still, the criteria of 
multiple-user satisfaction, and of the fulfilment of a 
further task, should be examined in the future. 
- the XML representation framework is versatile in 
matters of extensibility, scoping, and embedding; still, 
one will have to test it for possible overlap problems 
- the annotation scheme is inherently multilingual; the 
annotation process requires language-specific lexical 
and syntactic resources 
- the mark-up addresses argumentative texts: those 
reporting collaborative or contradictory discussants 
-(in direct speech, indirect speech, or synthetic 
rephrasing), those where the author gives claims and 
rationales, and those texts meeting both criteria. More 
finely, it may be useful to characterize different 
subtypes of texts: a reader does not look for the same 
information in different types of texts; and not with 
the same cues. This has not been considered as yet. 
6.2. Assumptions, risks, limitations 
There is a risk of running into problems of overlap: 
To address this, we have to : 
- use the tag classification 
- handle specific priorities where applicable 
- for the rest, use one the existing means for handling 
overlap, e.g. as proposed for the Bergen Wittgenstein 
Archive (Sperberg-McQueen & Huitfeldt 98), or by the 
scoping methods of GDA. 
7. Background and Advancement 
We had the idea of HTML-style mark-up of discourse in 
early 1998, before we were even acquainted with XML. 
The original purpose was mostly one of highlighting, for 
applying critical-thinking analyses in a standardizable way. 
Experiments have been done manually by the author on 42 
texts, 10 to 600 words long, consisting of news articles (Le 
Monde, CNN Online), encyclopedia entries (E. Britannica 
online) and on texts containing more in-depth analyses 
(The Economist; editorials in Le Monde; articles of Le 
Monde Diplomatique and the New York Review Of 
Books). 
The analyzer is being prototyped in Perl to be associated to 
the aummarizer in (Delannoy et al. 98). It includes an 
operational parser of XML, and a chart-parser of English, 
in progress. 
The system of tags can probably be improved by more 
manual mark-up and evaluation. The DTD is not yet 
mature; it should be refined and validated on a large 
number of texts, and preferably in a collaborative way.. 
8. Conclusion 
This proposal is, to our knowledge, the first to explicitly 
apply XML mark-up to argumentation mark-up. It claims 
that the mark-up can be useful both for human critiquer 
and for further processing, thus being a multi-purpose, 
intermediate-level representation. 
Future work includes: 
• tracking possible conflicts of scope between tags 
(overlap, vs. simple embedding) 
• a systematic evaluation of the tag set, its 
adjustment, and the definition of a DTD 
• completing the implementation 
• application of machine learning (cf. the AAAI-97 
Spring Symposium on ML for Discourse analysis; 
Hirschman et al. 98; Barker & Szpakowicz 98). 
• implementing an acquisition tool with which a 
user could simply highlight a passage and click 
the appropriate tag's button. 
• exploration of information extraction based on a 
combination of HTML tags, (for structure) and 
discourse XML; comparison with the automatic 
mark-up of online catalogs 
• test with the students class on critical thinking 
24 
• testing onpolylogal discourse 
• designing a critiquing software, and especially a 
recognizer of fallacies 
• gather more data on speed reading or plain text, 
and on reading/speed reading of annotated text 
• last but not least, engaging in a collaborative 
effort towards a common DTD for argumentation 
mark-up, which is an important part of discourse 
mark-up. 

References 
Barker, Ken & Szpakowicz, Stan, "Semi-Automatic 
Recognition of Noun Modifier Relationships", 
Proceedings of COLING-ACL '98. Montr6al, 96-102. 

Barker, Ken, Yllias Chali, Terry Copeck, Stan Matwin, 
Stan Szpakowicz: "The Design of a Confignrahle Text 
Summarization System", Technical Report TR-98-04, 
SITE, U. Ottawa, 1998 \[Barker et al. 98a\] 

Barker, Ken, Sylvain Delisle and Start Szpakowicz (1998) 
"Test-driving TANKA: Evaluating a semi-automatic 
system of text analysis for knowledge acquisition". 
Proc. AI'98, Twelfth Canadian Con./" on Artificial 
Intelligence. Vancouver, May 1998 \[Barker et al. 98b\] 

Carver, Ronald P. Reading Rate. U. Kansas Press, 1990 
Cremmins Edward T. The Art of Abstracting. ISI Press, 
1982 

Delannoy et al. "Flexible Summarization", working paper, 
Worskhop on Intelligent Text Summarization, AAA 98 
Spring Symposium, 1998 

Delannoy (1999): Description du eolloque "L'importance 
des outils de recherche et de traitement de rinformation 
textuelle", 67e congr6s ACFAS, 
http://www.csi.uottawa.ca/delannoy/ACFAS/description.html 

Generalized Document Markup (GDA): presentation at 
http://www.etl.go.jp/etl/nl/GDA, and tag set at 
http://www.etl.go.jp/etl/nl/GDA/tagset.html 

Greffenstette, Gregory: "Producing Intelligent Telegraphic 
Text Reduction to provide an Audio Scanning Service 
for the Blind", Workshop on Intelligent Text 
Summarization, AAA198 Spring Symposium, 1998 

Groarke Leo A., Tindale Christopher A., Fisher Linda. 
Good Reasoning Matters.t A Constructive Approach to 
Critical Thinking. Oxford University Press, (1996) 1997 

Hirschman Lynette, Patricia Robinson, John Burger, Marc 
Vilain, "Automatic Coreference: The Role of Annotating 
Training Data", Wkp. on Applying Machine Learning to 
Discourse Processing, AAAI 98 Spring Symposium, 
1998 

Jewinski Ed & Jewinsky Judi. How to Write an Executive 
Summary. U. of Ottawa Press, 1990. 

Judson Horace & Schaill William S. The Techniques of 
Reading. An Integrated Program for Improved 
Comprehension and Speed. Harcourt Brace Jovanovich, 
(1954) 1973 

Mann, W. & Thompson, S., "Rhetorical structure theory: 
Towards a functional theory of text organization", Text 
8(3), 241-281 

Marcu, Daniel: The Rhetorical Parsing, Summarization, 
and Generation of Natural Language Texts, Ph.D. 
Dissertation, U Toronto, 1997 

Moulin, Bernard: "Discourse spaces: a pragmatic 
interpretation of contexts", ICCS 1995 

Ostler, Tim: "TextLight: Towards an Intelligent Text 
Browser", dissertation for M.Sc. in Cognitive Science 
and Intelligent Systems, U. of Westminster, UK, Jan 
1998. Online at http://www.eogarch.com 

Sperberg-McQueen, C.M. & Huitfeldt C., "Concurrent 
Document Hierarchies in MECS and SGML" (draft of 
communication at ALLCACH '98), 1998. Online at 
http://lingua.arts.kite.hu/allcach98/abst/abst47.htm 
Text Encoding Initiative (TEI) : 
http://www.uie.edu/orgs/tei/ 

Toulmin Stephen, Richard Rieke, Allan Janik: An 
Introduction to Reasoning, MacMillan, 1978 / 1984 

Walton Douglas, Argument Structure. A Pragmatic 
Theory. University of Toronto Press, 1996.
