This thesis is an inquiry into the nature of the high-level, rhetorical structure of unrestricted natural language texts, computational means to enable its derivation, and two applications (in automatic summarization and natural language generation) that follow from the ability to build such structures automatically 
The thesis proposes a first-order formalization of the high-level, rhetorical structure of text The formalization assumes that text can be sequenced into elementary units; that discourse relations hold between textual units of various sizes; that some textual units are more important to the writer's purpose than others; and that trees are a good approximation of the abstract structure of text The formalization also introduces a linguistically motivated compositionality criterion, which is shown to hold for the text structures that are valid 
The thesis proposes, analyzes theoretically, and compares empirically four algorithms for determining the valid text structures of a sequence of units among which some rhetorical relations hold Two algorithms apply model-theoretic techniques; the other two apply proof-theoretic techniques 
The formalization and the algorithms mentioned so far correspond to the theoretical facet of the thesis An exploratory corpus analysis of cue phrases provides the means for applying the formalization to unrestricted natural language texts A set of empirically motivated algorithms were designed in order to determine the elementary textual units of a text, to hypothesize rhetorical relations that hold among these units, and eventually, to derive the discourse structure of that text The process that finds the discourse structure of unrestricted natural language texts is called rhetorical parsing 
The thesis explores two possible applications of the text theory that it proposes The first application concerns a discourse-based summarization system, which is shown to significantly outperform both a baseline algorithm and a commercial system An empirical psycholinguistic experiment not only provides an objective evaluation of the summarization system, but also confirms the adequacy of using the text theory proposed here in order to determine the most important units in a text The second application concerns a set of text planning algorithms that can be used by natural language generation systems in order to construct text plans in the cases in which the high-level communicative goal is to map an entire knowledge pool into text

The rhetorical parsing, summarization, and generation of natural language texts

The notion that a text is coherent in virtue of the `relations' that hold between its component spans currently forms the basis for an active research programme in discourse linguistics. Coherence relations feature prominently in many theories of discourse structure, and have recently been used with considerable success in text generation systems. However, while the concept of coherence relations is now common currency for discourse theorists, there remains much confusion about them, and no standard set of relations has yet emerged. The aim of this thesis is to contribute towards the development of a standard set of relations. We begin from an explicitly empirical conception of relations: they are taken to model a collection of psychological mechanisms operative during the tasks of reading and writing. This conception is eshed out with reference to psychological theories of skilled task performance, and to Rosch's notion of the basic level of categorisation. A methodology for investigating these mechanisms is then presented, which takes as its starting point a study of cue phrases|the sentence/clause connectives by which they are signalled. Although it is conventional to investigate psychological mechanisms by studying human behaviour, it is argued here that evidence for the constructs modelled by relations can be sought in an analysis of the linguistic resources available for marking them explicitly in text. The methodology is based on two simple linguistic tests: the test for cue phrases and the test for substitutability. Both tests are functional in inspiration: the former test identi es a heterogenous class of phrases used for linking one portion of text to another; and the latter test is used to discover when a writer is willing to substitute one of these phrases for another. The tests are designed to capture the judgements of ordinary readers and writers, rather than the theoretical intuitions of specialised discourse analysts. The test for cue phrases is used to analyse around 200 pages of naturally occurring text, from which a corpus of over 200 cue phrases is assembled. The substitutability test is then used to organise this corpus into a hierarchical taxonomy, representing the substitutability relationship between every pair of phrases. The taxonomy of cue phrases lends itself neatly to a model of relations as featurebased constructs. Many cue phrases can be interpreted as signalling just some features of relations, rather than whole relations. Small extracts from the taxonomy can be used systematically to determine the alternative values of single features; complex relation de nitions can then be formed by combining the values of many features. The thesis delivers results on two levels. Firstly, it sets out a methodology for motivating a set of relation de nitions, which rests on a systematic analysis of concrete linguistic data, and demands a minimum of theoretical assumptions. Also provided are the relation de nitions which result from applying the methodology. The new de nitions give an interesting picture of the variation that exists amongst cue phrases, and o er a number of innovative insights into text coherence. ii Acknowledgements I would like to take this opportunity to express my deepest thanks to my supervisors, Chris Mellish, Robert Dale and Steve Isard, for their immense help and encouragement over the last four years. Robert rst prompted me to think about the theoretical status of coherence relations; he led me to appreciate the need for a rm foundation for these constructs, as well as overseeing the assembly of the taxonomy of cue phrases. Chris was a huge help in guiding me through the theoretical interpretation of the taxonomy, in all its details. Steve provided a valuable perspective on the thesis as a whole. I would also like to thank my examiners, Donia Scott, Graeme Ritchie and Alex Lascarides, for their detailed comments and suggestions on the manuscript, and for a very interesting discussion during the viva. Many other people have been in uential in the development of the ideas presented here. A big thank you to all of the following, whose feedback and suggestions were much appreciated: Jean Carletta, Ed Hovy, Elisabeth Maier, Kathy McCoy, Johanna Moore, Megan Moser, Jon Oberlander, C ecile Paris, Orin Percus, Ted Sanders, Karen Sparck Jones, Wilbert Spooren, Manfred Stede, and Keith Vander Linden. Special thanks to Ted, for long discussions in several countries, over many glasses of Belgian beer. It seems to be easier to start a PhD than to nish one. This thesis would certainly not have been nished without the concerted e orts of a large and ruthless team of people, and I would like to say thank you to all of them for managing to force me into submission. Firstly to Chris, for the diverse rhetorical strategies he has employed to encourage me to nish, ranging from rational arguments and encouragements, through thinly veiled warnings, to direct threats and emotive appeals; and nally for eliciting from me a solemn oath to stop my research and start writing. Next, to the Engineering and Physical Sciences Research Council (EPSRC), for their generous support over three years; and also for their Faustian `four year deadline', which I have watched approaching with increasing horror but which has nally|I think|had its intended e ect on me. And lastly, to the assorted PhD students and denizens of the AI department, who have made my life on E oor most pleasant. Thanks are due to Alberto, Chris, Ed, Elena, Ian, Ion, Janet, Jean, Jeremy, Kim, Nam Seog, Ste an, and Wamberto, who have all been a great in uence on me during the past few years. Needless to say, any defects that remain in my character are solely my own responsibility. I would like to give special thanks to my family. Their constant support, particularly during the past year when I was running on empty, has been very important to me. Thank you Mum, Dad, Beck, Har, Jule. Thanks also to my wider family, in Britain, in New Zealand and in Tonga. Above all I would like to thank Mele, who has had to put up with a lot of ranting and raving on the subject of discourse structure just recently. Thank you for your love, and cheerfulness, and steadfastness. And especially for the notice pinned above my bed, reminding me of the important fact that No-One Reads PhD Theses. iii Declaration I hereby declare that I composed this thesis entirely myself and that it describes my own research. Alistair Knott Edinburgh December 3, 1996 iv

A Data-Driven Methodology for Motivating a Set of Coherence Relations

Coherent texts are not just simple sequences of clauses and sentences, but rather complex artifacts that have highly elaborate rhetorical structure. This paper explores the extent to which well-formed rhetorical structures can be automatically derived by means of surface-form-based algorithms. These algorithms identify discourse usages of cue phrases and break sentences into clauses, hypothesize rhetorical relations that hold among textual units, and produce valid rhetorical structure trees for unrestricted natural language texts. The algorithms are empirically grounded in a corpus analysis of cue phrases and rely on a first-order formalization of rhetorical structure trees.The algorithms are evaluated both intrinsically and extrinsically. The intrinsic evaluation assesses the resemblance between automatically and manually constructed rhetorical structure trees. The extrinsic evaluation shows that automatically derived rhetorical structures can be successfully exploited in the context of text summarization.

http://www.cs.toronto.edu/pub/gh/Marcu-2000c.pdf

The rhetorical parsing of unrestricted texts: a surface-based approach

Rhetorical Structure Theory (RST) is a theory of text organization that has led to areas of application beyond discourse analysis and text generation, its original goals. In this article, we review...

/pdf/applications-of-rhetorical-structure-theory-38yygvy9eq.pdf

Applications of Rhetorical Structure Theory

Discourse markers as signals (or not) of rhetorical relations

In order to generate cohesive discourse, many of the relations holding between text segments need to be signalled to the reader by means of cue words, or {\em discourse markers}. Programs usually do this in a simplistic way, e.g., by using one marker per relation. In reality, however, language offers a very wide range of markers from which informed choices should be made. In order to account for the variety and to identify the parameters governing the choices, detailled linguistic analyses are necessary. We worked with one area of discourse relations, the Concession family, identified its underlying pragmatics and semantics, and undertook extensive corpus studies to examine the range of markers used in both English and German. On the basis of an initial classification of these markers, we propose a generation model for producing bilingual text that can incorporate marker choice into its overall decision framework.

Ma(r)king concessions in English and German

Natural language generation aims at automatically verbalizing a “deep” representation of content, so that coherent and cohesive text originates. To produce such cohesive discourse, it is important to signal many of the relations holding between text segments to the reader by means of cue words, which we call discourse markers. Current generation systems usually do this in a simplistic way, e.g., by using one marker per relation. In reality, however, language offers a wide range of markers from which informed choices should be made. This paper suggests a method for equipping generators with the knowledge to select the most appropriate discourse marker from a set of candidate expressions. We concentrate on one area of discourse relations, the CONCESSION family, and identify its underlying semantics and pragmatics. On the basis of extensive corpus studies, we propose a new classification of CONCESSION markers in English and German, and then suggest a generation model for producing bilingual text that can inc...

Ma(r)king Concessions in English and German.

A method is described for the operation of an automatic sector information system (Yellow Pages Service) in which a user by means of a terminal (1) connected to the system transmits search parameters to the system. The system then uses these search parameters to search for subscriber data (TD) stored in a system database (3) of suppliers of a sector desired by the user, and outputs corresponding subscriber data (TD) found in the database (3) to the user. The system determines information (SI) on the location of the user and the locations of the suppliers found in the database (3) and for each supplier determines a distance measurement which corresponds to a shortest route between the user and the supplier or the time required to cover the route. The subscriber data of these suppliers are output as a function of the distance measurements of the suppliers concerned. In addition, a corresponding automatic sector information system is claimed.

Method of operating an automatic sector information system

Beschrieben wird ein Verfahren zum Betrieb eines automatischen Branchen-Auskunftssystems
(Gelbe-Seiten-Service), bei welchem ein Nutzer mittels eines mit dem System verbundenen
Endgerats (1) Suchparameter an das System ubermittelt Das System sucht dann
anhand der Suchparameter in einer Datenbank (3) des Systems gespeicherte Teilnehmerdaten
(TD) von Anbietern einer vom Nutzer gewunschten Branche und gibt entsprechende, in der
Datenbank (3) gefundene Teilnehmerdaten (TD) an den Nutzer aus Das System ermittelt
Informationen (SI) uber den Standort des Nutzers und uber die Standorte der in der
Datenbank (3) gefundenen Anbieter und ermittelt fur jeden Anbieter ein Entfernungsmas,
welches einer kurzesten Wegstrecke zwischen dem Nutzer und dem Anbieter oder einer Zeit
entspricht, die benotigt wird, um die Wegstrecke zuruckzulegen Die Ausgabe der
Teilnehmerdaten der Anbieter erfolgt in Abhangigkeit von den Entfernungsmasen der
jeweiligen Anbieter Daruber hinaus wird ein entsprechendes automatisches Branchen-Auskunftssystem
beansprucht

Nils Lenke

Papers

Ma(r)king concessions in English and German

Ma(r)king Concessions in English and German.

Method of operating an automatic sector information system

Verfahren zum Betrieb eines automatischen Branchen-Auskunftssystems

System zur Erkennung gesprochener Ziffernfolgen