LINGUIST List 12.688
Mon Mar 12 2001
Review: Reiter & Dale, Building NL Generation Systems
Editor for this issue: Terence Langendoen <terrylinguistlist.org>
What follows is another discussion note contributed to our Book Discussion
Forum. We expect these discussions to be informal and interactive; and
the author of the book discussed is cordially invited to join in.
If you are interested in leading a book discussion, look for books
announced on LINGUIST as "available for discussion." (This means that
the publisher has sent us a review copy.) Then contact Simin Karimi at
simin
linguistlist.org or Terry Langendoen at terry
linguistlist.org.
Directory
[iso-8859-1] Korn�l R. Bangha, review: Reiter, Ehud, and Robert Dale (2000) Building Natural
Language Generation Systems
Message 1: review: Reiter, Ehud, and Robert Dale (2000) Building Natural
Language Generation Systems
Date: Mon, 12 Mar 2001 14:52:42 -0500
From: [iso-8859-1] Korn�l R. Bangha <banghakMAGELLAN.UMontreal.CA>
Subject: review: Reiter, Ehud, and Robert Dale (2000) Building Natural
Language Generation Systems
Reiter, Ehud, and Robert Dale (2000) Building Natural
Language Generation Systems, Cambridge University Press, 248pp.
Reviewed by: Kornel Bangha, University of Montreal
SYNOPSIS
This book explains how to build Natural Language Generation
(NLG) systems - computer software systems which use
techniques from artificial intelligence and computational
linguistics to automatically generate understandable texts
in English or other human languages. Typically starting
from some non-linguistic representation of information of
input, NLG systems use knowledge about language and the
application domain to automatically produce documents,
reports, explanations, help messages, and other kinds of
texts. The book is based on one particular architectural
decomposition of the NLG task which consists of three
modules: document planning, microplanning, and surface
realisation.
Chapter 1 - Introduction
NLG is presented both from a research perspective and from
an applications perspective. From a research perspective,
NLG is a subfield of natural language processing which in
turn can be seen as a subfield of both computer science and
cognitive sciences. The authors make a comparison between
generation and understanding, the other part of natural
language processing. From an applications perspective, most
current NLG systems are used either to present information
to users, or to (partially) automate the production of
routine documentation. Six examples of NLG systems are
presented in this chapter followed by a short history of
NLG.
Chapter 2 - Natural Language Generation in Practice
This chapter considers alternatives to NLG and examines the
circumstances under which it is appropriate to use NLG
systems. The first question considered is when an NLG
system is indeed the most appropriate. Advantages and
disadvantages of alternatives like use of graphics, mail-
merge systems and human authoring are considered. An
important issue for the question is how to create a corpus
to determine user requirements. Evaluating and fielding NLG
systems are also discussed.
Chapter 3 - The Architecture of a Natural Language
Generation System
This chapter gives an overview of the inputs and the
outputs of NLG and of a specific architecture that embodies
one particular decomposition of the process into distinct
modules. One can caracterise the input of an NLG system as
a four-tuple containing the knowledge source, the
communicative goal, the user model and the discourse
history. The output of the generation is a text. However,
this is much more than just a stream of ASCII text. The
generation process can be decomposed into three component
modules, which will be referred to as the document planner,
the mircoplanner, and the surface realiser. Each of these
three modules accomplishes a content task and a structure
task. An overview of these modules is presented here, each
of them will be studied in the following chapters.
Chapter 4 - Document planning
This chapter is about document planning: it presents what
the task of the first component of an NLG system is and how
it works. The document planner is responsible for deciding
what information - messages - to communicate (this being
the task of content determination) and determining how this
information should be structured for presentation (this
being the task of document structuring). Content
determination usually involves one or more of selecting,
summarising and reasoning with data. Very few NLG systems
simply generate massages that communicate all the input
data. Document structuring is important because documents
are not just random collections of sentences. They possess
coherence and thematic structure, which is to say that the
content is expressed in a way that is easy for humans to
read and understand.
Chapter 5 - Microplanning
This chapter presents why microplanning is important and
what it is concerned with: lexicalisation, aggregation and
referring expression generation. Lexicalisation is the
process of choosing word and syntactic structures to
communicate the information. Aggregation takes a set of
simple phrase specification and combines them to permit the
generation of more complex sentence structures. Referring
expression generation turns knowledge base entities into
semantic content of noun-phrase referring expressions that
will be sufficient to identify the intended referents to
the hearer.
Chapter 6 - Surface Realisation
This chapter describes the surface realisation which is the
third module of the NLG system proposed in this book. It
has two parts: the linguistic realisation and the structure
realisation. Linguistic realisation is the task of
converting abstract representations of sentences into a
real text; it corresponds to the content task of surface
realisation. Structure realisation is the task of
converting abstract structures such as paragraphs and
sections into the mark-up symbols understood by the NLG
system being used; this corresponds to the structural side
of surface realisation. The tasks accomplished by the
surface realisation are illustrated here by three NLG
systems.
Chapter 7 - Beyond Text Generation
This final chapter looks beyond text generation and
examines some of the issues that arise when one considers
the generation of text contained within some medium. The
authors present the role of typography, graphics and
hypertext in NLG and the problems that arise when they are
implemented in real systems. The speech output is also
considered.
CRITICAL EVALUATION
The authors are concerned with both theoretical and
practical questions about NLG. It is clear however, that
their main interest is practical: how to construct NLG
systems which are useful and work efficiently. This also
means that the authors are concerned less with fundamental
linguistic questions than with knowledge representation and
with the mapping of knowledge structures into linguistic
representations.
Most of the time, the content of the book is clear and easy
to read. A great number of examples is provided.
The focus of the book is very large and each of the
specific aspects of NLG can not be discussed in details: it
is rather intended as a general introduction to the topic
of NLG. Pointers to further readings provided at the end of
each chapter can help the reader.
About the reviewer:
Kornel Bangha prepares a Ph. D. of linguistics and
Artificial Intelligence at the University of Montreal. His
research is about how the process of interpretation of
linguistic units in discourse is influenced not only by
semantic factors but also by the context and by knowledge
about the world.