The Counselor Project at the Un/versity of Massachusetts 
David D. McDonald & James D. Pustejowky 
Department of Computer and Information Science 
Un/versity of Massachusetts, 
Amherst, Massachusetts 01003 
Participants in the Counselor Project, Fall 1984 through Summer 1986: 
PrincipAl Investigators: Edwina L. Risslemd. l~tvid D. McDonedd, Wendy G. Lehnert 
Research Associates: Beverly Woolf, James e. Pustejovsky 
Gradutte Students: Marie M. Vaughem, Briim Stucky, Penelope Sibun. Seth Rosenberg, Yelly 
MurreLy, Kevin Getllagher, JoAnn M. Brooks, John Brolio, Sabine Bergler, \[evin D. Ashley, Scott 
D. Anderson 
Introduction 
The COUNSELOR PROJECT began in the fall of 1984 with the goal of exploring basic 
problems in discourse structure and text processing within an integrated interface to a 
strong expert system. The program that we have developed, COUNSELOR, integrates 
separately developed components for natural language generation (MUMeLE see \[7\], \[8\], 
\[9\]), parsing (PLUM \[5\]), and case-based legal reasoning (Hvpo \[1\], \[2\]). It adds a newly 
developed component, CICERO (\[ 10\]), positioned between the two text processors and 
the expert system; CIc~o is responsible for managing textual inferences ("reading 
between the lines") by using common sense models of legal events. Coussez~t can 
provide advise to an attorney about how to argue cases involving violations of trade 
secret law in the computer field. The attorney presents the facts of their case to the 
system, which may ask questions to elicit other facts that it knows to be relevant. The 
system then suggests lines of argument that the attorney might use, drawing on its 
library of litigated cases to find ones with analogous dimensions. 
At its present state of development, COUNSELOR can handle simple variations on a 
single scenerio, exemplified by the following dialog: 
User: 
Counselor: 
User: 
Counselor: 
User: 
Counselor: 
I represent a client named Hacklnc, who wants to sue Swipelnc and Leroy 
Soleil for misappropriating trade secrets in connection with software 
developed by my client. Hacklnc markets the software, known as Autotell, a 
program to automate some of a bemk teller's fUnctions, to the banking 
industry. 
Did Soleil work for Hacklnc.? 
Yes. he was s key employee on the Autotell project. 
Did he later work for Swipelnc.? 
Yes. 
You can argue that there is an implied agreement arising out of Soleil's 
employment with Hscklnc. that he not disclose any trade secret 
inform ttion to which he gained eu:cess by virtue of his employment. 
26 
Motivations 
Consequential results in natural language research will only come from working 
with a strong underlying program whose communicative needs will challenge the 
capabilities of state of the art of language interfaces. As a group, we are not 
interested in building yet another question answering system: our goal is to 
understand the structure of discourse. We believe that an effective place to begin is 
with task specific, mixed initiative dialog where the particiants' goals cannot be 
satisfied by single utterances. 
Working with a legal reasoning system like Kevin Ashley and Edwina Rissland's 
Hvpo provides particular challenges to natural language research: 
(I) Legal text is structurally complex. The need to avoid ambiguity leads to deeply 
embedded clauses and heavy noun phrases. 
(2) As both the user and the system have a thorough knowledge of the law, they 
communicate vastly more information in conversations about legal arguments than 
ever appears in their literal utterances. 
(3) Hvpo's role as an advisory system creates a natural motivation to communicate 
through language. 
(4) Legal cases are large, complex objects that can be viewed from many 
alternative perspectives. The purpose for which a case is being described strongly 
influences which of its attributes are salient and how that information should be 
structured as a text. 
Component Parts 
We began the project with three partially developed components, Hvpo, MUMBLE, 
and PLUM, each designed with independent motivations. An initial tension was 
whether to convert aspects of these programs that did not seem apt in their new 
setting, or alternatively to interpose new components between them to smooth out the 
differences. We concluded that the motivations underlying each component were 
strong enough that we should not change them just because they were now working 
together. 
H~o reasons with cases and hypotheticals. Actually litigated legal cases are 
encoded and indexed by "dimensions", which capture the utility of a case for making a 
particular kind of argument. When evaluating new cases, Hvpo first analyzes them in 
terms of the dimensions they involve. Relevant cases are then retrieved to guide the 
reasoning. The system may ask pertinent questions about facts now found to be 
relevant. When the analysis is complete, H~o describes the arguments available to 
the user, and responses and counter responses that may follow. 
MUHSLS, the linguistic component for generation, is responsible for realizing 
conceptual specifications as grammatical text cohesive with the discourse which 
proceeds it. MU~LE works within a description directed framework. Its input 
specification is a description of the message the underlying program wants to 
27 
communicate. This description is executed incrementally, producin~ an intermediate 
llnguistic representation wl~ich defines the text's grammatical relatlons and imposes 
constraints on further realization. This surface structure description is concurrently 
executed, producing the actual text. 
PLUM is a conceptual analyzer which has been given a well defined schematic 
structure so that it can be easily extended. It parses by doing prediction and 
completion over semantic concepts implied by the words rather than over syntactic 
categories. As in other conceptual analyzers, no explicit surface structure is recovered. 
Pt,uM's output is the set of completed frames. 
Clcmto is a new component, a discourse and inference manager between the 
language components and the expert system. From the understanding side, CICE20 
must integrate the clause by clause output of the parser into the larger discourse 
context, recognizing, for example, when noun phrases refer to the same object. In 
interpreting these small, lexically derived frames, CIcEio draws on its own 
representation of events which bridges the gap between the way such information is 
expressed in language and the way it is organized for expert legal reasoning. For 
generation, CIcEI~o is responsible for planning the message that is given to the 
generator. In particular, it determines what information should be included and what 
may be omitted as inferable, and it selects pivotal lexical items with appropriate 
perspective and rhetorical force. 
Future Directions 
While the accomplishments of the individual components of Counselor are 
interesting in their own right, the greatest effect of the project has been to provide a 
workbench for studying the problems of language in an integrated context. Perennial 
problems in anaphora, lexical semantics, aspect, etc. become more tractable in an 
integrated system where there is a discourse context and intensional motivation. 
There are also semantic generalizations between the level at which the text processors 
operate and the level of the expert system which are more easily captured when 
parsing and generation can be studied in unison. On a larger scale, an explicit 
discourse manager, a requisite for more complex dialogs, can only be developed once 
an integrated system exists. 
References 
\[I\] Ashley, Kevin D. (1986) "Modelling Legal Argument: Reasoning with csses and hypotheticals 
-- s thesis proposal", Technical Report I0, The Counselor Project, Department of Computer 
and Informstion Science, University of Msssschusetts st Amherst. 
\[Z\] Ashley, Kevin D. and Edvins L. Rissland (1985) "Toward Modelling Legal Argument'. 
Proceedings of the 2rid International Congress LOGICA, IIqFORMATICA, DIRITTO. Institute 
~: Per I,e, Documentazione Giuridics, Florence, Italy. 
\[3\] Brooks, JoAnn M. (1985) "Themis: A Discourse Manager",unpubfished Master's thesis, 
Department of Computer and Information Science, University of Massachusetts st jtm h erst. 
28 
\[41 Gallagher, gevin (1986) "The Design and Implementation of CICEIIO', unpublished Master's 
thesis, Depsrtment of Computer and Information Science, University of Massachusetts st 
Amherst. 
\[5\] Lehnert, Wendy G. and Seth Rosenberg (1985) "The PLUM User's Manual" TechnicalReport I, 
The Counselor Project. Department of Computer and Information Science, University of 
Massachusetts at Amherst. 
\[6\] McDonald, David D. (1986) "Natural Language Generation: Complexities and Techniques", to 
appear in Nirenburg (ed.) Theoretic&l and Methodologic&l Issues in M&chine 
Transl~tion. Cambridge University Press. 
\[7\] McDonald, Dzvid D. and James Pustejovsky (1985) "Description Directed Nstural Langauge 
Generation', Proceedings of IJCAI-85, pp. 799-805. 
\[8\] McDenald, David D. and Jsmes Pustejovsky (1985) "TAGs as a Grammatical Formalism for 
Generation", Proceedings of the 23rd Meeting of the Association for Computational 
Linguistics, pp. 94--103. 
\[91 McDonald, David D. and Jxmes Pustejovsky (1985) "SAMSON: A ComputaLional Theory of Prose 
Style for Natoral Language Generation', Proceedings oi" the 1985 meeting of the European 
Association for Computational Linguistics. 
\[ I0\] Pustejovsky. James (1986) "An Integrated Theory Discourse Analysis", Technical Report I I. 
The Counselor Project, Depa~'tment off Computer and Information Science, University of 
Massachusetts at Amherst. 
\[II\] Risslsnd Edwina L., Edward Valcarce, and Kevin Ashley (1984) "Explaining and Arguing with 
Examples", Proceedings of AAAI-84. 
\[12\] Vaughan, MaJ'ie M, and David D. McDonald (1986) "A Model off Revision in Natural Language 
Generation", Proceedings of the 24th Meeting of the Association for Computational 
Linguistics. 
29 
