RESEARCH IN NATURAL LANGUAGE PROCESSING 
University of Pennsylvania 
Department of Computer and Information Science 
This a brief report 
publications. 
FACULTY 
STUDENTS 
FACILITIES 
summarizing our work to date, our intermediate and long term goals, and a summary of some of our 
Aravind Joshi, Tim Finin, Dale Miller, Lokendra Shastri, and B°nnie Webber 
Brant Cheikes, John Dowding, Amy Felty, Ellen Hays, Robert Kass, Ron Katriel, Sitaram Lanka, Megan 
Moser, CGopalan Nadathur, MaryAngela Papalaskaris, Martha Pollack, Robert Rubinoff, Yves Schahes, Ethel 
Schuster, Sunil Shende, Jill Smudski, Vijayshankar, David Weir, Blair Whitaker 
LINC (Langauge, Information, and Computation) laboratory, which consists of a dedicated VAX 11/785, I0 
Symbolics Lisp machines, 7 HP 68020 based AI workstations, a SUN workstation, several Macintoshes, and 
a laser printer. These machines are networked together and to other research facilities in the department. 
MAJOR THRUST 
Natural language interfaces providing support for many different communicative functions. 
• Providing definitions of concepts 
* Recognizing and correcting user misconceptions 
* Providing explanations 
• Offering to provide information later, when known 
• Verifying and demonstrating understanding 
• Exploiting and enriching the context of natural language discourse between user and system. 
WORK-TO-DATE 
• Integration of RUS-TEXT-MUMBLE (RTM) - This effort involves integrating three natural language system 
components (BBN's RUS parser-interpreter, McKeown's TEXT system (developed at Penn), and McDonald's 
MUMBLE system (received from U. Mass in January 1985). This integration of three independently developed 
systems has required substantial effort. The version of RTM (to be completed in May 1986) \[1\] accepts a limited 
number of English language requests for definitions of, descriptions of, or comparisions between terms in the ONR 
database used by Kathy MeKeown in her development of TEXT; \[2\] formulates appropriate reponses using TEXT 
and outputs those responses in English using MUMBLE; and \[3\] runs on a SYMBOLICS Lisp machine. This work 
has been done by Moser, Whitaker and Rubinoff. 
• Initial work on incorporating a sense of relevance in monitor offers. Mays' dissertation work on monitor offers was 
limited to issues of competancy. This work is being done by Cheikes and Schabes. 
• Completion of McCoy's dissertation work on correcting certain types of object-related misconceptions and 
implementation of a system called ROMPER which generates such corrections. (MUMBLE is used as the tactical 
generation component of this system as well.) 
• Completion of Hirsehberg's dissertation work on scalar implicatures and their use in constructing non-misleading 
responses. 
• Completion of Pollack's dissertation work on plan inference in which user's and system's beliefs about actions and 
plans is dccoupled. 
• Continuation of work on integrating scalar-implicature-based reasoning within a general framework of 
circumscription-based non-monotonic reasoning. 
• Development of methods for converting proofs in a system akin to first-order resolution into natural deduction (ND) 
• proofs, which are then reorganized into cohesive paragraphs using Chester's 1976 algorithm. 
30 
• Development of methods of converting modal resolution proofs into modal ND proofs and higher-order resolution 
proofs into higher-order ND proofs. 
• Initial development of domain-independant tools for expressing and reasoning about user models - in particular, for 
defining hierachies of stereotypical users, representing individual users, and drawing inferences about them using a 
default logic. 
• Continuation of basic research on local coheLeuce of discourse using the notions of centering and syntax, semantics, 
and parsing of tree adjoining grammars. 
FUTURE PLANS 
Having gained the experience of integrating three natural language systems and carrying out some of the basic research as 
described in the previous section, we have now developed the plan described below, which summarizes the near term and long 
term goals. 
Near Term Goals 
We have three tangible goals for the next year: 
• Completing the RTM demonstration system (using the existing domain and knowledge representation) and 
producing a videotape which explains and demouslrates it. 
• Developing TEXT into a more modular tool for defining and comparing terms, on the order of RUS and MUMBLE. 
This will eliminate its tie to a particular knowledge representation and increase its portability. 
• Acquiring familiarity with the PENMAN approach to NL generation through acting as a beta-test site for NIGEL. 
Long Term Goals 
Support for NL Definitions - Enriched Knowledge Representation 
In our original proposal, we stated our intention of employing a richer knowle.xlge representation as the basis for our work on text 
generation, especially for constructing definitions. Our original idea was to make use of BBN's NIKL systenx In the past year 
though, we have become aware of some of NIKL's limitations, which essentially make it non-optimal, even as a next step, for 
our text generation work. On the other hand, we have identified several features with which a NIKL-like language could be 
enriched to make it more suitable for our work: 
• associating non-definitional information with concepts in a way that maintains the underlying structure of that 
information, without interferring with NIKL's automatic classification mechanism. 
• associating "evidential" information with concepts, especially frequency information - how often the concept is 
known to display particular features. 
• allowing for what appears to be conflicting information coming down through inheritance - e.g., information that is 
contrary to expectations grounded in an alternative perspective on a concept 
• allowing mutual definition of concepts - each being defined with reference to the others in a set 
• incorporating notions of time and change - allowing the defming properties and evidential properties of concepts to 
include how they change over time 
• allowing assertions about usual relations between properties of subtypes 
Work on an enriched knowledge representation that includes all these features in a well-motivated way will take several years. 
However one that includes at least the first three of them can probably be developed over the next two years, with work on 
employing it in text generation beginning after the first six months to a year after the start of that work. 
Support of NL Definitions - Use of Discourse and User Models 
The TEXT system, as it is currently structured, will produce the same definition for a concept (or comparison between two 
concepts) whenever it is asked. It does not take into account what the user may have already found out about the concept, or what 
it is implicitly being contrasted with (e.g., some other concept the user has recently asked about), or what the user's goal is in 
making the request. Hence, other directions in which we would like to take this definitional/clarificational capability is to 
increase its sensitivity to (1) the discourse history, to avoid repetition and possibly to take advantage of the additional clarity 
brought by contrasting a new term to one explained before; (2) the user's level of expertise, to avoid either stating the obvious or 
going more deeply into a concept than the user can understand; and (33) the user's goals, to focus on those aspects of the concept 
being defined (or concepts being compared) which are significant to the current tasL (The latter is related to the notion of 
"perspective" used in Kathy McCoy's recent thesis here.) For both these apects of user modelling (in contrast with the first point, 
which can be developed using the current discourse alone), we will draw on the other work being done here on domain- 
independent user-modelling mechanisms. This proposed work must be done in a domain in which tasks can be characterized and 
recognized. Thus we plan to do this initially in investment advising dmnain that we have started to develop. Work on 
31 
incorporating and using discourse history will involved about a one-year effort, once the knowledge base is built. Work on 
incorporating and using a model of a user's expertise and goals will take more time, on the order of two to three years. 
Explanations 
Again in our original proposal, we proposed work on constructing natural language explanations - more specifically, on ways to 
loosen the current tight coupling between the form of the system's proof of some statement to the form of its explanation of why 
the statement is true. This coupling has kept systems which should be able to explain their reasoning from employing stronger 
proof methods which do not have a natural, understandable form of presentation to their human users. 
Our immediate goals involve: 
* developing a demonstration system which responds to NL queries posed to RUS by doing an efficient first-order 
resolution-based proof, transforming that proof into an ND proof, organizing that proof according to an improved 
version of the Chester algorithm, and then producing an English version of the text using MUMBLE or NIGEL. 
• abstracting from the three separate sets of proof conversion methods (noted under WORK-TO-DATE) into general 
methods of transforming any resolution-style proof in any logic into its corresponding ND proof. 
• determining whether existing methods of organizing fwst-order ND proofs into paragraphs are applicable to ND 
proofs in these stronger logics or whether more must be done to produce high-quality, cohesive, understandable text. 
Our loog-term goals remain as stated in our original proposal - the production of explanations sensitive to users' beliefs, 
expertise, desired level of detail and expectations. In this long-term research, we see taking expertise and desired level of detail 
into account in determining how much of the ND proof is made explicit. Of more interest is how users' beliefs and expectations 
should affect the explanations. Work on scientific explanation has shown that central to the explanation of what is the case is a 
set of alternative situations which are not the case. One explains what is in contrast to what is not. However, this requires 
additional work. to prove of each of the alternatives (which may be given explicitly by the user - "why this and not thatT - or 
inferred from the system's model of the user's expectations) that it is not true. Our planned approach involves guiding the 
(failing) proof of each alternative against the successful prooL The point is that although there may be many failing proofs of 
each alternative, the most relevant of these in the current situation is the one which is analogous - up to the point of failure - to 
the original successful proof not only should this technique provide relevant information, but is should also be efficient in 
reducing the search space. We expect this work to take on the order of two to three years, provided we have enough resources to 
pursue it in parallel with our more near-term goals. 
Natural Language Parsing and Generation 
While continuing to use the RUS system, we will continue our work on tree adjoining grammar (TAG) both from the parsing and 
generation points of view. TAGs lead to some attractive approaches to parallelizing parsing and also seem to provide natural 
planning units for generation. This work will be integrated with our future work on parsing and generation. Our first language 
generator (used by TEXT) was one based on Kay's Functional Unification Grammar. While theoretically elegant, it was 
unacceptably slow (in its straightforward implementation), leading us last year to import the MUMBLE generator from 
McDonald at University of Massachusetts and adapt it to work with TEXT. Using MUMBLE has produced a 60-fold speed-up 
in generation time. However, adapting MUMBLE to work with TEXT and, independently, with two other systems has made us 
aware of MUMBLE's limitations, primarily its lack of knowledge of words or grammar. EssentiaUy, MUMBLE's knowledge is 
limited to how to realize particular message units (i.e., to choose an acceptable one from an a priori specified set of choices), 
given constraints already imposed by message units that have already been realized. The large amount of work that must be 
invested in building a MUMBLE lexicon and the lack of inter-application portability of anything but the control structure comes 
from this fact - that one has to completely specify each set of choices beforehand for each message unit and the sets are 
completely application specific. We propose to work on the development of a new architecture, including our work on tags, that 
avoids these limitations by having more knowledge of syntax and words and hence is more portable between applications. The 
time frame for this project is approximately three years. 
Anaphora Resolution 
The RUS parser/interpreter we received from BBN uses a limited method of resolving definite pronouns and noun phrases that is 
only a bit more advanced than the one originally developed for BBN's LUNAR system back in 1971. Since then, there have been 
major theoretical advances in our understanding of discourse anaphora (in the works of Grosz (at SRI), Joshi, Sidner (at BBN), 
Webber, and Weinstein), but these theoretical advances have not yet found their way into natural language understanding 
systems. We feel strongly qualified to undertake this work, having two of the major participants (Joshi and Webber) here at Penn 
already, and want to do so. For us, it is both of research interest and of practical importance, since it can mean a major 
improvement in system's understanding abilities. We will also integrate our work on tags with this effort as it relates to parsing 
and generation. This work wiU also complement additional work being done here on a theoretical and computational account of 
anaphoric reference to actions and events. We see this work as taking about two to two and a half years. 
32 
User Modeling 
The need for systems to model the knowledge and beliefs of their users has already been pointed out. We plan to address a 
number of issues which underly the succesful development and encorporation of explicit user models. Our current domain- 
independent user-modelling system, GUMS, provides mechanisms for defining hierachies of stereotypical users, representing 
individual users, and drawing inferences about them using a rich default logic. We will continue to develop this system as a tool 
which will support the user modeling needs of various applications. We also plan to study the problem of bow new knowledge of 
individual users can be derived from their regular interaction - that is, how relevent information about users can be inferred from 
their queries and responses. In other situations it may become necessary for the system to explicitly pose a few crucial questions 
to the user to determine what he or she does and does not know. 
System Integration 
Finally, we plan to begin work on system integration. In recent years, we have identified many types of behavior that interfaces to 
database systems and expert systems should demonstrate. Beginning with Kaplan's work on recognizing and responding to 
existential presupposition failures in his COOP system, we have developed and produced several modules, each demonslrating 
another type of desired behavior. These include the ability to recognize and respond to type failures, the ability to respond to 
object-related misconceptions, the ability to calculate and offer competant database monitors, the ability to use scalar 
implicatures to convey additional information, and the ability to respond to a class of "inappropriate" queries, and various 
paraphrase abilities. 
Following the publication of Kaplan's thesis, the features of his COOP system were soon incorporated into several database 
interfaces (both natural slanguage and formal query language). This gave the resulting systems the ability to give two types of 
responses: either a direct answer, ff there was one, or a statement concerning the abscnse of individuals satisfying some 
description in the given query. Now we plan to tackle the more significant problem related to this: 
Given a system that is able to call upon a variety of response strategies, how does it decide what to do in a given circumstance? 
This is the issue we plan to explore by investigating the integration of multiple communicative behaviors. Given a system with 
several different types of useful behaviors, which can be combined in various ways, can one efficiently and effectively coordinate 
a response that is better (i.e., more useful, more helpful and more understandable) than simply a (direct) answer. While we 
speculate that it will be the case that identifying what one might consider the best response might take complex reasoning about 
the user's goals, level of expertise and need-to-know with respect to what the answer (if any) actually is, we also plan to look at 
how, with more limited resuurees, we can still improve system behavior. 
This aspect of our future plans is the most long term, involving both the actual component integration itself (in which, in many 
cases, it is only the basic ideas that can be carried over, where the component must be re-programmed entirely to fit into the 
integrated system) and the development of that part of the total system that reasons about what kind of response(s) to give. The 
time frame here is approximately four years. 
Architecture 
We plan to investigate parallel and connectionist architectures and algorithms for realizing our systems, especially those for 
knowledge representation, reasoning, explanations, and integrated parsing and generation. 
Abstracts of Recent Technical Reports 
INTERACTIVE CLASSIFICATION A Technique for the Aquisition and Maintenance of Knowledge Bases, Tim Flnln and 
David Silverman, MS-CIS-84-17. 
The practical application of flame-based knowledge-based systems, such as in expert systems, requires the maintenance of 
potentially very large amounts of declarative knowledge stored in their knowledge bases (KBs). As a KB grows in size and 
complexity, it becomes more difficult to maintain and extend. Even someone who is familiar with the representation and the 
contents of the existing KB may introduce inconsistencies and errors whenever an addition or modification is made. 
This paper describes an approach to thisproblem based on a tool called an interactive classifier. An interactive classifier uses the 
contents of the existing KB and knowledge about its representation to assist the person who is maintaining the KB in describing 
new KB objects. The interactive classifier will identify the ~ppmpriate taxonomic location for the newly described object and 
add it to the KB. The new object is allowed to be a generalization of existing KB objects, enabling the system to learn more 
about existing objects. The ideas have been tested in a system call KuBIC, for Knowledge Base Interactive Classifier, and are 
being extended to a more complete knowledge representation language. 
Correcting Object-Related Misconceptions: How Should The System Respond?, Kathleen F. McCoy, MS-CIS-84-1& 
This paper describes a computational method for correcting users' misconceptions concoming the objects modeled by a computer 
system. The method involves classifying object-related nnsconceptions according to the knowledge-base feature involved in..the 
incorrect information. For each resulting class sub-types are identified, according to the structure of the knowledge base, wmcn 
indicate what information may be supporting the misconception and, therefo.re, what informatio 9 to tnelufle .m the rysponse. 
Such a characterization, along with a model of what the user knows, enables IRe system to reason m a aomam-moepenoent way 
about how best to correct the user. 
33 
Default Reasoning in Interaction, Aravind Joshi, Bonnie Webber, and Ralph Welschedel, MS-CIS-84-58 
Nonmonotonic reasoning is usually studied in the context of a logical system in its own right or as reasoning done by an agent, in 
which the agent reasons about the world from partial information andhence may draw conclusions unsupported by traditional 
logic. The main point of departure here is looking at nonmonotonic reasoning in the context of interacting with another agent. 
This information is partial, in that the other agent neither will not can make everything explicit. Knowing this, the agent may 
attempt to derive more from the interaction than what has been made explicit, by. reasoning by default about what has been m~£te 
explicit (often by contrast with what he assumes would have been made explicit, were something else the case). Thus there can 
be rules for default reasoning that are operative m~'d~'~'~-ac~"~ situation Cinteractional defaults") that are not operative with 
only a single agent. 
Preventing False Inferences, Aravind Josh\[, Bonnie Webber, and Ralph M. Weischec~e!, MS..CIS.84-5~ 
In cooperative man-machine interaction, it is taken as necessary that a system truthfully and informatively respond to a user's 
question. It is not, however, sufficient. In particular, if the system has reason to believe that its planned response might lead the 
user to draw an inference that l~ows to be false, then it must block it by modifying or adding to its response. The problem is 
that a system neither can nor should explore all conclusions a user might possibly draw: its re~oning must be constrained in 
some systematic and weU motivated way. 
Living Up To Expectations: Computing Expert Responses Aravind Josh\[, Bonnie Webber, and Ralph Weischedel, 
MS-CIS-84-60 
In cooperative man-machine interaction, it is necessary but not suJJiclent for a sustem to respond truthfully and informatively to 
a user's question. In particular, if the system has reason to believe that its planned response might mislead the user, then it must 
block that conclusion by modifying its response. This paper focusses on identifying and avoiding potentially misleading 
responses by acknowledging types of "informing behavior" usually expected of an expert. We air.erupt to give a formal account 
of several types of assertaUons that should be included in response to questions concerning the achievement of some goal (in 
addition to the simple answer), lest the questioner otherwise be misled. 
A Modal Temporal Logic for Reasoning About Changing Databases with Applications to Natural Language Question Answering, 
Eric Mays, Aravind Joshl, Bonnie Webber, MS-CIS-85-01. 
A database which models a changing world must evolve in correspondence to the world. Previous work on natural language 
UeStion answering systems for databases has largely ignored the issues which arise when the database is viewed as a dynamic 
ather than a static) object. We investigate the question answering behaviors that become possible with the ability to represent 
and reason about thepossible evolution of a database. These behaviors include offering to monitor for a possible future state of 
the database as an indirect response to a query, and directly answering questions about prior and futare possibility. We apply a 
propositional modal temporal logic that captures possibility and temporality to represent and reason about dynamic databases, 
and present a sound axiomatization and proof and proof procedure. 
Explaining Concepts in Expert Systems: The CLEAR System, Robert Rublnoff, MS-CIS-85-06, LINC LAB 02 
Existing expert systems provide limited explanatory ability. They can explain the specific reasoning the system uses, but if the 
user is confused about the concepts and terms the system is using, no help is available. The CLEAR system allows users to ask 
for explanations of specific concepts. The system generates the explanaU~ns by examining the rule base, selecting rules that are 
relevant to the concept asked about. These rules are then turned into Engfish by various simple translation schemes and 
presented to the user, providing an explanation of how the concept is used by the system. 
The Linguistic Relevance of Tree Adjoining Grammars, Anthony S. Kroch, Department of Linguistics, and Aravind 
K. Joshl, Department of Computer and Information Science, MS-CIS-85-16, LINC LAB 03 
In this paper the linguistic significance of the Tree Adjoining Grammar (TAG) has been investigated. An important property of 
TAG is that it defines a constrained theory of syntactic embedding, one requiring that embedded structures be composed out of 
elementary structures in a fixed way, and one which forces co-occurence relations between elements that are separated in surface 
constituent structures to be stated broadly as constraints on elementary trees in which those elements are copresent. The extra 
generative power of TAG beyond context-free grammar emerges as a corollary of factoring recursion and co-occurence relations. 
The linguistic details specifically discussed are raising constructions, passive, and WH-movements. 
A Computational Logic Approach to Syntax and Semantics Dale A. Miller and Gopalan Nadathur, MS-CIS-85-17 
It is well known that higher-order logics are very expensive, and for this reason have been used to represent many problems in 
mathematics and theoretical computer science. In the latter domain, higher-order logics are often used to describe the semantics 
of first-order logics, natural languages, or programs, since the formalization of such semantics needs a recourse to quantification 
over the domain of functions and sets. In these settings, higher-order logic has generally been limited to a descriptive role. Once 
the formalization is made little has been made of it computationally, largely because there is abundant evidence that theorem 
proving in higher-order logics is very difficult. In this paper we shall look at a sublogic of a particular higher-order logic that is 
derived from Church s Theory of Types, and examine its representational power and its computational tractability. This sublogic 
can also be described as Horn clauses logic extended with quantifications over function variables and R-contraction. We shall 
present a sound and complete theorem prover for this logic, which uses higher-order unification and may be described as an 
extension of a unification procedure for the typed R-calculus. There are at least three ways in which this logic is different from 
the first-order logic that it generalizes. First l't possesses function variables which can be mstantiated with ~.-terms and evaluated 
through ~.-contractions. This provides the logic with a new source of computation. Second, since Z-terms do not have most 
general unifiers, the process off'mding appropriate unifiers must branch, and hence involves real search. This facet provides a 
new source of noncteterminism in specifying computations. Finally, this log.ic can directly encode f'trst-order logic m its term 
stucture and can manipulate such terms in logically meaningful ways. We illustrate this with examples taken from knowledge 
representation and natural language parsing. 
The Role of Perspective In Responding to Property Misconceptions, Kathleen F. McCoy, MS-CIS-85-31, May 1985 
In order to ad.ecluately respond to misconceptions involving an object's properties, we must have a ccontext-sensitive method for 
determining object similarity. Such a method is intrnduced here. Some of the necessary contextual information is captured by a 
new notion of object perspective It is shown how object perspective can be used to account for different responses to a given 
34 
misconception in different contexts. 
Some Computational Properties of Tree Adjoining Grammars, ViJayshenkar and Joshl, MS-CIS-85-07 
Tree Adjoining Grammar (TAG) is a formalism for natural language grammars. Some of the basic notions of TAG's were 
introduced in \[Joshi,Levy, and Takahashi 1975\] and by \[Joshi, 1983\]. A detailed investigation of the linguistic relevance of 
TAG's has been carried out in \[Kroch and Joshi,1985\]. In this paper, we will describe some new results for TAG's, especially in 
the following areas: (1) parsing complexity of TAG's, (2) some closure results for TAG'S, and (3) the relationship to Head grammars. 
Grammar, Phrase Structure, Aravtnd K. Josht, MS-CIS-85-45 
Phrase-slructure trees (phrase-markers) provide structural descriptions for sentences. Phrase-structure trees can be generated by phrase-structure grammars. 
Phrase-structure trees can be shown to be appropriate to characterize structural descriptions for 
sentences, including those aspects which are usually characterized by transformational grammars, by making certain amendations 
to CFG's, without increasing their power, or by generating them from elementary trees (phrase-markers) by a suitable rule of 
composition, increasing the ~powcr only mildly beyond that of CFG's. Structural descriptions provided by phrase-structure trees 
are used explicitly or implicltly in natural language processing systems. 
Uestion, Answer and Responses: Interacting with Knowledge Base Systems, Bonnie Lynn Webber, MS-CIS-85-50, LINC O4 
The purpose of this chapter is to examine the character of information-seeking interactions between a user and a knowledge base 
system (KBS). In doing so, I advocate that a clear distinction be made between an answer to a c.luestion and a response. The 
chapter characterizes questions, answers, and responses, the role they play in effective information interchanges, and what is 
involved in facilitating such interactions between user and KBS. 
A Theory Of Scalar Implicature,Julla Bell Hlrschberg, MS-CIS-85-56. 
The Relationship Between Tree Adjoining Grammars And Head Grammars, K. ViJay-Shanker, David J. Weir and Aravind 
K. Joshl, MS-CIS-86-01, LINC LAB 06 
Tree Adjoining Grammars (TAG) and Head Grammars (HG) were introduced to capture certain structural properties of natural 
languages. These formalisms, which were developed independently, appear to be quite different notationaliy. In this paper we 
discuss the formal relationship between the class of languages generated by TAG's (TAL) and the class of languages generated 
by HG's (HL). In particular, we show that HL's are included in TAL's andthat TAG's are equivalent to a modification of HG:s 
called Modified Head Grammars (MHG's). The inclusion of MHL in HI.,, and thus the equivalence of HG's and TAG's, in the 
most general case remains to be established. We show that this relationship is very close both linguistically and formally, the 
difference hinging on the status of heads of empty swings and whether one deals with heads directly or with the left and right 
wrapping positions around the head. 
Natural Language Interactions With Artflcial Experts, Tim Finin, Aravlnd K. Jeshl and Bonnie Lynn Webber, MS- CIS-86-16, LINC LAB 08). 
The aim of this paper is to justify why Natural Language (NL) interaction, of a very rich functionality, is critical to the effective ] 
se of Expert Systems and to describe what is needed and what has been done to support such interaction. Interactive functions 
iscussed here include defining terms, paraphrasing, correcting misconceptions, avoiding misconceptions and modifying questions. 
Higher.Order Logic Programming, Dale A. Miller and Gopalan Nadathur, MS-CIS-86-17 
In this paper we consider the problem of extending Prolog to include predicate and function variables and typed ~.-terms. For this 
purpose, we use a higher-order logic to describe a generalization to first-order Horn clauses. We show that this extension 
possesses certain desirable computational properties. Specifically, we show that the familiar operational and least fixpoint 
semantics can be given to these clauses. A language, ~.Prolong that is based on this generalization is then presented, and several 
examples of its use are provided. We also discuss an interpreter for this language in which new sources of branching and 
backtracking must be accommodated. An experimental interpreter has been constructed for the language, and aLl the examples in this paper have been tested using it. 
Some Uses of Higher.Order Logic in Computational Linguistics, Dale A. MiLler and Gopalan Nadathur, MS-CIS-86-31, LINC LAB 08 
Consideration of the question of meaning in the framework of linguistics often requires and allusion to sets and other 
higher-order notions. The traditional approach to representing and reasoning about meaning in a computational setting has been 
to use knowledge representation systems that are either based on first-order logic or that use mechanisms whose formal 
justifications are to be provided after the fact. In this paper we shall consider the use of a higher-order logic for this task. We 
first present a version of definite clauses (positive Horn clauses) that is based on this logic. Predicate and function variables may 
occur m such clauses the terms in the language are the typed/-terms. Such term structures have a richness that may be exploited 
in representing meanings. We also describe a higher-order logic programming language, called /Prolog, which represents 
programs as hijgher-order definite clauses and interprets them using a depth-first interpreter. A virtue of this language is that it is 
possible to write programs in it that integrate syntactic and semantic analyses into one computational paradigm. This is to be 
cont~'asted ~vith the more common pFactice of using two entirely different computation paradigms, such as DCGs or ATNs for 
parsing ann frames or semantic nets for semantic processing. We illustrate such and integration in this language by considering a 
stmple example, and we claim that its use makes the task of providing formal justifications for the computations specified much more airect. . 
Some Aspects Of Default Reasoning In Interactive Discourse, Aravlnd K. Joshl, Bonnie L. Webber and Ralph 
M. Weischedel, MS-CIS-86-27 (revised version of MS-CIS-84-58) 
In cooperative inter.action_, it is taken as necessary that a system truthfully and informatively respond to a user's question. It is 
not, however, sur.nClent. In par~cuiar, ff the system has reason to befieve that its planned response might lead the user to draw an inference mat it knows to be false, then it must block it by modifying or adding to its response. In this paper we investigate 
several aspects of such reasoning in interactive discourse. 
35 
Adapting MUMBLE: Experience with N~tuml Language Generation, Robert Rublnoff, MS-CIS-86-32, LINC LAB 09} 
This paper describes the construction of a MUMBLE-based \[McDonald 83b tactical component for the TEXT text 
generation system \[McKeown 85\]. This new component, which produces fluent English sentences from the sequence of 
structured message units output from TEXT's strategic component, has produced a 60-fold speed-up in sentence production. 
Adapting MUMBLE required work on each of the three parts of the MUMBLE framework: the interpreter, the grammar, and the 
dictionary.. It also provided some insight into the generation process and the consequences of MUMBLE's commitment to a 
determlmsttc model. 
GUMS 1 : A General User Modeling System, Tim Finin and David Drager, MS-CIS-86-35 
This paper describes a general architecture of a domain independent system for building and maintaining long term models of 
individual users. The user modeling system is intended toprovide a well defined set of services for an application system which 
is interacting with various users andhas a need to build andmaintain models of the-re. As the application system interacts with a 
user, it can acquire knowledge of him and pass that knowledge on to the user model maintenance system for incorporation. We 
describe a prototype general user modeling system which we have implemented in Prolog. This system satisfies some of the 
desirable characteristics we discuss. 
Breaking the Primitive Concept Barrier, Robert Kass, Ron Katrlel, and Tim Finln, MS-CIS-86-36 
Building and maintaining a large knowledge base of general information requires a knowledge representation system with \]?recise 
semantics and an easy knowledge acquisition procedure. Systems such as KL-ONE meet these criteria by using a classifier to 
install new concepts into a taxonomic sWacture. These systems use a formal notion of a definition for concepts. Unfortunately, 
many concepts do not seem to have such precise definittons, and end up represented as primitive concepts. Primitive concepts 
form a barrier to classification, forcing the user to manually classify a new concept with respect to all primitive concepts in the 
knowledge base. 
We propose an extension to ~NE which retains its soundness and greatly reduces the burden on the user during knowledge 
acquisition. This extension consists of adding an explicit definitional component to concepts and relaxing the strictness of 
concept definitions themselves. The relaxed definition reduces the number primitive concepts in a knowledge base, enables the 
classifier to handle concepts that do not have complete definitions and enhances the usefulness of an interactive classifier. 
36 
