Uwe Reyle

Dealing with Ambiguities by Underspecification: Construction, Representation and Deduction

In this paper we develop a theory of language meaning that represents scope ambiguities by underspecified structures. The set of possible meanings of a sentence, or text is determined by a set of meta-level constraints that restricts the class of semantic representations appropriately. Thus the way ambiguities are represented does not correspond to any of the usual concepts of formalizing ambiguities by means of disjunctions. A sound and complete proof theory is provided that relates these structures directly, without considering cases.

N-Prolog: An extension of Prolog with hypothetical implications. I.

Abstract An extension of PROLOG called N-PROLOG is presented. N-PROLOG allows hypothetical implications in the clauses. For clauses without implication, N-PROLOG acts like PROLOG. Examples are given to show the need for N-PROLOG. N-PROLOG is a self-reflecting language; it is equal to its own metalanguage. N-PROLOG is more suitable for expressing temporal behavior (change in time). Ordinary PROLOG is conceptually weaker than N-PROLOG.

Natural Language Parsing and Linguistic Theories

Seperating Linguistic Analyses from Linguistic Theories.- Applicability of Indexed Grammars to Natural Languages.- A Natural Language Toolkit: Reconciling Theory with Practice.- An Extension of LR-Parsing for Lexical Functional Grammar.- An Efficiency-Oriented LFG Parser.- Parsing with a GB-Grammar.- Combining Categorial Grammar and Unification.- A feature-Based Categorial Morpho-Syntax for Japanese.- The Treatment of the French adjectif detache in Lexical Functional Grammar.- Some Problems of Coordination in German.- German Word Order and Universal Grammar.- Nonlocal-Dependencies and Infinitival Constructions in German.- GPSG and German Word Order.- Nested Cooper Storage: The Proper Treatment of Quantification in Ordinary Noun Phrases.- Compositional Semantics for LFG.

Principle based semantics for HPSG

The paper presents a constraint based semantic formalism for HPSG. The syntax-semantics interface directly implements syntactic conditions on quantifier scoping and distributivity.1 The construction of semantic representations is guided by general principles governing the interaction between santax and semantics. Each of these principles acts as a constraint to narrow down the set of possible interpretations of a sentence. Meanings of ambiguous sentences are represented by single partial representations (so-called U(nderspecified) D(iscourse) R(epresentation) S(tructure)s) to which further constraints can be added monotonically to gain more information about the content of a sentence. There is no need to build up a large number of alternative representations of the sentence which are then filtered by subsequent discourse and world knowledge. The advantage of UDRSs is not only that they allow for monotonic incremental interpretation but also that they are equipped with truth conditions and a proof theory that allows for inferences to be drawn directly on structures where quantifier scope is not resolved.

A calculus for first order Discourse Representation Structures

This paper presents a sound and complete proof system for the first order fragment of Discourse Representation Theory. Since the inferences that human language users draw from the verbal input they receive for the most transcend the capacities of such a system, it can be no more than a basis on which more powerful systems, which are capable of producing those inferences, may then be built. Nevertheless, even within the general setting of first order logic the structure of the “formulas” of DRS-languages, i.e. of the Discourse Representation Structures suggest for the components of such a system inference rules that differ somewhat from those usually found in proof systems for the first order predicate calculus and which are, we believe, more in keeping with inference patterns that are actually employed in common sense reasoning.This is why we have decided to publish the present exercise, in spite of the fact that it is not one for which a great deal of originality could be claimed. In fact, it could be argued that the problem addressed in this paper was solved when Gödel first established the completeness of the system of Principia Mathematica for first order logic. For the DRS-languages we consider here are straightforwardly intertranslatable with standard formulations of the predicate calculus; in fact the translations are so straightforward that any sound and complete proof system for first order logic can be used as a sound and complete proof system for DRSs: simply translate the DRSs into formulas of predicate logic and then proceed as usual. As a matter of fact, this is how one has chosen to proceed in some implementations of DRT, which involve inferencing as well as semantic representation; an example is the Lex system developed jointly by IBM and the University of Tübingen (see in particular (Guenthner et al. 1986)).In the light of the close and simple connections between DRT and standard predicate logic, publication of what will be presented in this paper can be justified only in terms of the special mash we have tried to achieve between the general form and the particular rules of our proof system on the one hand and on the other the distinctive architecture of DRS-like semantic representation. Some additional justification is necessary, however, as there exist a number of other proof systems for first order DRT, some of which have pursued more or less the same aims that have motivated the system presented here. We are explicitly aware of those developed by (Koons 1988), (Saurer 1990), (Sedogbo and Eytan 1987), (Reinhart 1989), (Gabbay and Reyle 1994); perhaps there are others. (Sedogbo and Eytan 1987) is a tableau system, and (Reinhart 1989) and (Gabbay and Reyle 1994) are resolution based, goal directed. These systems may promise particular advantages when it comes to implementing inference engines operating on DRS-like premises. But they do not aim to conform to certain canons of actual inferencing by human interpreters of natural language; and indeed the proof procedures they propose depart quite drastically from what one could plausibly assume to go in the head of such an interpreter. Only (Koons 1988) and (Saurer 1990) are, like our system, inspired by the methods of natural deduction. But there are some differences in the choice of basic rules. In particular both (Koons 1988) and (Saurer 1990) have among their primitive rules the Rule of Reiteration, which permits the copying of a DRS condition from a DRS to any of its sub-DRSs. In our system this is a derived rule (see Section 4 below).We will develop our system in several stages. The necessary intuitions and the formal background are provided in Sections 1 and 2. (The formal definitions can be found also in the first two chapters of (Kamp and Reyle 1993). The first system we present is for a sublanguage of the one defined in Section 2, which differs from the full language in that it lacks identity and disjunction. The core of the paper consists of Section 3, where the proof system for this sublanguage is presented, and Section 5, which extends the system for the full language, including disjunctions (Section 5.1) and identity (Section 5.2) and then establishes soundness and completeness for the full system. Section 4 deals with certain derived inference principles.

On reasoning with ambiguities

The paper adresses the problem of reasoning with ambiguities. Semantic representations are presented that leave scope relations between quantifiers and/or other operators unspecified. Truth conditions are provided for these representations and different consequence relations are judged on the basis of intuitive correctness. Finally inference patterns are presented that operate directly on these underspecified structures, i.e. do not rely on any translation into the set of their disambiguations.

Ontology-driven discourse analysis for information extraction

This paper presents a novel approach to discourse analysis within information extraction systems. It makes use of DRT as formal representation of the linguistic context as well as of a domain-specific ontology as a basis to compute conceptual relations between extracted events thus establishing discourse coherence. The approach has been implemented within GenIE, an information extraction system with the aim of extracting information about biochemical pathways, about sequences, structures and functions of genomes and proteins. The approach is evaluated against a semantically hand-annotated set of Swiss-Prot protein function descriptions and shows very promising results.

Developing a protein‐interactions ontology

The prediction and analysis of a protein’s function is an ongoing challenge in the field of genomics. With upcoming datasets on protein interactions [9], it is becoming evident that the function of a protein can only be understood when taking its interaction with other molecules into account. Most current approaches to the classification and description of protein function, such as the Gene Ontology [8], focus on single proteins. These annotation efforts should be paralleled by the development of ontologies dealing with the interactions of a protein with other biomolecules. Currently, most approaches to building such ontologies focus on metabolism [3,6]. So far, for interactions, only high-level classifications have been created [4], developed to assist information extraction from text. In addition to assisting text mining, a more fine-grained (in comparison to these classifications) ontology on protein interactions could be helpful in database development and information mining. As an ontology captures domain knowledge in a computer-understandable way, it can be used for inferencing, i.e. deriving new knowledge from existing data. There are two important points to consider in developing such a formal ontology: (a) it should be independent of its final use; and (b) it should not only restrict itself to a controlled vocabulary but the concepts should be related to each other in a semantically consistent manner, and rules governing these definitions and relations should be incorporated whenever necessary. Here we describe our approach for developing such an ontology.

A General Reasoning Scheme for Underspecified Representations

Underspecified semantic representations have attracted increasing interest within computational linguistics. Several formalisms have been developed that allow to represent sentence or text meanings with that degree of specificity that is determined by the context of interpretation. As the context changes they must allow for (partial) disambiguation steps performed by a process of refinement that goes hand in hand with the construction algorithm. And as the interpretation of phrases often1 relies on deductive principles and thus any construction algorithm must be able to integrate the results of deductive processes, any semantic formalism should be equipped with a deductive component that operates directly on its semantic forms.

Compositional Semantics for LFG

The two levels of syntactic representation induced by lexical functional grammars (LFG) consist of c-structures and sets of attribute-value pairs called f-structures. F-structures are built up from the syntactic surface structure of a sentence by unification of the attribute-value graphs that are associated with each word and each constituent of the c-structure of that sentence. Thus (2) shows an f-structure for the sentence (1).