[Return to programme grid]
Chair: Sebastian Loebner
[Return to programme grid]
Abstracts Invited Lectures
Nicola Guarino On the semantics of functional roles
In some previous papers (Guarino 1992, Guarino 2009), I argued that the concepts suitable to be interpreted as attributes in a frame, through the mechanism that I labelled relational interpretation, can denote parts, qualities, or roles. I will focus here on the latter category. Despite a lot of work on the ontology of roles recently (Boella et al. 2007), still there are a number of issues open. Typical examples of role terms are student or president, which are usually assumed to denote unary properties resulting from abstractions of binary relations (such as student-of and president-of), which in addition are anti-rigid, in the sense that such unary properties are contingent for all their instances. So, according to the OntoClean terminology (Guarino and Welty 2002), the meta-properties that characterise roles are anti-rigidity and dependence. I will argue in these talk that this characterisation is too coarse to account for the cognitive intuitions about roles and the linguistic behaviour of certain role terms. [Continued]
Frank Richter Redundancy and Identity in Semantic Theory and Computation
Manfred Sailer's preceding talk investigated a case of redundant semantic marking in the lexicon that concerns heads and their dependents, and he argues for its specification in a particular feature-logic based framework (LRS). In this talk I want to put optional, obligatory or impossible redundant marking in lexical semantics in a broader context: Comparing Manfred's analysis to other instances of redundant marking in compositional semantics (wh questions in German, tense in Afrikaans, negative concord in Polish, Romanian, French and German), I discuss details of their feature-logical specification and relate these details to the implementation of a constraint-based semantics in the Constraint Language for Lexical Resource Semantics (joint work with Gerald Penn). Moreover, I want to draw attention to the connection of these issues to similar questions that have recently been raised by Ovchinnikova (2011) with respect to the implementation of a natural language understanding system which relies on the identification and merging of redundant information at the discourse level by means of inference. As we will see, the relevant discourse redundancies also originate with lexical resources, in this case with axioms gained from WordNet and from FrameNet.
Manfred Sailer Concord and Contribution Constraints in Lexical Semantics
In previous work we have shown that the syntax semantics interface of natural languages makes use of redundancy and non-redundancy in the morphosyntactic marking of semantic operators. In Richter/Sailer 2006 we illustrate this with the marking of negation in various languages: Standard German does not allow negative concord, so every element that expresses a negation contributes a negation operator of its own. Polish is a strict negative concord language. This means that if several expressions may contribute a negation, they must contribute the same negation operator. French, an optional negative concord language, allows for both options. The framework of Lexical Resource Semantics (LRS, Richter/Sailer 2004) was developped to capture variation of this kind in a natural way. We argued in various places that we find a similar pattern of redundant and non-redundant marking with other semantic operators, such as the interrogative operator and tense.
LRS is a constraint-based mechanism that works with feature structures as semantic representations. As such, there is a close connection to the modelling of frames in Petersen 2007. So far, however, the two approaches differ in the empirical domain that they have been applied to: LRS has been primarily used to analyze phenomena of combinatorial semantics that had to do with scope and clausal semantic operators. In contrast to this the phenomena treated within the frame-framework have been primarily lexical.
In this talk we will look at phenomena that concern semantic contributions of heads and their immediate dependents. We argue that in this domain, too, there is motivation for the combinatorial mechanism of LRS, i.e., that we find obligatory non-redundant marking, optional redundancy and obligatory redundancy.
In English, a number of verbs show an alternation between a PP complement headed by a meaningful prepositon and an NP complement ("discuss the issue/ about the issue"). In the case of a PP complement the choice of the preposition is not arbitrary but in line with the semantic relation between the verb and the complement. However, this marking is redundant: As the transitive version shows, the verb alone already encodes this relation. We also find verbs such as "talk", which show an alternation between an about-PP and an of-PP. The of-PP is taken as semantically vacuous. Consequently, at the semantic side, such verbs are to be treated like "discuss", but they differ syntactically in not allowing an NP complement, i.e. as cases of an optionally redundant marking.
In German, the verb "diskutieren" ("discuss") behaves like its English counterpart, whereas "sprechen" ("talk") only allows for the analogue of an about-PP. We could either say that this is an instance of an obligatorily redundant marking or an instance of non-redundant marking. To decide among these alternatives, we will consider the obligatorily transitive "be-" derivations of German talk verbs, such as "besprechen".
After a discussion of the data in terms of redundant and non-redundant encoding, we will present an LRS analysis that makes use of the techniques known from the treatment of operators at the syntax-semantics interface. [Slides]
Abstracts Contributed Lectures
Sebastian Löbner Why frames
The talk briefly introduces the new Collaborative Research Centre (CRC/SFB) 991 “The Structure of Representations in Language, Cognition, and Science”, financed by the Deutsche Forschungsgemeinschaft and started at Düsseldorf University in July 2011. The CRC presently comprises 16 research projects from semantics, computational linguistics, psycho- and neurolinguistics, philosophy, neurology, and psychiatry. The central and overarching objective of the CRC is the development of a general frame theory of con- cepts. This is attempted by developing a formal theory of frames, based on Barsalou's notion of frame, as a general format for cognitive representations. The representations investigated include linguistic representations in lexical and compositional semantics and syntactic representations from a cognitive point of view; we further investigate concept formation and operation at the psychological and neurological level, as well as institu- tional concept formation and concept shifts in science. The focus of the CRC is on semantics and computational linguistics. For semantics, it aims at grounding composi- tion in decomposition, i.e. explicit lexical meaning representations; those in turn are to be grounded in cognition. If successful, the approach will yield an information preserving approach to composition yielding rich structured representations of propositions. [Slides]
Wiebke Petersen Towards a radically attribute-centered perspective on frames
According to Barsalou (1992), frames, understood as recursive attribute-value structures, “provide the fundamental representation of knowledge in human cognition”. The attributes in a concept frame are the general properties or dimensions by which the respective concept is described. Their values are concrete or underspecified specifications. Barsalou’s frames are recursive in the sense that the value of an attribute can itself be further described by attributes. Being motivated primarily by empirical research, Barsalou’s focus in developing his frame theory was not on providing a formal theory. In recent years, based on the frame hypothesis that all representations in the human cognitive system correspond to frames, a new interest has developed towards a frame theory that is cognitively adequate and empirically founded as well as formally rigid (cf. Löbner 2011, Petersen & Werning 2007). [Continued]
Tanja Oswald Frames in category theory
This talk gives a representation of frames in category theory. All of the con- stituents of a frame — from the underlying universe of objects with thei attributes to the frame itself including constraints on its values — can be expressed in cate- gories. Categories are employed in mathematics to regard structures from a very abstract point of view (MacLaine, 1998). Thus, they can be used to examine the structural properties of frames. In short, categories are given by a class of objects and by the morphisms between all pairs of objects. If a morphism does not uniquely determine its pair of objects we speak of a pre-category. [Continued]
Thomas Gamerschlag, Wiebke Petersen and Stroebel Liane Sitting, Standing, and Lying in Frames: a frame-based approach to stative verbs of location
Stative verbs of location (SVLs) such as stand, sit, and hang encode gestalt properties and positio positional information, which are perceived by means of cognitive modules such as gestalt recognition and spatial orientation. Like spatial prepositions and dimensional adjectives, stative verbs of location are thus an excellent object of the investigation of cognition and language. The properties that are relevant to the choice of a specific SVL in German were established, among others, by Berthele 2004, established, Kaufmann 1995, Kutscher & Schultze Berndt 2007, and Serra Borneto 1996. These properties i Schultze-Berndt include (i) the way the localized object is kept in its position (e.g. support from below in the case of e sitzen ‘sit’ and support from above in the case of hängen ‘hang’), (ii) the state of matter of the su supporting medium (e.g. schwimmen auf ‘be afloat on’ versus liegen auf ‘lie on’) and (iii) the orient e orientation of the most prominent object axis (e.g. die Leiter steht ‘the ladder is standing’ versus die Leiter liegt ‘the ladder is lying’). Moreover, properties of the localized object such as animacy determine ). the choice of the SVL. Kaufmann (1995) proposes an analysis in which these properties are expl explicitly implemented as conjuncts in predicate logic representations. Following Lang ( ly (1989), she assumes that part of the spatial requirements that are imposed by the SVL on the localize object is localized captured in object schemata. [Continued]
Laura Kallmeyer and Rainer Osswald Syntax-Driven Semantic Frame Composition in Lexicalized Tree Adjoining Grammars
Semantic frames have been established as an expressive way to capture detailed aspects of meaning. However, so far, they are mainly used to describe the meanings of single lexical items. This paper concentrates on frame-based semantic composition and the interaction with syntactic operations. We propose a framework that integrates Lexicalized Tree Adjoining Grammars (LTAG) with Frame Semantics. There are two reasons for choosing LTAG in the context of semantic frame composition: Firstly, the elementary trees in LTAG represent entire subcategorization frames, which facilitates the linking of the syntactic components and the semantic frame components. Secondly, the underlying ``metagrammatical'' specification of an LTAG allows a strong factorization of the syntactic and semantic information of elementary trees and thereby enables to capture the specific meaning contributions of fragments of constructions.