Online publications

The Processing of Information Structure in SYNPHONICS

File(s): postscript (2900 KBytes) or gzipped postscript (294 KBytes)
Author(s): Carsten Günther , Claudia Maienborn, Andrea Schopp
Language: english
Page(s): 20
Abstract: The paper sketches an integrated view on processing information structure in a computational linguistic model of language production. The approach presented here extends from computing the information structure of an utterance out of a conceptual structure and a relevent contextual embedding to its corresponding prosodic realisation. Apparently conflicting requirements of structurally oriented theories of information structure on one hand and incrementally as on mayor processing property on the other hand are shown to be reconcilable within the language production system advocated here.
Conference: Interdisciplinary Conference in Celebration of the 10th Anniversary of the Journal of Semantics "Focus & Natural Language Processing", 1994, Schloß Wolfsbrunnen

Incremental computation of information structure and its empirical foundation

File(s): postscript (411 KBytes) or gzipped postscript (88 KBytes)
Author(s): Carsten Günther , Andrea Schopp Soenke Ziesche
Language: english
Page(s): 25
Abstract: This paper describes the incremental processing of information structure in a computational linguistic concept-to-speech system and its empirical evaluation through a comparison to empirically collected human speech data. The approach presented here shows that the SYNPHONICS result of computing the information structure of an utterance from conceptual structure and a relevant contextual embedding to its corresponding prosodic realization is consistent with the data collected in experiments according to HCRC Map Task guidelines.
Conference: 5th European Language Generation Workshop, 1995, Leiden

Pronominalisierung

File(s): postscript (234 KBytes) or gzipped postscript (71 KBytes)
Author(s): Andrea Schopp
Language: german
Page(s): 23
Abstract: In diesem Beitrag wird ein Modellierungsvorschlag für die Erzeugung von Pronomen im Rahmen des psycholinguistisch fundierten Sprachproduktionsmodells SYNPHONICS vorgestellt. Die psycholinguistische Ausgangsbasis des Systems kommt in den Annahmen zum inkrementellen Prozeßverlauf und zur strikten Modularität der konzeptuellen und sprachlichen Verarbeitungskomponenten zum Ausdruck. Es wird gezeigt, welche Probleme im Verlauf der Verbalisierung von Pronomen zu bewältigen sind und wo diese im Prozeßverlauf lokalisiert werden müssen. Die Zielsetzung des Beitrags besteht darin, eine durchgängige Modellierung der an der Verbalisierung von Pronomen beteiligten konzeptuellen und sprachlichen Komponenten vorzustellen und damit die Abhängigkeiten und das Zusammenspiel der verschiedenen Ebenen sprachlicher Strukturbildung als Folge einer charakteristischen konzeptuellen Ausgangskonstellation aufzuzeigen. Damit werden die auf den verschiedenen Ebenen wirkenden Prozesse zur Erzeugung von Pronomen durch eine computerlinguistische Modelliierung im Rahmen eines Sprachproduktionssystems in einen einheitlichen Funktionszusammenhang gestellt, der im Gegensatz zu herkömmlichen linguistischen Klassifikationsrastern eine Erklärung für den Zusammenhang zunächst heterogen wirkender Phänomene im Bereich der Pronominalisierung bereitstellt.

Adjuncts in HPSG

File(s): postscript (119 KBytes) or gzipped postscript (32 KBytes)
Author(s): Bernd Abb, Claudia Maienborn
Language: english
Page(s): 15
Abstract: Within linguistic theory and its computational modelling very little effort is spent on a theory of adjuncts. In current grammatical frameworks like HPSG, adjuncts are treated as behaving basically either as like arguments of a functor or like functors. Both strategies mix lexical information with purely combinatorial information within the lexicon and are therefore unsuitable to cope with the combinatorial variability that is characteristic for adjuncts. We present an alternative approach for adjuncts in HPSG which strictly distinguishes lexical and combinatorial parts of information and therefore avoids the existing difficulties of the HPSG analysis. The main emphasis is put on the development of a compositional semantics for adjuncts in comparison with semantic composition of arguments and copula constructions. Locative adjuncts are used for the illustration of our proposal.
Conference: KONVENS 94, Wien

Towards a Compositional Semantics for Locative Modifiers

File(s): postscript (114 KBytes) or gzipped postscript (36 KBytes)
Author(s): Claudia Maienborn
Language: english
Page(s): 24
Abstract: Within linguistic theory, very little effort has been devoted so far to a comprehensive theory of the syntax and semantics of modifiers. It is predominantly assumed that modifiers make a constant semantic contribution to the meaning of the constituent that is modified, irrespective of any conveivable syntactic differentiations. This account cannot cope with a multitude of empirical data and should therefore be replaced by a theory of modification that pays more attention to the exact nature of a given syntactic configuration and its impact on semantic composition. The approach advocated here is illustrated with locative modifiers of verb phrases. It will be shown that these expressions have a large range of meaning variability. Using German data, it is argued that differences in meaning can be correlated with differences in syntactic structures. On this basis, a compositional semantics for locative modifiers is proposed.
Conference: Salt 94

Formalization of Context within SYNPHONICS and Computations based on it

File(s): postscript (155 KBytes) or gzipped postscript (60 KBytes)
Author(s): Soenke Ziesche
Language: english
Page(s): 16
Abstract: This paper deals with a computational linguistic approach to the modelling of language production. In particular it addresses the fact that a single propositional content can be uttered in various ways depending on the context of the utterance which has to be defined. According to this three tasks arise: First of all, an analysis of the correlation between contextual parameters and their effects has to be made. Then, an adequate formal representation of the context has to be supplied. Finally, the computation of these effects has to be modelled on this base. In the following the language production system SYNPHONICS is introduced which handles the tasks described above. The conceptual structure of the proposition as well as detailed representation of the relevent context serve as input to a module by the name Semantic Encoder. Using this input, various settings are computed on the level of semantics leading to different syntactic or phonological structures respectively different lexical access. Examples for such phenomena are focus-background-structure, ellipsis, topic-comment-structure, linearization, referential specification and selection.
Conference: Workshop Context in Natural Language Processing, IJCAI-95, Montreal