How Many are many? Exploring Context-Dependence of few and many with Probabilistic Computational Models

DSpace Repositorium (Manakin basiert)


Dateien:

Zitierfähiger Link (URI): http://hdl.handle.net/10900/77197
http://nbn-resolving.de/urn:nbn:de:bsz:21-dspace-771976
http://dx.doi.org/10.15496/publikation-18598
Dokumentart: Dissertation
Erscheinungsdatum: 2017-07-20
Sprache: Englisch
Fakultät: 5 Philosophische Fakultät
Fachbereich: Allgemeine u. vergleichende Sprachwissenschaft
Gutachter: Jäger, Gerhard (Prof. Dr.)
Tag der mündl. Prüfung: 2017-07-19
DDC-Klassifikation: 150 - Psychologie
400 - Sprache, Linguistik
420 - Englisch
Schlagworte: Semantik , Pragmatik , Experiment , Stochastisches Modell , Kontext , Bayes-Verfahren , Erwartung
Freie Schlagwörter:
few
many
semantics
pragmatics
context-dependence
prior expectations
computational modeling
Bayesian inference
Lizenz: http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=de http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=en
Gedruckte Kopie bestellen: Print-on-Demand
Zur Langanzeige

Inhaltszusammenfassung:

Diese Dissertation untersucht die Kontextabhängigkeit der Wörter /few/ (`wenige') und /many/ (`viele') und die Interaktion zwischen deren linguistischer Bedeutung und statistischem Weltwissen. Konkret wird eine Theorie von Fernando und Kamp (1996) vorgestellt, die präzise Vorhersagen darüber macht, wie die Informationen aus dem Kontext in the Semantik integriert werden. Diese Theorie nimmt an, dass die sogenannte "Überraschungs-Lesart" von /few/ und /many/ ausdrückt, dass eine Zahl oder Proportion kleiner oder größer ist als erwartet. A priori Erwartungen im jeweiligen Kontext werden als Wahrscheinlichkeitsverteilungen P_E über natürliche Zahlen oder Proportionen formalisiert. Welche Zahlenwerte als /few/ oder /many/ gelten, wird bestimmt indem feste, kontextunabhängige Grenzwerte theta_few and theta_many auf die kumulierten Wahrscheinlichkeiten dieser Verteilungen angewendet werden. Mit anderen Worten, /few/ und /many/ enthalten eine feste Kernbedeutung, die erklären kann, warum Sprecher und Hörer so erfolgreich mit diesen kontextabhängigen Begriffen kommunizieren und wie Kinder ihre Verwendung erlernen können. Fernando und Kamps (1996) Theorie wird getestet indem sie in ein probabilistisches Modell übersetzt wird, welches die Grenzwerte als latente Parameter betrachtet. Deren Werte können nicht gemessen werden, sondern sie werden mit Bayesianischer Inferenz basierend auf experimentellen Daten geschätzt. In mehreren Versuchsreihen werden a priori Wahrscheinlichkeiten im jeweiligen Kontext und die Produktion und Interpretation von Sätzen mit /few/ und /many/ gemessen. Insbesondere wird die kardinale und die proportionale Lesart betrachtet, sowie die Modifikation von /few/ und /many/ mit /surprisingly/ (`überraschend') und /compared to/ (`im Vergleich zu') Konstruktionen, die die "Überraschungs-Lesart" explizit markieren.

Abstract:

This dissertation investigates the context-dependence of the quantity words /few/ and /many/ and the interaction between their linguistic meaning and rich statistical world knowledge. Concretely, one theory by Fernando and Kamp (1996) is explored which makes precise predictions about how contextual information might be integrated in the semantics. This theory assumes that the "surprise reading" of /few/ and /many/ expresses that a number or a proportion is lower or higher than expected. Context-dependent prior expectations about a relevant quantity are formalized as probability distributions P_E over cardinalities and the cardinalities which count as /few/ or /many/ are determined by applying fixed, context-independent thresholds theta_few and theta_many to the cumulative density mass of these distributions. In other words, /few/ and /many/ comprise a stable core meaning, which explains why speakers and listeners manage to successfully communicate with these context-dependent expressions and how children can acquire proficiency in their use. Fernando and Kamp's (1996) theory is tested by couching it in a probabilistic model of language use in which the threshold parameters are treated as latent parameters. Their values cannot be directly observed, but are estimated based on experimental data by applying Bayesian inference. In several series of experiments prior expectations are elicited and the production and interpretation of sentences with few and many are measured. In particular, the cardinal and the proportional reading of the quantity words is examined as well as the effect of overtly marking surprise with adverbs like /surprisingly/ or /compared to/ constructions.

Das Dokument erscheint in: