[print]

Institut für Linguistik: Anglistik

Graduiertenkolleg - SS 2006

27.07.06 :: Edit Doran
The Semantics of the Middle Voice

It is known from the typological literature that languages which overtly mark transitivity alternation vary in their choice of the marked alternant. The transitive verb may be marked as causative, or the intransitive verb may be marked as middle, a marking often identical to the marking of lexical reflexivity. In view of the typological observation that causatives and middles often mark a single alternation, it is compelling to look for a unified analysis. I present such a unified analysis, based on Semitic template morphology. In the Semitic languages, causative and middle verbs alike are derived by particular templates. The Semitic templates denote voice (of which middle is one possible value) and a thematic dimension which I call agency (of which causative is one possible value). The analysis of middle verbs in terms of a middle template is not equivalent to an analysis in terms of argument reduction. Under the latter approach, the identity found cross-linguistically between the middle and the reflexive morphemes remains a mystery. Moreover, argument reduction faces a serious semantic problem, i.e. that there is no way to 'celiminate' the semantic contribution of the transitive verb's external argument.

 

05.07.06 :: Lucie Barque
A hybrid lexicon for the treatment of polysemy

Our study concerns the modelization of polysemy links from an enumerative lexicon of French, the BDif (for Base de Difinitions).

First, we will introduce the BDif definitions whose main feature is the formal description of the internal structuring of lexical meanings. Having defined a polysemy domain, i. e. the set of lexical units in polysemy relations, we will then present polysemy patterns extracted from pairs of Bdif definitions.
Finally, we will show how these lexicographic descriptions, definitions, and polysemy patterns are implemented in LKB (Linguistic Knowledge Building System), respectively as typed feature structures and pairs of underspecified typed feature structures.

 

20.06.06 :: Horst Mueller
Zur kortikalen Repräsentation des Lexikons: Die Sonderstellung der Eigennamen

Zu den wichtigen Forschungsfragen der gegenwärtigen Kognitionswissenschaft gehört die Untersuchung der kortikalen Repräsentation von lexikalischem und konzeptuellem Wissen. Erkenntnisse zur Binnenstruktur des Lexikons und zur kognitiven Realisierung von Sprache sind beispielsweise für Theorien zum Spracherwerb, für die Sprachtherapie von Aphasikern oder für die Konstruktion sprachverstehender Systeme innerhalb der KI-Forschung von Bedeutung.
In dem Vortrag werden Untersuchungsergebnisse zur Binnenstruktur der Nomen, speziell die Sonderstellung der Eigennamen (Nomina propria) vorgestellt. Nach einer kurzen Begründung für die seit über 2000 Jahren theoretisch begründete und kontrovers diskutierte Sonderstellung der Eigennamen, soll die Rolle der "Identifikatoren" auch aus biologischer Sicht diskutiert werden. Weiterhin werden neuropsychologische Befunde vorgestellt, die bei Patienten mit bestimmten Hirnschäden eine selektive Beeinträchtigung ausschließlich bei der Verarbeitung von Eigennamen oder aber von Gattungsbezeichungen zeigen. Im Hauptteil des Vortrages werden Ergebnisse neurolinguistischer Studien zur Verarbeitung von Eigennamen vorgestellt, bei denen psycholinguistische und neurowissenschaftliche Methoden eingesetzt wurden. Befunde zu Verarbeitungsunterschieden zwischen Gattungsbezeichnungen, Personenvornamen, Ortsnamen und Warennamen werden diskutiert.

 

29.06.06 :: Bernard Jacquemin
Word sense and textual meaning:
Dictionary- and corpus-oriented approaches for question answering

Word sense specification in context is a major issue in question answering, not only to identify the meaning of each utterance, but also to reformulate them as much as possible, keeping the original meaning of the sentence.
This talk presents two ways to use the word senses to find, in a textual base, one ore more answers for a question asked in natural language (French). The first one uses electronic dictionaries, morpho-syntactic analysis and word sense disambiguation to build a structure of the texts, which stores all indentified or reformulated pieces of information.
The second one mixes statistic and linguistic tools to process the corpus to question. This process makes it possible to build a specilized lexico-semantic ressource for the corpus, and to automatically identify the meaning of each word in context. The matching process is done through a projection in a multidimensional semantic space.

 

22.06.06 :: Fabienne Martin
Psych-verbs and agent oriented adverbs

As is well-known, only a subset of French causative psych-verbs are compatible with agent-oriented manner adverbs like prudemment 'cautiously' or patiemment 'patiently':

  • (1)
    • a. Marie l'a prudemment séduit.
      Marie cautiously seduced him.
    • b. Marie l'a prudemment attiré.
      Marie cautiously attracted him.
  • (2)
    • a. Il les a patiemment encouragés.
      He encouraged them patiently.
    • b. Il les a patiemment stimulés.
      He stimulated them patiently.

Traditionally, the problem illustrated in (1b)-(2b) is analysed as follows: in the case of stimuler-verbs, S (the entity denoted by the subject) cannot act with sufficient volition, intent and control to provoke the expected reaction of the Experiencer (van Voorst, Ruwet), or S is an abstract entity, thus unable to act (Bouchard). The problem of the first explanation is that it doesn't explain why stimulate-verbs, contrary to uncontroversially non-agentive verbs like suffer or know, can often be used in constructions which are also said to require the presence of an Agent:

  • (3) C'était stupide de la part de Marie de souffrir/ intelligent de la part de Marie de connaître la réponse.
    It was stupid of Mary to suffer/ clever of Mary to know how to answer.
  • (4) C'était intelligent de la part de Marie de les stimuler avec un peu plus d'argent.
    It was clever of Mary to stimulate them with a bit more money.

The problem of the second explanation is that it doesn't explain why the subject of stimulate-verbs can corefer with the implicit subject of an agentive gerund, which is obviously an Agent:

  • (5) En faisant cela, Marie a stimulé tout le monde.
    In doing this, Mary stimulated everybody

I will argue that the problem of (1b)-(2b) comes from the fact that stimulate-verbs presuppose the action of S, and only assert the psychological reaction of the Experiencer. It will be shown that on this point, attirer-verbs are similar to what Zybatow (2000) calls right achievement verbs, like persuader 'persuade' and to what I will call strictly strong accomplishment verbs like convaincre 'convince/persuade' or guérir 'cure':

  • (6) Il m'a patiemment persuadé/convaincu d'accepter l'offre.
    He patiently persuaded/convinced her to accept the offer.
  • (7) Il m'a prudemment guérie.
    He cautiously cured me.

I will provide an event-semantics analysis of stimulate-verbs and convaincre-verbs in the line of the analysis of right achievement verbs by Malink (2005) and Pinon (2005).

 

01.06.2006 :: Agnes Bende-Farkas
TWO NEGATIONS, ONE SENTENCE (AND ONE FOCUS)

This talk will offer some hypotheses (and a semantic analysis for simple cases) concerning the puzzles presented by Hungarian sentences like (1).

(1)

  • Ja'nos nem [a Hamletet]_F nem olvasta (hanem a Ro'meo' e's Ju'lia't) `It was not Hamlet John has not read (but Romeo and Juliet)'

(1) contains preverbal Focus and two occurrences of the negative particle `nem'. These particles correspond to two separate, independent instances of semantic negation, just as in the corresponding English it-cleft. That is to say, (1) is not a double negation sentence (it does not mean that John did read Hamlet). The first puzzle is how two independent negations are possible in the same (syntactically simple) clause.

The second puzzle is how (1) is possible in a strict Negative Concord (NC) language like Hungarian, where sentences can contain any number of n-words (provided they also contain `nem') that all contribute to ONE negation.

(2)

  • a. Nem la'tott senki semmit sehol (= It is not the case that anyone saw anything anywhere)
  • b. *Senki la'tott semmit sehol (Intended: same as (2a), bad because of absence of `nem'. Becomes grammatical with insertion of `nem'.)

In the talk I would like to discuss the following issues relating to (1), negation and Focus in Hungarian:

  • Sentences like (1) are possible only if the Focus position is filled. The guiding assumption (A) for a semantic analysis is that Focus `splits' sentence material into a presupposition and an assertion part. Preverbal `nem' contributes to the assertion and postverbal `nem' to the presupposition. (So (1) presupposes that there was something John did not read, and asserts that that something was not Hamlet.)
  • A consequence of (A) is that variants of (1) that also involve n-words will have to be explained by a theory of presupposition accommodation. These cases open up a separate field of investigation (the interaction of presuppositions with quantifiers or scope-bearing n-words) which will not be investigated here. (The beginnings of a discussion can be found in the short paper Focus and Negative Concord in Hungarian, downloadable from http://www.ims.uni-stuttgart.de/~agnes/papers.html)
  • The talk will focus on a problem involving syntax and the syntax--semantics interface: is the licensing domain of negation determined solely in the syntax, or can it be determined (as in these Hungarian cases) at the level of semantic representation, or at Logical Form?

In sentences like (1) Focus creates two SUBSENTENTIAL domains for negation. `Nem' and n-words are licensed in the post-Focus domain independently from the pre-Focus domain, just as they are licensed in subordinate (finite or non-finite) clauses.

The simple Hungarian sentence (1) thus resembles an English it-cleft in another way: the post-Focus domain is comparable to the English relative clause where the licensing of negation is concerned.

At this point I can see two alternative solutions to this problem.

  1. The two `domains' correspond to assertion and presupposition, i.e. to two separate `compartments' in the semantic representation of the sentence. One can conclude from this that negation and n-words can be licensed at this level of of the grammar. To my knowledge this is proposal has not been made before (consensus on these matters says that this is decided in the syntax).
  2. The syntactic structure of Hungarian sentences with Focus is that of a complex clause. That is, there is a component of the syntax of Hungarian Focus responsible for the creation of two syntactic domains and for introducing a `barrier' that prevents the interaction of pre-Focus and post-Focus `nem' and n-words. An argument in favour of (ii) is that no other presupposition triggers are like Hungarian Focus where the licensing of negation is concerned.

 

18.05.06 :: Jaye Padgett
Sound change, allophony, and perception

Padgett and Zygis (to appear) analyzed a rather mysterious unconditioned sound change that affected Polish (and Russian probably independently) several hundred years ago: a series of palatoalveolars became retroflex. This change is hard to understand, since it was not assimilatory, and retroflexes are regarded as more marked than palatoalveolars. We argued that it was motivated by perceptual dispersion: acoustic and typological data support that claim that retroflexion was triggered by the presence in the Polish inventory of a relatively new series of alveolopalatals. A contrast between alveolopalatals and palatoalveolars is perceptually dubious, and retroflexion improved this state of affairs.

In this talk (also in collaboration with Marzena Zygis), I present results from perception experiments that support the historical account, and also allow us to draw general conclusions about the 'perceptual map' of a complex set of sibilants. An AX discrimination task involving Polish sibilants as stimuli was carried out on Polish and American English listeners. Some of the results are likely due to the different phoneme categories that Polish and English speakers have (Polish having the richer sibilant inventory). But other results, particularly from reaction time data, are similar across the languages and are best accounted for by a language-independent perceptual map.

 

04.05.06 :: Jonny Butler
Tense and Sequence of Tense

I examine Sequence of Tense (SOT) phenomena in English including the so-called simulataneous reading of Past embedded under Past (1) and the double-access reading of Present embedded under Past (2).

  1. Mary said that she was pregnant = Mary said "I am pregnant"
  2. Mary said that she is pregnant = Mary said "I am pregnant", and Mary is still pregnant now

There are two issues to deal with: one is that a restriction on these constructions is often stated that the embedded predicate must be stative; the other is that there is a range of approaches to the simultaneous reading in (1), some claiming that the embedded tense really is Past as it appears, some that despite appearances it is really present, and some that it is null.

As to the first point, I examine the data to sharpen up the definition of what the restriction on the embedded predicate really is. This allows clarification on the second point, since the restriction on interpretation turns out to apply also to non-embedded Present clauses in English, obviously pushing a treatment where the embedded tense is Present.

I then present a theoretical treatment based on the syntactic approach to tense developed by Stowell (1996) and Demirdache & Uribe-Etxebarria (2001) which explains the observed restrictions.

 

27.04.06 :: Regine Eckardt
Wh-Phrases as inherently focussed objects

In this talk, I propose a semantic spell-out for the claim that wh-phrases be "inherently focussed". I suggest that in languages like German, wh-phrases have sets of alternatives as their ordinary semantic value. These alternatives license an alternative-semantic interpretation procedure like the one first elaborated in Rooth (1985), yet as part of the ordinary semantic interpretation. I will argue that focus semantic interpretation, even though formally identical, has to be kept strictly separate from the semantic processing of inherently focussed material. The analysis captures core (non)-intervention facts, wh-interpretation in situ, Reinhart's "philosopher" puzzle sentences, and offers an elegant basis to explain the typological closeness between wh-pronouns and indefinites. The ingredients of the analysis have been taken up in other recent work, specifically Kratzer/Shimoyama 2002 and Beck (2006). I will argue that the present analysis offers a very robust starting point that can be refined to capture their data while offering a solid core analysis for inherent focus.

 

26.04.06 :: Alain Polguère
The Structure of Lexical Systems

Two types of models for natural language lexica are presently available. First, there are dictionary-like models, that represent the lexicon by means of a huge text, which is a collection of much smaller texts called "dictionary entries." Each entry focuses on one word, or word sense. Second, we have net-like models (e.g. WordNet or FrameNet), that represent the lexicon as graphs made up of lexical entities (nodes) and lexical links (oriented arcs). Even though the current trend favors the net approach, most net-like databases can be criticized for taking a single-minded view of the lexicon. Usually, one specific range of phenomena is the target of description, generating lexical structures that are designed accordingly and offer very little flexibility when other lexical phenomena have to be taken into consideration.

In this talk, we present a new type of lexical structures of the net family, called "lexical systems." They are pure directed graphs, not constrained by any pre-established hierarchical structure. Additionally, lexical systems can take into account the potential fuzziness of lexical knowledge and are conceived of as organic, evolutional structures. We illustrate the approach using a lexical system that has been generated from the DiCo database of French semantic derivations and collocations. Rather than focusing on formal or implementation aspects of this work, we concentrate on the linguistic nature of lexical entities and lexical links that have been used to model data taken from the DiCo, with special attention paid to Meaning-Text lexical functions and to the DiCo system of semantic labels.