• Keine Ergebnisse gefunden

We base the definition of the thematic roles supported in our model implementation on the lists of thematic roles in Dowty (1989, p. 69) and L¨obner (2003, p. 174).

We deliberately avoid a discussion of the granularity and appropriateness of these thematic roles by choosing comparatively general definitions for a set of roles that is widely accepted as standard (cf. also Ferretti et al., 2001). The purpose of our model implementation is not to demonstrate the correctness or appropriate-ness of specific thematic role definitions. Rather, we wish to show with our model that inherently semantic generalisations over verbal argument slots can be used as constituents of semantic representations that mediate the cross-modal interaction between non-linguistic and linguistic modalities. We acknowledge that the thematic role definitions in Table 5.1 are to some extent arbitrary and not sharply delineated – as are all thematic role definitions that attempt to capture semantic generalisa-tions over verbal arguments. As long as the assumption that underlies the concept of thematic roles remains unchallenged, namely that semantic generalisations over verbal argument structures are indeed possible, the precise semantic delineation of these roles or the labels attached to them have no fundamental impact on the va-lidity of our model.2

These role definitions deserve a few further comments: Our role definitions do not build on a differentiation between events, actions, states or processes. Instead, we adopt the more embracing term situation as used in the situation semantics of Barwise and Perry (1983, pp. 7) to subsume all of the aforementioned notions. A situation type or concept hence is taken to denote an abstract constellation by which participating individuals with certain properties are related to each other. An ex-ample of a situation concept is bark which only involves a barker as participant.

1Strictly, the well-formedness rules for context model representations also contribute to the fulfilment of this requirement. These are described in Section6.4.

2Clearly, the choice of thematic roles considered does correlate with the constraints defined in the role-assigning grammar. A more fine-grained differentiation in role definition will also require more finely differentiated role-assigning constraints in the grammar. This aspect, however, does not question the validity of our model as such.

AGENT The participant specified as doing, causing, having, being or experiencing something in a situation.

Example: He is eating an apple.

He−−−−−−−−−−−→isAGENT f or eat

THEME The participant that something is happening to in the situation or that is immediately affected by the situation.

Example: He is eating an apple.

apple−−−−−−−−−−−→isTHEME f or eat

RECIPIENT The participant that the result of the situation is directed to.

Example: She gave him a book.

he−−−−−−−−−−−−−−→isRECIPIENT f or give

INSTRUMENT The entity enabling or facilitating the occurrence or progress of a situation.

Example: He opened the door with a key.

key−−−−−−−−−−−−−−−−→isINSTRUMENT f or open

OWNER The entity extending any sort of ownership or belonging relation towards another participant.

Example: She has Kirsa’s book.

Kirsa−−−−−−−−−−−→isOWNERf or book

COMITATIVE The entity that physically or figuratively accompanies another participant.

Example: He went to the cinema with her.

she−−−−−−−−−−−−−−−→isCOMITATIVE f or he

Table 5.1: Overview over the thematic role definitions in our model.

The situation in which a participant isbarked at involves an enitybarked-at in addi-tion to the barker and thus constitutes a related, yet different, situation bark.at. Barwise and Perry consider the instances of barking observed in the real world to be instantiations of the abstract situation types barkand bark.at.

By including the component of experience in the role definition of AGENT, we incor-porate the aspects which Dowty (1989) lists for the separate role of EXPERIENCER

into the role definition of AGENT. Our definition of THEME is in line with that of L¨obner (2003) and treats the rolesTHEMEand PATIENT as semantically equivalent.1 For terminological clarity, we henceforth differentiate between entities and partici-pants. By the term ‘entity’ we denote anything that takes a thematic role in the context of a situation. We use the term ‘participant’ to refer specifically to those entities in a situation that engage in a direct and semantically mandatory thematic relation with an instance of the situation concept. We therefore denote entities taking an AGENT, RECIPIENTor THEME role as participants while entities taking an

OWNER, COMITATIVE or INSTRUMENT role are considered situation entities but not participants.

We limit our modelling scope to these thematic roles since they are sufficient for the study of a number of interesting and notoriously difficult-to-parse syntactic phenomena such as PP-attachment or subject-object ambiguity of German plural nouns. Also, this set of thematic roles results in a manageable number and complex-ity of hand-written constraints in the role-assigning grammar. The role-assigning grammar used in the experimental runs reported in Chapters8to11contains about 140 individual constraints — as opposed to approximately 1050 active constraints in WCDG1’s large-coverage syntactic grammar for German.

In principle, our model permits to extend or modify the list of supported thematic roles. Any extension or modification that is not simply a reduction of the set of thematic roles supported may, however, incur the need to add or change constraints in the role-assigning grammar.2 Thematic role assignment in our model is subject to the following modelling decisions and constraints:

• Thematic dependencies originate from the role filler.

• The AGENT dependency is assigned on its own level of analysis.

• The THEME dependency is assigned on its own level of analysis.

• The dependenciesRECIPIENT,OWNER,INSTRUMENTandCOMITATIVEare modelled as mutually exclusive and are assigned on a separate level of analysis.

• Thematic dependencies under the same regent are unique.

• The verb-centred semantic dependenciesAGENT,THEMEandRECIPIENTcan only be assigned to verb forms that have a corresponding semantic valence.

1Jackendoff (1990, p. 129) vehemently argues against this practice. Given the widely acknowledged fuzziness in defining thematic roles, we choose to disagree with his point of view.

2An arbitrarily large extension of the list clearly is not possible in our model. Every implementation is limited, not only by its algorithmic design, but also by the capabilities of the hardware on which it is executed. Unless stated otherwise, we assume that limitations arising from the hardware environment can be neglected in the assessment of the validity of our model.

• A verb’s semantic valence must be saturated by the assignment of the cor-responding semantic dependencies.

• Attachment restrictions for the semantic dependencies have been formulated based on the dependant’s and regent’s part of speech rather than based on genuinely semantic criteria. The role-assigning grammar imposes no selectional restrictions as a function of the role fillers’ conceptual category. Our model hence fails to meet Requirement R23 which demands meaning-based selectional restrictions for thematic role fillers.