The systematic assumption of faithfulness violations in Optimality Theory implies an infinite space of candidates. Under the methodological principle of trying to explain as much possible through constraint interaction, control over this infinite space should be exerted by the constraints. Assuming the subsumption-based candidate denition of OT-LFG, the candidate space is indeed sufficiently structured to facilitate computational processing according to this principle. However, the parsing direction in the standard production-based optimization model is not subject to optimization, so for the parsing task, a decidability issue arises. Adopting a bidirectional optimization model is one way of solving this problem, but the required strong concept of bidirectional optimization may not be linguistically desirable. Other possible conclusions are discussed briefly.

This paper describes an architecture for exploiting implicit information about the grammar of the languages included in a parallel corpus. By initially applying statistical word alignment and defining an appropriate representation format for cross-linguistic structural correspondence, this implicit information can feed a system for bootstrapping grammars. The proposed architecture will be underlying in the new PTOLEMAIOS project. Dieses Papier beschreibt einer Architektur, mit der die implizit in Parallelkorpora enthaltene Information über die Grammatiken der beteiligten Sprachen ausgenutzt werden soll. Wenn vorab eine statistische Wortalignierung angewandt wird und ein geeignetes Repräsentationformat für die crosslinguistische Strukturkorrespondenz definiert wird, kann diese implizite Information in einem Bootstrapping-Ansatz zum Grammatiklernen verwertet werden. Die vorgeschlagene Architektur wird im neuen PTOLEMAIOS-Projekt zur Anwendung kommen. 1.

Creation of high-quality treebanks requires expert knowledge and is extremely time consuming.

A single speaker's production of variable outputs for the same input often reflects categorical generalizations across grammars. Variation in subject-verb agreement in English dialects and varieties has just this property. For example, previous studies of verb agreement in `nonstandard' English varieties have observed a reduction of variation with plural (vs. singular) subjects or negative (vs. armative) sentences; these are typologically marked grammatical contexts in which contrasts are often categorically

This paper describes an architecture for exploiting implicit information about the grammar of the languages included in a parallel corpus. By initially applying statistical word alignment and defining an appropriate representation format for cross-linguistic structural correspondence, this implicit information can feed a system for bootstrapping grammars. The proposed architecture will be underlying in the new PTOLEMAIOS project. Dieses Papier beschreibt einer Architektur, mit der die implizit in Parallelkorpora enthaltene Information über die Grammatiken der beteiligten Sprachen ausgenutzt werden soll. Wenn vorab eine statistische Wortalignierung angewandt wird und ein geeignetes Repräsentationformat für die crosslinguistische Strukturkorrespondenz definiert wird, kann diese implizite Information in einem Bootstrapping-Ansatz zum Grammatiklernen verwertet werden. Die vorgeschlagene Architektur wird im neuen PTOLEMAIOS-Projekt zur Anwendung kommen.

computational architecture for corpus-based studies of syntax 1 This paper describes a computational architecture for accessing implicit information about the grammar of the languages included in a parallel corpus and exploiting it in an Optimality Theory-style learning approach. Previous work on OT learning presupposes the existence of training data in which the underlying input has been annotated. This is an idealization that does not reflect the natural learning situation; and it also requires considerable effort to produce such training data for learning experiments with syntactic/semantic grammar models. In the proposed bootstrapping architecture, which will be underlying in the new PTOLEMAIOS project, the training data are sentences from a parallel corpus; manual annotations are only provided for a small set of seed sentences. The translations of the sentence into the other languages serve as clues for zeroing in on the assumed underlying meaning representations, which can be used as the input in OT-style learning. 1 Introduction: the goal for the PTOLEMAIOS project