Department
of Linguistics - Simon Fraser University
Colloquium
Series
Ruhr-Universität Bochum, Germany
Department of Linguistics
Thursday, August 18, 2011
11:30 am, EDB 8651
Abstract
The nominal complement of a Preposition-Noun-Combination (PNC) contains a singular count noun, but no determiner, despite grammatical rules that require the presence of a determiner for singular count nouns. PNCs are sometimes called ‘determinerless PPs’ or ‘bare PPs’, but this is a misnomer, at least if elementary concepts of locality are considered. Examples from English and German are given below.
(1) |
on disk, from jail, at school | |||||||||||||||||||||||
(2) |
|
|||||||||||||||||||||||
The examples from German in (2b, c) show that the construction should be analysed as a collocation. In (2b), we find instances of adjectival modification, and (2c) shows an instance of postnominal complementation.
PNCs cannot be classified as a rare construction, yet speakers are reluctant to judge or coin new PNCs in isolation. For some time, they have even been classified as listed exceptions. Such a conclusion is however neither warranted from a typological perspective, nor from a computational of productive or regular processes. Typological studies show that the ‘irregular’ construction can be found abundantly in languages that ordinarily require singular count nouns to be accompanied by a determiner (cf. Himmelmann 1998), and they have also been shown as being regular, so that an approach in terms of ‘listing’ is barred (Dömges et al. 2007).
PNCs are also interesting from the perspective of foreign language learning, as a command of PNCs identifies an advanced level of language fluency. As long as grammars list PNCs as belonging to special registers, foreign speakers cannot learn them. Thus the identification of the determining properties of PNCs is not only worthwhile from the perspective of theoretical linguistics, but also needed to define practical grammars.
PNCs show an interesting bandwidth of properties if individual languages are compared. While analyses focussing on English (Stvan 1998, Baldwin et al. 2006) refer to possibly defective nouns (jail, school), and operations of semantic or pragmatic enrichments, recent analyses of German have shown that the construction is not restricted to peculiar noun types and also does not make use of pragmatic enrichment.
We present an analysis of PNCs by use of ‘annotation mining’. In annotation mining, large data sets receive automatic and manual annotations, and the annotations are fed into classifiers from statistical or machine learning. The result of the classification process is not used to predict ‘unseen’ cases but to identify discriminative features of the construction. Our analysis will focus on the preposition ohne (without).