Corpus-based Analysis of Parenthetical Reporting Clauses

In this paper, the syntactic properties of parenthetical reporting clauses in Dutch are investigated by means of a small corpus. It is shown that an analysis in which the quote is looked upon as the direct object of the reporting verb is inadequate. Therefore, an alternative analysis is proposed, viz. one in which the quote and the reporting clause are taken to be adjoined. Such an analysis, however, is not unproblematic. Here we discuss the two main problems with this analysis. 1 Introduction A construction that so far has not received much attention in automatic syntactic analysis is the parenthetical reporting clause (rc). The construction has been the subject of a number of theoretical studies including Ross (1973), Luif (1990) and Collins and Branigan (1997), but none of these came near a formalisation, let alone an implementation. For the purpose of this article the term`parenthetical clause' is deened as a linguistic entity that does not have any syntactic relation to the sentence it interrupts. Paren-thetical clauses have restrictions, though: they can only appear at certain positions in the main clause (they usually do not appear within the premod-iier of a noun phrase (NP) or between a preposition and its complement, for example) and not every construction is allowed to become parenthetical. An example of a parenthetical clause is given in 1): (1) He must, I thought, be a wonderful person. The parenthetical clausèI thought' does not fulll any thematic role of the main sentence, nor does one clause appear to be subordinated to the other. Nevertheless, example 1) is a syntactically correct sentence. The fact that the relation between the main sentence and the parenthetical clause is unclear implies that this relationship-and therefore these constructions-cannot be implemented in an automatic syntactic parser. In the AMAZON-parser, a surface based parser for Dutch, the parenthetical construction has not been implemented yet. In order to extend AMAZON with at least one variant of the parenthetical construction, a study was made of the variant that seemed easiest to identify: the parenthetical reporting clause. This study consisted of compiling a 50.000-word corpus with a relatively high proportion of these constructions and analysing these real-life observations. Unless stated otherwise, examples present in the paper originate from this