📄 The Centre and Periphery of Discourse Connectives

Summary

Discourse connectives have no agreed definition, and authors use competing labels for them. This paper treats them as a category with a centre and a periphery, based on discourse annotation of Czech newspaper texts in the Prague Dependency Treebank (PDT, about 50 thousand sentences).

The centre consists of primary connectives. These are mostly one-word, lexically frozen conjunctions and structuring particles whose main role is linking text units. They were annotated earlier in the Prague Discourse Treebank 1.0.

The authors concentrate on the less studied periphery: secondary connectives, also called AltLexes. These are free collocations or fixed phrases built around autosemantic words, such as nouns meaning "reason" or verbs meaning "explain". The authors annotated them manually, with some candidates found automatically. Secondary connectives prove more flexible than primary ones. They can take a nominalized second argument, form a separate sentence, or sit syntactically above the second argument.

A universality criterion keeps context-independent forms (because of this) as connectives and excludes context-bound ones (because of this increase). The result is a corpus-grounded way to draw the boundaries of the category, which the authors consider applicable beyond Czech.

Key points

• Data: PDT 2.5 (Bejček et al., 2012). Poláková et al. (2012) released the 2012 annotation of primary connectives with verbal arguments as version 1.0 of the Prague Discourse Treebank, which covers 20,255 such expressions.
• The new annotation of secondary connectives adds 1,161 relations to PDT:
  • 924 have verbal arguments and can often be swapped for a primary connective (e.g. z tohoto důvodu ↔ proto "therefore").
  • 237 have a nominalized second argument (e.g. důsledkem je "the consequence is"). The authors see this pattern as typical of written language.
• Primary connectives account for 95 % of explicitly signalled discourse relations in PDT, which is why they form the centre of the category.
• Universal secondary connectives (a preposition plus an anaphoric "this") are described as close to primary ones and as candidates for grammaticalization. English therefore (from there + fore) is given as a historical parallel.
• The approach follows the Penn Discourse Treebank view of connectives as two-argument expressions. Part-of-speech membership is rejected as a strict criterion for what counts as a connective.