Abstract
Full text discourse parsing relies on texts comprehensively annotated with discourse relations. To this end, we address a significant gap in the inter-sentential discourse relations annotated in the Penn Discourse Treebank (PDTB), namely the class of cross-paragraph implicit relations, which account for 30% of inter-sentential relations in the corpus. We present our annotation study to explore the incidence rate of adjacent vs. non-adjacent implicit relations in cross-paragraph contexts, and the relative degree of difficulty in annotating them. Our experiments show a high incidence of non-adjacent relations that are difficult to annotate reliably, suggesting the practicality of backing off from their annotation to reduce noise for corpus-based studies. Our resulting guidelines follow the PDTB adjacency constraint for implicits while employing an underspecified representation of non-adjacent implicits, and yield 62% inter-annotator agreement on this task.
Highlights
Empirical approaches for modeling discourse relations rely on corpora annotated with such relations, such as the Penn Discourse Treebank (PDTB) (Prasad et al, 2008), the RST-DT (Carlson et al, 2003), and the ANNODIS corpus (Afantenos et al, 2012)
This paper describes our experiments in annotating cross-paragraph implicit relations in the PDTB (Section 2), with the goal of producing a set of guidelines (Section 3) to annotate such relations reliably (Section 4) and produce a representative dataset annotated with complete sequences of inter-sentential relations
Matched arguments show an increase to 42% from 24% in Phase Two and there are fewer disagreements due to supra-sentential overlapping spans, which have reduced to 13% from 30% in Phase Two
Summary
Empirical approaches for modeling discourse relations rely on corpora annotated with such relations, such as the PDTB (Prasad et al, 2008), the RST-DT (Carlson et al, 2003), and the ANNODIS corpus (Afantenos et al, 2012). The PDTB is currently the largest of these annotated corpora and widely used for theoretical and empirical research on discourse relations. It does not provide exhaustive annotation of its source texts (Prasad et al, 2014). While the PDTB provides annotations of explicit inter-sentential relations within and across paragraphs, and of implicit relations between adjacent sentences within paragraphs, it ignores cross-paragraph implicit relations. Ex. (1) illustrates the problem in a PDTBannotated text, showing 6 sentences (S1-S6) in the first four paragraphs of a longer article. (Empty lines indicate paragraph boundaries.) While all annotation elements are not shown here, the key issue to note is that the relations of sentences S2 and S3 with the prior text are left unannotated because they are paragraph-initial sentences lacking any inter-sentential explicit connectives
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
