Abstract

The discourse structure of a (multi-sentential) text is primarily represented in terms of relational coherence, which explains how the text segments (sentences or clauses) are linked to each other by coherence relations. Coherence relations encode a meaning relationship (e.g., causal, additive, temporal) between two text segments, which represent propositions (facts, events or situations), speech acts (e.g., claim) or beliefs (of the speaker/writer). For example, in the sentence “John could not go to the party because he was busy with his work”, the two segments (clauses) represent two events which are linked by a Reason relation. Coherence relations are signaled by discourse connectives (e.g., although, if, thus, when) and also by many other textual features (e.g., parallel syntactic constructions as a signal of Contrast relations). In this talk, I briefly introduce the core theoretical constructs of discourse structure, and outline some important characteristics of relational coherence. Then, I discuss how coherence relations are indicated by relational signals, and report on my previous research on this topic. Finally, I present our present work on the Bangla discourse structure, and report on the development of a discourse-annotated corpus and a lexicon of Bangla discourse connectives.

Have you ever seen the visual representation of rapid speech? You can see it on any digital recording software. It is a continuous waveform without any helpful pauses to subdivide the whole into discrete units of sound or meaning. Ergo, speech segmentation from acoustic data is a very complicated task for machines. However, for a human child, it is effortless. A human mind will not only segment it into individual sounds that it categorically identifies, it will also have a rather good intuition of where to expect word boundaries even though the person might have never heard that language or know any words in it.

The knowledge that we humans use to do this, is studied under the rubric of phonological universals. It focuses on the compositional principles of the human sound system. In this talk we will discuss one of the major compositional principles of the human perceptual system, called sonority and see the typology of phonological grammars organized around this principle.