SCOPE OF THE SYMPOSIUM
The ability to engage in a dialogue has been trumpeted as a good indicator of general intelligent behaviour since Turing (1950). Philosophers such as Hamblin (1970), Brandom (1994) and Davidson (2001) can be said to have proposed various types of linguistic rationalism, the notion that linguistic abilities are a pre- or co-requisite for rationality: in Brandom's terms, the hallmark of rationality is the ability to take part in the game of “giving and asking for reasons”. Indeed, Deligiorgi (2002) already finds in Kant the notion that “rationality cannot be exercised by a solitary thinker” but depends on the communication of publicly criticisable judgments. The capacity to engage in a dialogue could very well be AI-complete, i.e., employ all the skills abilities that constitute human-level intelligence.
The year 2014 marks several significant anniversaries: one of them is the 20th year since the publication of Brandom's Making it Explicit, a large, complex and difficult work in the philosophy of language which Jürgen Habermas likened to Rawls' Theory of Justice in terms of its scope and importance within its field. It is fair to say that this work has so far had little direct influence on computational or formal approaches to language, though some partial formalisations have been offered by Lance and Kremer, Kibble and Piwek. This symposium will be loosely organised around various themes arising from Brandom's work, or questions provoked by it, though participants will not necessarily be expected to directly engage with his original texts. As noted, Brandom sees the game of “giving and asking for reasons” as central to human rationality or sapience, but it turns out that he has rather little to say about questions or any other speech acts apart from assertion. Brandom stresses the importance of shared “material” inferences for successful communication, though it is far from clear how this common background understanding could be encoded in a computer system.
TOPICS OF INTERESTSuitable subjects will include, but are not limited to:
- Inference in dialogue
- Commitments, norms, discourse obligations and dialogue games
- Intentionality: can discourse and dialogue be modelled without reference to mentalistic notions of intention and belief?
- Comparison of formalisms for discourse analysis: e.g. RST, SDRT
- Argumentation: analysis and representation of argument structure
- what is an appropriate response at a given point in a dialogue?
- what is the optimal ordering of propositions in a discourse?
- how should predicates, referring expressions and rhetorical relations be realised (verbally and/or non-verbally) so that the resulting utterance can be interpreted naturally and fluently?
- Annotation schemes for discourse relations
- Questions and answers:
- Does an account of questions presuppose a model of assertion (or vice versa)?
- Cognitive and computational models of question generation (QG)
- Question taxonomies
- Data collection and preparation for developing, training, and testing QG systems
- Annotation schemes and processes
- Evaluation of generated questions: metrics and methods (human, automatic, semi-automatic)