Non-Sentential Utterances in Dialogue: Experiments in Classification and\n Interpretation

Non-sentential utterances (NSUs) are utterances that lack a complete\nsentential form but whose meaning can be inferred from the dialogue context,\nsuch as "OK", "where?", "probably at his apartment". The interpretation of\nnon-sentential utterances is an important problem in computational linguistics\nsince they constitute a frequent phenomena in dialogue and they are\nintrinsically context-dependent. The interpretation of NSUs is the task of\nretrieving their full semantic content from their form and the dialogue\ncontext. The first half of this thesis is devoted to the NSU classification\ntask. Our work builds upon Fern\\'andez et al. (2007) which present a series of\nmachine-learning experiments on the classification of NSUs. We extended their\napproach with a combination of new features and semi-supervised learning\ntechniques. The empirical results presented in this thesis show a modest but\nsignificant improvement over the state-of-the-art classification performance.\nThe consecutive, yet independent, problem is how to infer an appropriate\nsemantic representation of such NSUs on the basis of the dialogue context.\nFern\\'andez (2006) formalizes this task in terms of "resolution rules" built on\ntop of the Type Theory with Records (TTR). Our work is focused on the\nreimplementation of the resolution rules from Fern\\'andez (2006) with a\nprobabilistic account of the dialogue state. The probabilistic rules formalism\nLison (2014) is particularly suited for this task because, similarly to the\nframework developed by Ginzburg (2012) and Fern\\'andez (2006), it involves the\nspecification of update rules on the variables of the dialogue state to capture\nthe dynamics of the conversation. However, the probabilistic rules can also\nencode probabilistic knowledge, thereby providing a principled account of\nambiguities in the NSU resolution process.\n