[Paper Review] Non-Sentential Utterances in Dialogue: Experiments in Classification and Interpretation

Paolo Dragone|arXiv (Cornell University)|Nov 22, 2015

Natural Language Processing Techniques41 references3 citations

TL;DR

This paper proposes a hybrid framework for classifying and interpreting non-sentential utterances (NSUs) in dialogue by combining supervised machine learning with probabilistic rule-based reasoning. It improves NSU classification using enhanced features and semi-supervised learning, then introduces a probabilistic extension of Type Theory with Records (TTR) to resolve NSU semantics using dialogue context, achieving a principled handling of ambiguity and context dependency.

ABSTRACT

Non-sentential utterances (NSUs) are utterances that lack a complete sentential form but whose meaning can be inferred from the dialogue context, such as "OK", "where?", "probably at his apartment". The interpretation of non-sentential utterances is an important problem in computational linguistics since they constitute a frequent phenomena in dialogue and they are intrinsically context-dependent. The interpretation of NSUs is the task of retrieving their full semantic content from their form and the dialogue context. The first half of this thesis is devoted to the NSU classification task. Our work builds upon Fernández et al. (2007) which present a series of machine-learning experiments on the classification of NSUs. We extended their approach with a combination of new features and semi-supervised learning techniques. The empirical results presented in this thesis show a modest but significant improvement over the state-of-the-art classification performance. The consecutive, yet independent, problem is how to infer an appropriate semantic representation of such NSUs on the basis of the dialogue context. Fernández (2006) formalizes this task in terms of "resolution rules" built on top of the Type Theory with Records (TTR). Our work is focused on the reimplementation of the resolution rules from Fernández (2006) with a probabilistic account of the dialogue state. The probabilistic rules formalism Lison (2014) is particularly suited for this task because, similarly to the framework developed by Ginzburg (2012) and Fernández (2006), it involves the specification of update rules on the variables of the dialogue state to capture the dynamics of the conversation. However, the probabilistic rules can also encode probabilistic knowledge, thereby providing a principled account of ambiguities in the NSU resolution process.

Motivation & Objective

• Address the challenge of interpreting non-sentential utterances (NSUs), which are common in dialogue but lack full sentential structure.
• Improve NSU classification performance beyond existing state-of-the-art methods using enhanced features and semi-supervised learning.
• Develop a robust semantic interpretation framework for NSUs that accounts for context dependency and ambiguity.
• Overcome limitations of logic-based formalisms like TTR by integrating probabilistic modeling to handle partial observability and uncertainty in dialogue.
• Provide a proof-of-concept implementation of probabilistic resolution rules for NSUs grounded in dialogue state dynamics.

Proposed method

• Extend prior work by Fernandez (2007) using additional linguistic and contextual features for NSU classification.
• Apply semi-supervised learning techniques, including transductive SVM (TSVM), to improve classification with limited labeled data.
• Reimplement resolution rules from Fernandez (2014) using the probabilistic rules formalism (Lison, 2014) to model dialogue state updates.
• Integrate probabilistic knowledge into dialogue state transitions to represent uncertainty and ambiguity in NSU interpretation.
• Use Type Theory with Records (TTR) as a foundation for dialogue state modeling, enhanced with probabilistic inference for semantic resolution.
• Design a rule-based system that dynamically updates dialogue state variables based on NSU input and context, using probabilistic confidence scores.

Experimental results

Research questions

• RQ1 Can improved feature engineering and semi-supervised learning significantly enhance NSU classification accuracy over existing methods?
• RQ2 How can logic-based resolution rules from TTR be adapted to handle uncertainty and ambiguity in NSU interpretation?
• RQ3 To what extent does a probabilistic extension of TTR improve the robustness and accuracy of NSU semantic resolution?
• RQ4 Can probabilistic rules effectively model context-dependent, partially observable dialogue phenomena in NSU interpretation?
• RQ5 Is there a measurable improvement in NSU interpretation performance when combining classification and probabilistic resolution in a unified framework?

Key findings

• The proposed classification method achieves a modest but statistically significant improvement over the state-of-the-art, demonstrating the effectiveness of enhanced features and semi-supervised learning.
• The integration of transductive SVM (TSVM) and active learning (AL) leads to better generalization on low-resource NSU classification tasks.
• The probabilistic rules formalism successfully captures ambiguity in NSU interpretation by assigning confidence scores to possible semantic expansions.
• The framework provides a principled way to model context-dependent, non-deterministic phenomena in dialogue, overcoming limitations of purely logical TTR approaches.
• The proof-of-concept implementation shows that probabilistic resolution rules can dynamically update dialogue state with uncertainty-aware semantics.
• The approach enables more robust interpretation of NSUs such as 'OK', 'where?', and 'probably at his apartment' by combining classification with probabilistic context modeling.