Bootstrapping incremental dialogue systems from minimal data: the generalisation power of dialogue grammars

We investigate an end-to-end method for automatically inducing task-based dialogue systems from small amounts of unannotated dialogue data. It combines an incremental semantic grammar - Dynamic Syntax and Type Theory with Records (DS-TTR) - with Reinforcement Learning (RL), where language generation and dialogue management are a joint decision problem. The systems thus produced are incremental: dialogues are processed word-by-word, shown previously to be essential in supporting natural, spontaneous dialogue. We hypothesised that the rich linguistic knowledge within the grammar should enable a combinatorially large number of dialogue variations to be processed, even when trained on very few dialogues. Our experiments show that our model can process 74% of the Facebook AI bAbI dataset even when trained on only 0.13% of the data (5 dialogues). It can in addition process 65% of bAbI+, a corpus we created by systematically adding incremental dialogue phenomena such as restarts and self-corrections to bAbI. We compare our model with a state-of-the-art retrieval model, MemN2N. We find that, in terms of semantic accuracy, MemN2N shows very poor robustness to the bAbI+ transformations even when trained on the full bAbI dataset.

[1]  Oliver Lemon,et al.  Challenging Neural Dialogue Models with Natural Data: Memory Networks Fail on Incremental Phenomena , 2017, ArXiv.

[2]  Oliver Lemon,et al.  How domain-general can we be? Learning incremental Dialogue Systems without Dialogue Acts , 2014 .

[3]  Matthew Purver,et al.  Probabilistic Type Theory for Incremental Dialogue Processing , 2014, EACL 2014.

[4]  Oliver Lemon,et al.  Bootstrapping incremental dialogue systems: using linguistic knowledge to learn from minimal data , 2016, NIPS 2016.

[5]  Arash Eshghi,et al.  Feedback in Conversation as Incremental Semantic Update , 2015, IWCS.

[6]  Oliver Lemon,et al.  Interactional dynamics and the emergence of language games , 2017, FADLI@ESSLLI.

[7]  Arash Eshghi,et al.  Incremental Semantic Construction in a Dialogue System , 2011, IWCS.

[8]  David Vandyke,et al.  Multi-domain Neural Network Language Generation for Spoken Dialogue Systems , 2016, NAACL.

[9]  Arash Eshghi,et al.  Conversational Interactions: Capturing Dialogue Dynamics , 2012 .

[10]  Mark Steedman,et al.  Combined Distributional and Logical Semantics , 2013, TACL.

[11]  Robert F. Port,et al.  Dynamics of Language , 2009, Encyclopedia of Complexity and Systems Science.

[12]  David Vandyke,et al.  A Network-based End-to-End Trainable Task-oriented Dialogue System , 2016, EACL.

[13]  Arash Eshghi,et al.  Probabilistic Grammar Induction in an Incremental Semantic Framework , 2012, CSLP.

[14]  Quoc V. Le,et al.  A Neural Conversational Model , 2015, ArXiv.

[15]  Dov M. Gabbay,et al.  Dynamic syntax - the flow of language understanding , 2000 .

[16]  Robin Cooper,et al.  Records and Record Types in Semantic Theory , 2005, J. Log. Comput..

[17]  Arash Eshghi,et al.  Incremental Grammar Induction from Child-Directed Dialogue Utterances , 2013, CMCL.

[18]  Jason Weston,et al.  Learning End-to-End Goal-Oriented Dialog , 2016, ICLR.

[19]  Jianfeng Gao,et al.  A Neural Network Approach to Context-Sensitive Generation of Conversational Responses , 2015, NAACL.

[20]  Gabriel Skantze,et al.  Towards Incremental Speech Generation in Dialogue Systems , 2010, SIGDIAL Conference.

[21]  Eunsol Choi,et al.  Scaling Semantic Parsers with On-the-Fly Ontology Matching , 2013, EMNLP.

[22]  Julian Hough Incremental Semantics Driven Natural Language Generation with Self-Repairing Capability , 2011, RANLP Student Research Workshop.

[23]  E. Gregoromichelaki,et al.  Splitting the “ I ”s and Crossing the “ You ”s: Context, Speech Acts and Grammar , 2010 .

[24]  Matthew Purver,et al.  Dialogue and compound contributions , 2014, Natural Language Generation in Interactive Systems.

[25]  Arash Eshghi,et al.  DyLan : Parser for Dynamic Syntax , 2013 .

[26]  Jonathan Ginzburg,et al.  The interactive stance : meaning for conversation , 2012 .

[27]  James F. Allen,et al.  Incremental Dialogue System Faster than and Preferred to its Nonincremental Counterpart , 2007 .

[28]  Lutz Marten,et al.  The Dynamics of Language , 2005 .

[29]  Pietro Perona,et al.  One-shot learning of object categories , 2006, IEEE Transactions on Pattern Analysis and Machine Intelligence.