What to talk about and how? Selective Generation using LSTMs with Coarse-to-Fine Alignment

Proceedings of NAACL-HLT 2016, pages 720–730, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics What to talk about and how? Selective Generation using LSTMs with Coarse-to-Fine Alignment Hongyuan Mei UChicago, TTI-Chicago hongyuan@uchicago.edu Mohit Bansal TTI-Chicago mbansal@ttic.edu Matthew R. Walter TTI-Chicago mwalter@ttic.edu Abstract We propose an end-to-end, domain- independent neural encoder-aligner-decoder model for selective generation, i.e., the joint task of content selection and surface realization. Our model first encodes a full set of over-determined database event records via an LSTM-based recurrent neural network, then utilizes a novel coarse-to-fine aligner to identify the small subset of salient records to talk about, and finally employs a decoder to generate free-form descriptions of the aligned, selected records. Our model achieves the best selection and generation results reported to-date (with 59% relative improvement in generation) on the benchmark WEATHER- GOV dataset, despite using no specialized features or linguistic resources. Using an improved k-nearest neighbor beam filter helps further. We also perform a series of ablations and visualizations to elucidate the contributions of our key model components. Lastly, we evaluate the generalizability of our model on the ROBOCUP dataset, and get results that are competitive with or better than the state-of-the-art, despite being severely data-starved. 1 Introduction We consider the important task of producing a natu- ral language description of a rich world state rep- resented as an over-determined database of event records. This task, which we refer to as selective generation, is often formulated as two subproblems: content selection, which involves choosing a sub- set of relevant records to talk about from the ex- haustive database, and surface realization, which is concerned with generating natural language descrip- tions for this subset. Learning to perform these tasks jointly is challenging due to the uncertainty in decid- ing which records are relevant, the complex depen- dencies between selected records, and the multiple ways in which these records can be described. Previous work has made significant progress on this task (Chen and Mooney, 2008; Angeli et al., 2010; Kim and Mooney, 2010; Konstas and Lap- ata, 2012). However, most approaches solve the two content selection and surface realization sub- tasks separately, use manual domain-dependent re- sources (e.g., semantic parsers) and features, or em- ploy template-based generation. This limits do- main adaptability and reduces coherence. We take an alternative, neural encoder-aligner-decoder ap- proach to free-form selective generation that jointly performs content selection and surface realization, without using any specialized features, resources, or generation templates. This enables our approach to generalize to new domains. Further, our memory- based model captures the long-range contextual de- pendencies among records and descriptions, which are integral to this task (Angeli et al., 2010). We formulate our model as an encoder-aligner- decoder framework that uses recurrent neural net- works with long short-term memory units (LSTM- RNNs) (Hochreiter and Schmidhuber, 1997) to- gether with a coarse-to-fine aligner to select and “translate” the rich world state into a natural lan- guage description. Our model first encodes the full set of over-determined1 event records using …

Content truncated to 500 words