AI帮你理解科学

AI 生成解读视频

AI抽取解析论文重点内容自动生成视频


pub
生成解读视频

AI 溯源

AI解析本论文相关学术脉络


Master Reading Tree
生成 溯源树

AI 精读

AI抽取本论文的概要总结


微博一下
We propose slot-value informed sequence-tosequence models with attention for language generation in dialogue systems, where the slot values in the system action, as well as the slot tags are part of the input sequence

To Plan or not to Plan? Discourse Planning in Slot-Value Informed Sequence to Sequence Models for Language Generation.

INTERSPEECH, pp.3339-3343, (2017)

被引用24|浏览234
EI
下载 PDF 全文
引用
微博一下

摘要

Natural language generation for task-oriented dialogue systems aims to effectively realize system dialogue actions. All natural language generators (NLGs) must realize grammatical, natural and appropriate output, but in addition, generators for taskoriented dialogue must faithfully perform a specific dialogue act that conveys specific sem...更多

代码

数据

简介
  • Natural language generation for task-oriented dialogue systems aims to effectively realize system dialogue actions.
  • In order to train the NLG for task-oriented dialogue, it is necessary to provide training data that represents the mapping from semantics to surface realizations.
  • To date, this has meant that the data collection must target the particular task, whereas NLG for chatbots can often rely on large-scale, harvested, social media dialogues [3, 4].
  • The result is that datasets for training NLGs for task-oriented dialogue are typically smaller in scale than those used for chatbots [5, 6, 7]
重点内容
  • Natural language generation for task-oriented dialogue systems aims to effectively realize system dialogue actions
  • The result is that datasets for training natural language generators for task-oriented dialogue are typically smaller in scale than those used for chatbots [5, 6, 7]
  • In HUMAN-SPECIFIC, we attempt to match both slots and values, whereas in HUMAN-GENERAL we attempt to match slots. This is essentially template-based generation, where a template is induced from the human utterances, and used for realization, as in other work [26]
  • We propose slot-value informed sequence-tosequence models with attention for language generation in dialogue systems, where the slot values in the system action, as well as the slot tags are part of the input sequence
  • We investigate three ways of representing the values in the input sequence
  • We show that the slot values represented as vectors concatenated to vectors of slot names results in best overall quality
方法
  • Evaluation The authors' evaluation experiments solicit judgements from raters and trained annotators for both objective and subjective metrics.
  • The authors asked them separately to report whether there were any extraneous slots present (Slot precision)
  • This appeared to be difficult for raters, the authors conducted a separate evaluation with a trained annotator, on the mention representation results.
  • The authors asked the trained annotator to separately evaluate the correctness of scalar values assignment to slots: this is reported as Scalar precision in Table 7
结果
  • Results and Analysis

    The authors first compared various mention representation strategies in an LSTM setting akin to NONE plan supervision.
  • Given a frame to generate from, the authors attempt to select an utterance from the training set that matches this frame.
  • In HUMAN-SPECIFIC, the authors attempt to match both slots and values, whereas in HUMAN-GENERAL the authors attempt to match slots.
  • This is essentially template-based generation, where a template is induced from the human utterances, and used for realization, as in other work [26].
结论
  • The authors propose slot-value informed sequence-tosequence models with attention for language generation in dialogue systems, where the slot values in the system action, as well as the slot tags are part of the input sequence.
  • The authors investigate three ways of representing the values in the input sequence.
  • The authors study if the models can learn to plan the sentences given the limited amount of training data, or if they would sbenefit from being presented a plan in the form of multiple sentences as part of the input.
  • The authors show that the slot values represented as vectors concatenated to vectors of slot names results in best overall quality.
  • Adding a plan to the input further improves quality and diversity
表格
  • Table1: NLG for Restaurants I
  • Table2: Delexicalized Representations
  • Table3: Restaurant Entity Schema. Example values for categorical slots are shown. All scaled slots share the range of four possible values
  • Table4: NLG for Restaurants II
  • Table5: Input tokens for the various mention representations
  • Table6: Input tokens for the various plan supervision strategies. All slots are mentioned for NONE; slots are grouped into subsequences for FLAT and FINE, POSITIONAL adds tokens to indicate if a sentence is in the beginning or inside of an utterance
  • Table7: Results for mention representation and plan supervision. Values of CONCAT marked with * are significantly better than SEQ; values marked with † are significantly better than HUMAN-S. Bolded values of POSITIONAL are significantly better than NONE
Download tables as Excel
基金
  • Investigates sequence-tosequence models in which slot values are included as part of the input sequence and the output surface form
  • Studies whether a separate sentence planning module that decides on grouping of slot value mentions as input to the seq2seq model results in more natural sentences than a seq2seq model that aims to jointly learn the plan and the surface realization
  • Presents a dataset selected to encourage aggregation and discourse structuring
  • Introduces supervision for sentence planning phenomena as input to the model
引用论文
  • P. Shah, D. Hakkani-Tur, and L. Heck, “Interactive reinforcement learning for task-oriented dialogue management,” in NIPS 2016 Deep Learning for Action and Interaction Workshop, 2016.
    Google ScholarFindings
  • M. A. Walker, “An application of reinforcement learning to dialogue strategy selection in a spoken dialogue system for email,” Journal of Artificial Intelligence Research, vol. 12, pp. 387–416, 2000.
    Google ScholarLocate open access versionFindings
  • R. Lowe, N. Pow, I. Serban, and J. Pineau, “The ubuntu dialogue corpus: A large dataset for research in unstructured multi-turn dialogue systems,” in Proc. of the SIGDIAL 2015 Conference: The 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue, 2015.
    Google ScholarLocate open access versionFindings
  • R. Abbott, B. Ecker, P. Anand, and M. Walker, “Internet argument corpus 2.0: An sql schema for dialogic social media and the corpora to go with it,” in Language Resources and Evaluation Conference, LREC2016, 2016.
    Google ScholarLocate open access versionFindings
  • D. Hakkani-Tur, G. Tur, A. Celikyilmaz, Y.-N. Chen, J. Gao, L. Deng, and Y.-Y. Wang, “Multi-domain joint semantic frame parsing using bi-directional rnn-lstm,” in Proceedings of The 17th Annual Meeting of the International Speech Communication Association, 2016.
    Google ScholarLocate open access versionFindings
  • N. Dethlefs, H. Cuayahuitl, H. Hastie, V. Rieser, and O. Lemon, “Cluster-based prediction of user ratings for stylistic surface realisation,” EACL 2014, p. 702, 2014.
    Google ScholarLocate open access versionFindings
  • V. Rieser and O. Lemon, Reinforcement learning for adaptive dialogue systems: a data-driven methodology for dialogue management and natural language generation. Springer Science & Business Media, 2011.
    Google ScholarFindings
  • T.-H. Wen, M. Gasic, N. Mrksic, P.-H. Su, D. Vandyke, and S. Young, “Semantically conditioned lstm-based natural language generation for spoken dialogue systems,” in Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, 2015.
    Google ScholarLocate open access versionFindings
  • R. Lebret, D. Grangier, and M. Auli, “Generating text from structured data with application to the biography domain,” in Proc. of Empirical Methods in Natural Language Processing (EMNLP), 2016.
    Google ScholarLocate open access versionFindings
  • A. Ritter, C. Cherry, and W. B. Dolan, “Data-driven response generation in social media,” in Proceedings of the conference on empirical methods in natural language processing. Association for Computational Linguistics, 2011, pp. 583–593.
    Google ScholarLocate open access versionFindings
  • A. Sordoni, M. Galley, M. Auli, C. Brockett, Y. Ji, M. Mitchell, J.-Y. Nie, J. Gao, and B. Dolan, “A neural network approach to context-sensitive generation of conversational responses,” in Proc. of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2015.
    Google ScholarLocate open access versionFindings
  • O. Dusek and F. Jurcıcek, “Sequence-to-sequence generation for spoken dialogue via deep syntax trees and strings,” in The 54th Annual Meeting of the Association for Computational Linguistics, 2016, p. 45.
    Google ScholarLocate open access versionFindings
  • R. Barzilay and M. Lapata, “Aggregation via set partitioning for natural language generation,” in Proceedings of the main conference on Human Language Technology Conference of the North American Chapter of the Association of Computational Linguistics. Association for Computational Linguistics, 2006, pp. 359– 366.
    Google ScholarLocate open access versionFindings
  • M. Walker, O. Rambow, and M. Rogati, “Training a sentence planner for spoken dialogue using boosting,” Computer Speech and Language: Special Issue on Spoken Language Generation, vol. 16, no. 3-4, pp. 409–433, 2002.
    Google ScholarLocate open access versionFindings
  • R. Barzilay and L. Lee, “Bootstrapping lexical choice via multiple-sequence alignment,” in Proceedings of the ACL-02 conference on Empirical methods in natural language processingVolume 10. Association for Computational Linguistics, 2002, pp. 164–171.
    Google ScholarLocate open access versionFindings
  • A. Stent, “Rhetorical structure in dialog,” in Proc. of the first international conference on Natural language generation-Volume 14. Association for Computational Linguistics, 2000, pp. 247–252.
    Google ScholarLocate open access versionFindings
  • ——, “A conversation acts model for generating spoken dialogue contributions,” Computer Speech and Language: Special Issue on Spoken Language Generation, 2002.
    Google ScholarLocate open access versionFindings
  • M. Walker and O. Rambow, Eds., Computer Speech and Language: Special Issue on Spoken Language Generation. Academic Press, 2002, vol. 16: 3-4.
    Google ScholarFindings
  • O. Dusek and F. Jurcicek, “A context-aware natural language generator for dialogue systems,” in 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue, 2016.
    Google ScholarLocate open access versionFindings
  • D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine translation by jointly learning to align and translate,” arXiv preprint arXiv:1409.0473, 2014.
    Findings
  • M. A. Walker, A. Stent, F. Mairesse, and R. Prasad, “Individual and domain adaptation in sentence planning for dialogue,” Journal of Artificial Intelligence Research (JAIR), vol. 30, pp. 413– 456, 2007.
    Google ScholarLocate open access versionFindings
  • J. Novikova, O. Lemon, and V. Rieser, “Crowd-sourcing nlg data: Pictures elicit better data,” in International Conference on Natural Language Generation, 2016.
    Google ScholarLocate open access versionFindings
  • K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “Bleu: a method for automatic evaluation of machine translation,” in Proceedings of the 40th annual meeting on association for computational linguistics. Association for Computational Linguistics, 2002, pp. 311–318.
    Google ScholarLocate open access versionFindings
  • A. Belz and E. Reiter, “Comparing automatic and human evaluation of nlg systems.” in EACL, 2006.
    Google ScholarFindings
  • C.-W. Liu, R. Lowe, I. V. Serban, M. Noseworthy, L. Charlin, and J. Pineau, “How not to evaluate your dialogue system: An empirical study of unsupervised evaluation metrics for dialogue response generation,” in Proc. of Empirical Methods in Natural Language Processing (EMNLP), 2016.
    Google ScholarLocate open access versionFindings
  • G. Angeli, P. Liang, and D. Klein, “A simple domain-independent probabilistic approach to generation,” in Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2010, pp. 502– 512.
    Google ScholarLocate open access versionFindings
您的评分 :
0

 

标签
评论
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科