Elamur Research⌕
☰
home / papers / 2019 / Improving Natural Language Interaction with Robots Using Advice

Improving Natural Language Interaction with Robots Using Advice

Paper · 2019 · Nikhil Mehta and Dan Goldwasser

Summary

The authors work on the Blocks World task: an agent receives a natural-language instruction and the state of the board and must determine the coordinates of the block to be moved and the location where it should be placed. Such interaction is usually treated as one-shot: a human gives a command and a robot executes it. Here the human is allowed to give a short piece of advice after seeing the agent's error. Advice comes in two kinds. Restrictive advice indicates a region of the board. Corrective advice indicates a direction of shift: up, down, left or right. First, a separate LSTM network is trained to understand the advice text. It is then frozen and incorporated into the end-to-end model of Bisk et al. (2016). The authors compare four protocols that require progressively less human effort. In the cheapest of them, the model generates the advice itself, so no human is needed at all. Even simple advice noticeably reduces the error. All variants outperform the Bisk et al. baseline, and restrictive advice from a human gives a better result than Tan and Bansal. The work shows that short interactive advice is a cheap way to improve accuracy, and the approach can be applied to any similar architecture.

Key points

  • Error is measured as the distance to the correct point in block lengths. Restrictive advice with 4 regions (M5) gives a mean error of 2.21 for source and 2.19 for target. For Bisk et al. (M1) these values are 3.47 and 3.70, and for Tan and Bansal (M3) they are 2.21 and 3.07.
  • The modules that understand the advice text reach an accuracy of 99.99%. Without pre-training such a module (M4), the result is worse than the baseline model.
  • The human had to give advice in only 395 of the 720 test examples: only when the prediction fell into the wrong region. During training, a random half of the examples received advice in each epoch.
  • The "retry" protocol (M7): the model itself proposes the most likely region, and the human only asks it to try the next one. For target, the median error is 2.42, which is close to Tan and Bansal.
  • Fully automatic advice works without human involvement. Its best variant (M9) is a region centred on the model's initial prediction; it gives a mean error of 3.12 for source and 3.26 for target. This is better than Bisk et al.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksArts and Humanities · 17,959 worksLanguage and Linguistics · 17,789 worksPsychology · 1,673 worksSocial Psychology · 332 works

Cite

Cite as Nikhil Mehta and Dan Goldwasser. 2019. Improving Natural Language Interaction with Robots Using Advice. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/n19-1195 · arXiv 1905.04655

@inproceedings{mehta2019improving,
  title = {{Improving Natural Language Interaction with Robots Using Advice}},
  author = {Nikhil Mehta and Dan Goldwasser},
  booktitle = {Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)},
  year = {2019},
  doi = {10.18653/v1/n19-1195},
  eprint = {1905.04655},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/n19-1195}
}

References

17 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. James F Allen, Lenhart K Schubert, George Ferguson, Peter Heeman, Chung Hee Hwang, Tsuneaki Kato, Marc Light, Nathaniel Martin, Bradford Miller, Massimo Poesio, et al. 1995. The trains project: A case study in building a conversational planning agent. Journal of Experimental & Theoretical Artificial… find
  2. Yonatan Bisk, Deniz Yuret, and Daniel Marcu. 2016. Natural language communication with robots. In Proc. of the Annual Meeting of the North American Association of Computational Linguistics (NAACL). find
  3. Satchuthananthavale RK Branavan, Harr Chen, Luke S Zettlemoyer, and Regina Barzilay. 2009. Reinforcement learning for mapping instructions to actions. In Proc. of the Annual Meeting of the Association Computational Linguistics (ACL). find
  4. David L Chen and Raymond J Mooney. 2011. Learning to interpret natural language navigation instructions from observations. In Proc. of the National Conference on Artificial Intelligence (AAAI). find
  5. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. find
  6. Paul E Rybski, Kevin Yoon, Jeremy Stolarz, and Manuela M Veloso. 2007. Interactive robot task training through dialog and demonstration. In Proc. of the ACM/IEEE international conference on Human-robot interaction, pages 49–56. ACM. find
  7. Hao Tan and Mohit Bansal. 2018. Source-target inference models for spatial instruction understanding. In Proc. of the National Conference on Artificial Intelligence (AAAI). find
  8. Stefanie Tellex, Thomas Kollar, Steven Dickerson, Matthew R Walter, Ashis Gopal Banerjee, Seth J Teller, and Nicholas Roy. 2011. Understanding natural language commands for robotic navigation and mobile manipulation. In Proc. of the National Conference on Artificial Intelligence (AAAI). find
  9. TH Wen, D Vandyke, N Mrkšı́c, M Gašı́c, LM Rojas-Barahona, PH Su, S Ultes, and S Young. A network-based end-to-end trainable task-oriented dialogue system. In Proc. of the Annual Meeting of the European Chapter of the Association for Computational Linguistics (EACL).
  10. Terry Winograd. 1972. Understanding natural language. Cognitive psychology, 3(1):1–191. find
  11. Joohyun Kim and Raymond J Mooney. 2013. Adapting discriminative reranking to grounded language learning. In Proc. of the Annual Meeting of the Association Computational Linguistics (ACL), pages 218–227. find
  12. Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
  13. Andrew L Maas, Awni Y Hannun, and Andrew Y Ng. 2013. Rectifier nonlinearities improve neural network acoustic models. In Proc. of the International Conference on Machine Learning (ICML). find
  14. Matt MacMahon, Brian Stankiewicz, and Benjamin Kuipers. 2006. Walk the talk: Connecting language, knowledge, and action in route instructions. In Proc. of the National Conference on Artificial Intelligence (AAAI). find
  15. Cynthia Matuszek, Nicholas FitzGerald, Luke Zettlemoyer, Liefeng Bo, and Dieter Fox. 2012. A joint model of language and perception for grounded attribute learning. In Proc. of the International Conference on Machine Learning (ICML). find
  16. Dipendra K Misra, John Langford, and Yoav Artzi. 2017. Mapping instructions and visual observations to actions with reinforcement learning. In Proc. of the Conference on Empirical Methods for Natural Language Processing (EMNLP).
  17. Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. 2013. On the difficulty of training recurrent neural networks. In Proc. of the International Conference on Machine Learning (ICML), pages 1310–1318. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mehta-2019-improving-natural-language-interaction-robots/