BTPK-based interpretable method for NER tasks based on Talmudic Public Announcement Logic

What it is about

The authors ask how a bidirectional RNN recognises named entities, and propose BTPK, a binary version of Talmudic public announcement logic. The hidden states of the trained network are mapped to paths in a tree of possible labels, and a public announcement (a later state that settles an earlier undecided choice) lets the model return to the right branch. The case study trains the network on a Chinese NER dataset with 7 label categories (8,791 sequences extracted, 7,814 for training). To explain why "Hobbits" is labelled as a video name, they fix the hidden states of the word "movie" to zero, the label changes, and "movie" becomes the public announcement for that label. The parts read give no accuracy scores, so none are quoted here.

Ideas

I1. An intervention check as a deterministic explanation: for each entity label, remove or neutralise each content word in turn and record which removals flip the label. The words that flip it are the "reason" for the label, which is the rule-based counterpart of the paper's test, and it gives a why for each decision.
I2. Retroactive revision: when a later clause settles the reading of an earlier ambiguous entity (a name that could be a book, a film or a song), reopen the earlier branch and relink the entity, rather than overwriting it. Keep both readings in the world store so the change is visible.

For Mova

I1 goes into prag as an explanation output for entity or role labels. I2 goes into world, for revising an entity after a later clause. The neural model and the Chinese dataset are not used, and no languages/ chapter applies.

Topics: Named entity recognition

Cite

The paper itself, not this card. BibTeX from the library's record of the work. arXiv 2201.09523

@misc{chi2022btpk,
  title = {{BTPK-based interpretable method for NER tasks based on Talmudic Public Announcement Logic}},
  author = {Chi;, Yulin Chen; Beishui Liao; Bruno Bentzen; Bo Yuan; Zelai Yao; Haixiao and Gabbay, Dov},
  year = {2022},
  eprint = {2201.09523},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2201.09523}
}

References

11 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

Abraham, M., I. Belfer, D. M. Gabbay and U. Schild, Future determination of entities in talmudic public announcement logic, Journal of Applied logic 11 (2013), pp. 63–90. ↗
Agarwal, O., Y. Yang, B. C. Wallace and A. Nenkova, Interpretability analysis for named entity recognition to understand system predictions and how they can improve, Computational Linguistics 47 (2021), pp. 117–140. ↗
Alvi, M. Z. H., S. Zaman, J. R. Saurav, S. Haque, M. S. Islam and M. R. Amin, B-NER: A novel bangla named entity recognition dataset with largest entities and its baseline evaluation, IEEE Access 11 (2023), pp. 45194–45205. URL https://doi.org/10.1109/ACCESS.2023.3267746 ↗
Cornish, T. A. O., Mill’s methods for complete intelligent data analysis, in: X. Liu, P. R. Cohen and M. R. Berthold, editors, Advances in Intelligent Data Analysis, Reasoning about Data, Second International Symposium, IDA-97, London, UK, August 4-6, 1997, Proceedings, Lecture Notes in Computer Science 1280 (1997), pp. 65–76. URL https://doi.org/10.1007/BFb0052830 ↗
Fu, J., P. Liu and G. Neubig, Interpretable multi-dataset evaluation for named entity recognition, arXiv preprint arXiv:2011.06854 (2020). ↗
Fu, Y., N. Lin, Z. Yang and S. Jiang, Towards malay named entity recognition: an open-source dataset and a multi-task framework, Connect. Sci. 35 (2023). ↗
Hou, B.-J. and Z.-H. Zhou, Learning with interpretable structure from gated rnn, IEEE transactions on neural networks and learning systems 31 (2020), pp. 2267–2279. find
Krakovna, V. and F. Doshi-Velez, Increasing the interpretability of recurrent neural networks using hidden markov models, arXiv preprint arXiv:1606.05320 (2016). ↗
Li, J., A. Sun, J. Han and C. Li, A survey on deep learning for named entity recognition, IEEE Transactions on Knowledge and Data Engineering 34 (2022), pp. 50–70. ↗
Lin, B. Y., D.-H. Lee, M. Shen, R. Moreno, X. Huang, P. Shiralkar and X. Ren, Triggerner: Learning with entity triggers as explanations for named entity recognition, arXiv preprint arXiv:2004.07493 (2020). ↗
Wisdom, S., T. Powers, J. Pitton and L. Atlas, Interpretable recurrent neural networks using sequential sparse recovery, arXiv preprint arXiv:1611.07252 (2016). ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/langstream-2022-chen-btpk-based-interpretable-method-ner/