Elamur Research⌕
☰
home / papers / 2023 / Towards Explainable AI Writing Assistants for Non-native English Speakers

Towards Explainable AI Writing Assistants for Non-native English Speakers

Paper · 2023 · Yewon Kim, Mina Lee, Donghwi Kim et al.

Summary

Paraphrasing tools such as Grammarly, Wordtune or Quillbot usually offer several rewrites of a sentence but say nothing about how the options differ or why they were proposed. The authors argue that this gap hurts non-native English speakers (NNESs) most, because many of them cannot judge tone, nuance or shifts in meaning on their own. They ran a semi-structured interview study with 15 NNESs who regularly write English emails. Each participant composed an email from a scenario taken from prior work, using their everyday tools and thinking aloud, and then discussed the difficulties they met. The recordings were transcribed and analysed with open coding and thematic analysis. Participants distrusted suggestions whose rationale was hidden and worried about politeness and unintended readings. Those with lower proficiency struggled most. To cope, they checked dictionaries, example-sentence sites, search hit counts, Ngram trends and confidence percentages, or they gave up or asked native speakers. From these coping habits the paper derives four interface concepts that attach explanations to suggestions. This makes a user-grounded case for explainable writing support.

Key points

  • The participants were 15 NNESs with varied proficiency. Among them, 12 used Grammarly, 4 used Wordtune, 2 used Quillbot and 1 used Writefull. Seven also paraphrased indirectly by tweaking their first-language input to Google Translate or Naver Papago.
  • Eleven of the 15 reported trouble expressing their intended meaning in English. One example was choosing between "ratio" and "proportion", which share a single Korean translation. Five said their writing sounds unnatural.
  • Seven participants said they distrust AI tools, and four hesitated because the reasons behind suggestions were invisible. Seven had partly given up on reaching fully satisfying results.
  • The most common coping strategy, used by eight participants, was checking human-written examples from credible sources such as dictionaries, Ludwig and Thesaurus. Others used Google hit counts, the Google Ngram viewer and Writefull's confidence scores.
  • The paper proposes four interface designs:
    • (a) confidence scores for each output;
    • (b) example sentences from credible sources;
    • (c) usage-trend comparison between the original and the paraphrase;
    • (d) textual explanations.
  • The discussion covers efficiency vs. quality, learning effects, dialogue-based interaction and bias.

Interesting ideas

  • I1. Rank alternative outputs by model confidence and show the score, because a flat list of unranked options makes users feel the system chose carelessly.
  • I2. Justify a choice by pointing to real human-written usage examples rather than only asserting it, since users trust attested evidence more than model claims.
  • I3. Compare how frequent or trendy the original phrasing is against the replacement, while noting that raw counts such as search hits can be skewed.
  • I4. Separate local explanations, which justify one specific suggestion, from general explanations, which describe where an expression is typically used, and adapt both to the reader's level, for example by simplifying them or translating them into the first language.
  • I5. Treat explanation detail as a cost to tune: rich explanations improve decisions and learning but slow experts down, while oversimplified ones lower trust and encourage over-reliance.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (answer): When a reader returns several candidate answers or readings, Mova could order them by a calibrated score and show it, so users can see which one the system actually stands behind.
  • I2 (frames): A frame or semantic-role decision could cite matching attested sentences from a treebank or VerbNet example set as evidence that the verb is used that way.
  • I3 (back): During generation, Mova could check candidate phrasings against corpus frequency and expose that comparison, without treating frequency alone as proof of appropriateness.
  • I4 (answer): Each explained answer could carry a local trace for this particular question plus an optional general note on the rule involved, rendered in simpler English or in Ukrainian on request.
  • I5 (learn): Evaluation could measure explanation usefulness by explanation depth and user expertise, rather than assuming that more explanation is always better.

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

 Institutions

🎓Stanford UniversityStanford · United States434 works · 314,419 citations🎓Korea Advanced Institute of Science and TechnologyDaejeon · South Korea51 works · 3,167 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksHuman-Computer Interaction · 141 works

🧭 Topics 2Research subjects and languages the work is about.English · 2,263 worksInterpretability · 130 works

Cite

Cite as Yewon Kim, Mina Lee, Donghwi Kim and Sung-Ju Lee. 2023. Towards Explainable AI Writing Assistants for Non-native English Speakers. arXiv:2304.02625.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2304.02625 · arXiv 2304.02625

@misc{kim2023explainable,
  title = {{Towards Explainable AI Writing Assistants for Non-native English Speakers}},
  author = {Yewon Kim, Mina Lee, Donghwi Kim and Sung-Ju Lee},
  year = {2023},
  doi = {10.48550/arxiv.2304.02625},
  eprint = {2304.02625},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2304.02625}
}

References

56 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David Alvarez-Melis and Tommi Jaakkola. 2017. A causal framework for explaining the predictions of black-box sequence-to-sequence models. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Copenhagen, Denmark, 412–421. https://doi.org/10.18653/v1/D17-1042 ↗
  2. Fauzul Aufa. 2016. Explicit Pragmatic Instruction In Teaching English As A Foreign Language. Journal of English and Education 5, 1 (Apr. 2016), 37–44. https://journal.uii.ac.id/JEE/article/view/4458 ↗
  3. Dzmitry Bahdanau, Kyung Hyun Cho, and Yoshua Bengio. 2015. Neural machine translation by jointly learning to align and translate. 3rd International Conference on Learning Representations, ICLR 2015 ; Conference date: 07-05-2015 Through 09-05-2015. ↗
  4. Joeran Beel and Haley Dixon. 2021. The ‘unreasonable’effectiveness of graphical user interfaces for recommender systems. In Adjunct Proceedings of the 29th ACM Conference on User Modeling, Adaptation and Personalization. 22–28. find
  5. Daniel Buschek, Martin Zürn, and Malin Eiband. 2021. The Impact of Multiple Parallel Phrase Suggestions on Email Input and Composition Behaviour of Native and Non-Native English Writers. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (Yokohama, Japan) (CHI ’21). Association for Computing Machinery, New York, NY, USA, Article 732, 13 pages. https://doi.org/10.1145/3411764.3445372 ↗
  6. Adrian Bussone, Simone Stumpf, and Dympna O’Sullivan. 2015. The Role of Explanations on Trust and Reliance in Clinical Decision Support Systems. In 2015 International Conference on Healthcare Informatics. 160–169. https://doi.org/10.1109/ICHI.2015.26 ↗
  7. Jenny Cameron, Karen Nairn, and Jane Higgins. 2009. Demystifying Academic Writing: Reflections on Emotions, Know-How and Academic Identity. Journal of Geography in Higher Education 33, 2 (2009), 269–284. https://doi.org/10.1080/03098260902734943 arXiv:https://doi.org/10.1080/03098260902734943 ↗
  8. Mia Xu Chen, Benjamin N. Lee, Gagan Bansal, Yuan Cao, Shuyuan Zhang, Justin Lu, Jackie Tsay, Yinan Wang, Andrew M. Dai, Zhifeng Chen, Timothy Sohn, and Yonghui Wu. 2019. Gmail Smart Compose: Real-Time Assisted Writing. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (Anchorage, AK, USA) (KDD ’19). Association for Computing Machinery, New York, NY, USA, 2287–2295. https://doi.org/10.1145/3292500.3330723 ↗
  9. Michelene T.H. Chi, Miriam Bassok, Matthew W. Lewis, Peter Reimann, and Robert Glaser. 1989. Self-Explanations: How Students Study and Use Examples in Learning to Solve Problems. Cognitive Science 13, 2 (1989), 145–182. https://onlinelibrary.wiley.com/doi/pdf/10.1207/s15516709cog1302_1 ↗
  10. Dan Cosley, Shyong K Lam, Istvan Albert, Joseph A Konstan, and John Riedl. 2003. Is seeing believing? How recommender system interfaces affect users’ opinions. In Proceedings of the SIGCHI conference on Human factors in computing systems. 585–592. find
  11. Robert DeKeyser. 1998. Beyond focus on form: Cognitive perspectives on learning and practicing second language grammar. Focus on form in classroom second language acquisition 28 (1998), 42–63. find
  12. Robert DeKeyser. 2003. Implicit and explicit learning. The handbook of second language acquisition (2003), 312–348. ↗
  13. Robert DeKeyser. 2020. Skill Acquisition Theory. 83–104. https://doi.org/10.4324/9780429503986-5 ↗
  14. Catherine Doughty and Jessica Williams. 1998. Pedagogical choices in focus on form. Focus on form in classroom second language acquisition 3 (1998), 197–262. find
  15. Nick Ellis, Ute Römer, and Matthew O’Donnell. 2016. Usage-based Approaches to Language Acquisition and Processing: Cognitive and Corpus Investigations of Construction Grammar. find
  16. Muhammad Fareed, Almas Ashraf, and Muhammad Bilal. 2016. ESL learners’ writing skills: Problems, factors and suggestions. Journal of education and social sciences 4, 2 (2016), 81–92. ↗
  17. Google. 2023. Google Docs AutoCorrect. https://docs.google.com/document/u/0/. ↗
  18. Google. 2023. Google Ngram Viewer. https://books.google.com/ngrams/. ↗
  19. Google. 2023. Google Translate. https://translate.google.com/. ↗
  20. Grammarly. 2023. Grammarly: Free Online Writing Assistant. https://www.grammarly.com. ↗
  21. David Gunning and David Aha. 2019. DARPA’s Explainable Artificial Intelligence (XAI) Program. AI Magazine 40, 2 (Jun. 2019), 44–58. https://doi.org/10.1609/aimag.v40i2.2850 ↗
  22. Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. 2017. On Calibration of Modern Neural Networks. In Proceedings of the 34th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 70), Doina Precup and Yee Whye Teh (Eds.). PMLR, 1321–1330. https://proceedings.mlr.press/v70/guo17a.html ↗
  23. Benjamin Hoover, Hendrik Strobelt, and Sebastian Gehrmann. 2020. exBERT: A Visual Analysis Tool to Explore Learned Representations in Transformer Models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations. Association for Computational Linguistics, Online, 187–196. https://doi.org/10.18653/v1/2020.acl-demos.22 ↗
  24. Julie S. Hui, Darren Gergle, and Elizabeth M. Gerber. 2018. IntroAssist: A Tool to Support Writing Introductory Help Requests. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems (Montreal QC, Canada) (CHI ’18). Association for Computing Machinery, New York, NY, USA, 1–13. https://doi.org/10.1145/3173574.3173596 ↗
  25. Maurice Jakesch, Advait Bhat, Daniel Buschek, Lior Zalmanson, and Mor Naaman. 2023. Co-Writing with Opinionated Language Models Affects Users’ Views. arXiv preprint arXiv:2302.00560 (2023). ↗
  26. Kakao. 2023. Kakao i Translate. https://translate.kakao.com/. ↗
  27. Masahiro Kaneko, Sho Takase, Ayana Niwa, and Naoaki Okazaki. 2022. Interpretability for Language Learners Using Example-Based Grammatical Error Correction. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Dublin, Ireland, 7176–7187. https://doi.org/10.18653/v1/2022.acl-long.496 ↗
  28. Stephen Krashen. 1981. Second language acquisition. Second Language Learning 3, 7 (1981), 19–39. find
  29. Todd Kulesza, Simone Stumpf, Margaret Burnett, Sherry Yang, Irwin Kwan, and Weng-Keen Wong. 2013. Too much, too little, or just right? Ways explanations impact end users’ mental models. In 2013 IEEE Symposium on Visual Languages and Human Centric Computing. 3–10. https://doi.org/10.1109/VLHCC.2013.6645235 ↗
  30. Vivian Lai and Chenhao Tan. 2019. On Human Predictions with Explanations and Predictions of Machine Learning Models: A Case Study on Deception Detection. In Proceedings of the Conference on Fairness, Accountability, and Transparency (Atlanta, GA, USA) (FAT* ’19). Association for Computing Machinery, New York, NY, USA, 29–38. https://doi.org/10.1145/3287560.3287590 ↗
  31. LanguageTool. 2023. LanguageTool - Online Grammar, Style & Spell Checker. https://languagetool.org/. ↗
  32. Hajin Lim, Dan Cosley, and Susan R Fussell. 2022. Understanding Cross-lingual Pragmatic Misunderstandings in Email Communication. Proceedings of the ACM on Human-Computer Interaction 6, CSCW1 (2022), 1–32. ↗
  33. Ludwig. 2023. Ludwig • Find your English sentence. https://ludwig.guru/. ↗
  34. Ryo Nagata, Masato Hagiwara, Kazuaki Hanawa, Masato Mita, Artem Chernodub, and Olena Nahorna. 2021. Shared Task on Feedback Comment Generation for Language Learners. In Proceedings of the 14th International Conference on Natural Language Generation. Association for Computational Linguistics, Aberdeen, Scotland, UK, 320–324. https://aclanthology.org/2021.inlg-1.35 ↗
  35. Naver. 2023. Papago. https://papago.naver.com/. ↗
  36. David Nunan. 1989. Designing tasks for the communicative classroom. Cambridge university press. find
  37. Council of Europe. Council for Cultural Co-operation. Education Committee. Modern Languages Division. 2001. Common European framework of reference for languages: Learning, teaching, assessment. Cambridge University Press. find
  38. OpenAI. 2023. GPT-4 Technical Report. arXiv:2303.08774 [cs.CL] ↗
  39. OpenAI. 2023. Introducing ChatGPT. https://openai.com/blog/chatgpt. ↗
  40. Lourdes Ortega. 2003. Syntactic complexity measures and their relationship to L2 proficiency: A research synthesis of college-level L2 writing. Applied linguistics 24, 4 (2003), 492–518. ↗
  41. Eli Pariser. 2012. The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think. Penguin Books, USA. find
  42. Pearson. 2023. Longman Dictionaries: Corpus Network. http://www.pearsonlongman.com/dictionaries/corpus/. ↗
  43. Charlene G Polio. 1997. Measures of linguistic accuracy in second language writing research. Language learning 47, 1 (1997), 101–143. ↗
  44. Quillbot. 2023. Paraphrasing Tool - QuillBot AI. https://quillbot.com. ↗
  45. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. "Why Should I Trust You?": Explaining the Predictions of Any Classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (San Francisco, California, USA) (KDD ’16). Association for Computing Machinery, New York, NY, USA, 1135–1144. https://doi.org/10.1145/2939672.2939778 ↗
  46. Peter Gordon Roetzel. 2018. Information overload in the information age: a review of the literature from business administration, business psychology, and related disciplines with a bibliometric approach and framework development. Business Research (2018), 1–44. find
  47. Kenneth R. Rose. 2005. On the effects of instruction in second language pragmatics. System 33, 3 (2005), 385–399. https://doi.org/10.1016/j.system.2005.06.003 Pragmatics in Instructed Language Learning. ↗
  48. Tabnine. 2023. Tabnine: AI assistant for software developers. https://www.tabnine.com/. ↗
  49. Thesaurus.com. 2023. Thesaurus.com: Synonyms and Antonyms of Words. https://www.thesaurus.com/. ↗
  50. Evangeline Marlos Varonis and Susan M Gass. 1985. Miscommunication in native/nonnative conversation. Language in society 14, 3 (1985), 327–343. ↗
  51. Helena Vasconcelos, Matthew Jörke, Madeleine Grunde-McLaughlin, Tobias Gerstenberg, Michael Bernstein, and Ranjay Krishna. 2023. Explanations Can Reduce Overreliance on AI Systems During Decision-Making. arXiv:2212.06823 [cs.HC] ↗
  52. Jane A Vignovic and Lori Foster Thompson. 2010. Computer-mediated cross-cultural collaboration: Attributing communication errors to the person versus the situation. Journal of Applied Psychology 95, 2 (2010), 265. ↗
  53. Wordtune. 2023. Wordtune - Rewrite Text in Seconds. https://www.wordtune.com. ↗
  54. Writefull. 2023. Writefull. https://www.writefull.com/. ↗
  55. Ming Yin, Jennifer Wortman Vaughan, and Hanna Wallach. 2019. Understanding the Effect of Accuracy on Trust in Machine Learning Models. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI ’19). Association for Computing Machinery, New York, NY, USA, 1–12. https://doi.org/10.1145/3290605.3300509 ↗
  56. Yunfeng Zhang, Q. Vera Liao, and Rachel K. E. Bellamy. 2020. Effect of Confidence and Explanation on Accuracy and Trust Calibration in AI-Assisted Decision Making. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (Barcelona, Spain) (FAT* ’20). Association for Computing Machinery, New York, NY, USA, 295–305. https://doi.org/10.1145/3351095.3372852 ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kim-2023-towards-explainable-ai-writing-assistants/