--- title: "A Dialogue System for Second Language — Morioka 2026" description: "LLM-based conversation partners for language learners usually target one preset difficulty level." authors: ["Taku Morioka", "Junya Takayama", "Tomoyuki Kajiwara"] year: 2026 venue: "Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue" source: "https://aclanthology.org/2026.sigdial-1.13/" doi: "10.18653/v1/2026.sigdial-1.13" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/papers/morioka-2026-dialogue-system-second-language-learning/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/morioka-2026-dialogue-system-second-language-learning/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2026](https://md.research.elamur.ai/free/papers/2026/) / A Dialogue System for Second Language Learning with Dynamic Adaptation to In-Dialogue Difficulty Changes # A Dialogue System for Second Language Learning with Dynamic Adaptation to In-Dialogue Difficulty Changes **Authors:** Taku Morioka, Junya Takayama, [Tomoyuki Kajiwara](https://md.research.elamur.ai/free/authors/tomoyuki-kajiwara/) · **Year:** 2026 · **Venue:** Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.sigdial-1.13/) · [PDF](https://aclanthology.org/2026.sigdial-1.13.pdf) · [DOI 10.18653/v1/2026.sigdial-1.13](https://doi.org/10.18653/v1/2026.sigdial-1.13) · [OpenAlex](https://openalex.org/W7214049319) **Citations:** 0 citations (openalex-snapshot · 2026-10-14) ## Summary LLM-based conversation partners for language learners usually target one preset difficulty level. A learner's own output, however, becomes simpler or richer depending on how familiar the topic is. The authors build systems that estimate the learner's current level from their utterances during a dialogue and match it. They also design a self-chat evaluation task. An LLM user model simulating a learner switches between a familiar and an unfamiliar topic, each with its own difficulty instruction, at the dialogue's midpoint. The final USER–CHAT pair of each half is first screened for quality and then compared for lexical and syntactic difficulty. Three control methods are tested against a plain instruction baseline: two prompting variants (Detailed and Single-Block) and DPO trained on preference pairs built from automatically ranked outputs of several models. In both Japanese and English, all proposed methods beat the baseline on average. Single-Block is the most dependable prompt-only option, and DPO performs best when fine-tuning is affordable. Human judgments in Japanese agree well with the automatic labels. The work provides a reusable benchmark and practical recipes for adaptive tutoring dialogue. ## Key points - **Four-label scheme:** each pair is labeled Reject, UserHarder, Same, or ChatHarder. - Reject is assigned if the CHAT reply is unnatural, ignores the user, or merely repeats them. - Otherwise, separate lexical and syntactic verdicts are merged into one of the other three labels. - **Difficulty measures:** - Lexical difficulty uses CEFR-style levels for English and JLPT levels (N5–N1) for Japanese. - Syntactic complexity is measured with Mean Hierarchical Distance (MHD). - **Prompting variants:** - Detailed places the instructions in a system message and supplies the earlier conversation as alternating user/assistant turns. - Single-Block packs both instructions and history into one user message. - **DPO and alignment results:** - Because of its training cost, DPO was applied only to Qwen3-14B. - DPO gave the smallest MHD gap; Single-Block gave the closest vocabulary match. - In a case study, a Baseline reply differed from the user's utterance by 0.55 MHD. - **Validation and code:** - Human evaluation (Japanese only) showed high Spearman correlation and reasonable Cohen's κ with LLM-as-a-Judge. - Code is available at github.com/EhimeNLP/DiAL2. *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: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{morioka2026dialogue, title = {{A Dialogue System for Second Language Learning with Dynamic Adaptation to In-Dialogue Difficulty Changes}}, author = {Morioka, Taku and Takayama, Junya and Kajiwara, Tomoyuki}, booktitle = {Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue}, year = {2026}, doi = {10.18653/v1/2026.sigdial-1.13}, url = {https://doi.org/10.18653/v1/2026.sigdial-1.13} } ``` - [← previous A Dataset of Psychiatric Hospital Notes with Temporal Information Annotations](https://md.research.elamur.ai/free/papers/miller-2026-dataset-psychiatric-hospital-notes-temporal/) - [next → A Fine-Grained Linguistic Evaluation of Low-Resource Luxembourgish–English MT](https://md.research.elamur.ai/free/papers/rehlinger-2026-fine-grained-linguistic-evaluation-low/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/papers/2026/) ## References 17 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Mahmoud Abdi Tabari and Yizhou Wang. 2022. Assessing Linguistic Complexity Features in L2 Writing: Understanding Effects of Topic Familiarity and Strategic Planning Within the Realm of Task Readiness. Assessing Writing, 52:100605. [find](https://scholar.google.com/scholar?q=Mahmoud+Abdi+Tabari+and+Yizhou+Wang%2E+2022%2E+Assessing+Linguistic+Complexity+Features+in+L2+Writing%3A+Understanding+Effects+of+Topic+Familiarity+and+Strategic+Planning+Within+the+Realm+of+Task+Readiness%2E) 2. Mina Almasi and Ross Deans Kristensen-McLachlan. 2025. Alignment Drift in CEFR-Prompted LLMs for Interactive Spanish Tutoring. In Proceedings of the 20th Workshop on Innovative Use of NLP for Building Educational Applications, pages 70–88. [find](https://scholar.google.com/scholar?q=Mina+Almasi+and+Ross+Deans+Kristensen%2DMcLachlan%2E+2025%2E+Alignment+Drift+in+CEFR%2DPrompted+LLMs+for+Interactive+Spanish+Tutoring%2E+In+Proceedings+of+the+20th+Workshop+on+Innovative+Use+of+NLP+for+Building) 3. Bram Bulté and Alex Housen. 2012. Defining and Operationalising L2 Complexity. [find](https://scholar.google.com/scholar?q=Bram+Bult%C3%A9+and+Alex+Housen%2E+2012%2E+Defining+and+Operationalising+L2+Complexity%2E) 4. Kazuki Fujii, Taishi Nakamura, Mengsay Loem, Hiroki Iida, Masanari Ohi, Kakeru Hattori, Hirai Shota, Sakae Mizuki, Rio Yokota, and Naoaki Okazaki. 2024. Continual Pre-Training for Cross-Lingual LLM Adaptation: Enhancing Japanese Language Capabilities. In First Conference on Language Modeling. [find](https://scholar.google.com/scholar?q=Kazuki+Fujii%2C+Taishi+Nakamura%2C+Mengsay+Loem%2C+Hiroki+Iida%2C+Masanari+Ohi%2C+Kakeru+Hattori%2C+Hirai+Shota%2C+Sakae+Mizuki%2C+Rio+Yokota%2C+and+Naoaki+Okazaki%2E+2024%2E+Continual+Pre%2DTraining+for+Cross%2DLingual+LLM+Ad) [our card](https://md.research.elamur.ai/free/papers/fujii-2024-continual-pre-training-cross-lingual/) 5. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient Memory Management for Large Language Model Serving with PagedAttention. In Proceedings of the 29th Symposium on Operating Systems Principles, page 611–626. [find](https://scholar.google.com/scholar?q=Woosuk+Kwon%2C+Zhuohan+Li%2C+Siyuan+Zhuang%2C+Ying+Sheng%2C+Lianmin+Zheng%2C+Cody+Hao+Yu%2C+Joseph+Gonzalez%2C+Hao+Zhang%2C+and+Ion+Stoica%2E+2023%2E+Efficient+Memory+Management+for+Large+Language+Model+Serving+with+Page) 6. Llama Team. 2024. The Llama 3 Herd of Models. arXiv:2407.21783. [find](https://scholar.google.com/scholar?q=Llama+Team%2E+2024%2E+The+Llama+3+Herd+of+Models%2E+arXiv%3A2407%2E21783%2E) 7. Rina Miyata, Hyuga Koretak, Hiroki Yamauchi, Daiki Yanamoto, Tomoyuki Kajiwara, Takashi Ninomiya, and Yasuhiro Nishiwaki. 2024. MATCHA: Parallel Corpus for Japanese Text Simplification Based on Professionally Simplified Articles. Journal of Natural Language Processing, 31(2):590–609. [find](https://scholar.google.com/scholar?q=Rina+Miyata%2C+Hyuga+Koretak%2C+Hiroki+Yamauchi%2C+Daiki+Yanamoto%2C+Tomoyuki+Kajiwara%2C+Takashi+Ninomiya%2C+and+Yasuhiro+Nishiwaki%2E+2024%2E+MATCHA%3A+Parallel+Corpus+for+Japanese+Text+Simplification+Based+on+Profes) 8. Rina Miyata, Toru Urakawa, Hideaki Tamori, and Tomoyuki Kajiwara. 2025. Unsupervised Sentence Readability Estimation Based on Parallel Corpora for Text Simplification. In Proceedings of the 20th Workshop on Innovative Use of NLP for Building Educational Applications, pages 499–504. [find](https://scholar.google.com/scholar?q=Rina+Miyata%2C+Toru+Urakawa%2C+Hideaki+Tamori%2C+and+Tomoyuki+Kajiwara%2E+2025%2E+Unsupervised+Sentence+Readability+Estimation+Based+on+Parallel+Corpora+for+Text+Simplification%2E+In+Proceedings+of+the+20th+Works) 9. OpenAI. 2025. gpt-oss-120b & gpt-oss-20b Model Card. arXiv:2508.10925. [find](https://scholar.google.com/scholar?q=OpenAI%2E+2025%2E+gpt%2Doss%2D120b+%26+gpt%2Doss%2D20b+Model+Card%2E+arXiv%3A2508%2E10925%2E) [our card](https://md.research.elamur.ai/free/papers/agarwal-2025-gpt-oss-120b-gpt-oss/) 10. Dominik Glandorf, Peng Cui, Detmar Meurers, and Mrinmaya Sachan. 2025. Grammar Control in Dialogue Response Generation for Language Learning Chatbots. In Proceedings of the 2025 Conference Mostafa Papi and Hassan Khajavy. 2023. Second Language Anxiety: Construct, Effects, and Sources. Annual Review… [find](https://scholar.google.com/scholar?q=Dominik+Glandorf%2C+Peng+Cui%2C+Detmar+Meurers%2C+and+Mrinmaya+Sachan%2E+2025%2E+Grammar+Control+in+Dialogue+Response+Generation+for+Language+Learning+Chatbots%2E+In+Proceedings+of+the+2025+Conference+Mostafa+Pap) 11. Xuyan Qiu. 2020. Functions of Oral Monologic Tasks: Effects of Topic Familiarity on L2 Speaking Performance. Language Teaching Research, 24(6):745– 764. [find](https://scholar.google.com/scholar?q=Xuyan+Qiu%2E+2020%2E+Functions+of+Oral+Monologic+Tasks%3A+Effects+of+Topic+Familiarity+on+L2+Speaking+Performance%2E+Language+Teaching+Research%2C+24%286%29%3A745%E2%80%93+764%2E) 12. Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. 2023. Direct Preference Optimization: Your Language Model Is Secretly a Reward Model. In Proceedings of the 37th International Conference on Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Rafael+Rafailov%2C+Archit+Sharma%2C+Eric+Mitchell%2C+Stefano+Ermon%2C+Christopher+D%2E+Manning%2C+and+Chelsea+Finn%2E+2023%2E+Direct+Preference+Optimization%3A+Your+Language+Model+Is+Secretly+a+Reward+Model%2E+In+Proceed) 13. Hiroaki Sugiyama, Masahiro Mizukami, Tsunehiro Arimoto, Hiromi Narimatsu, Yuya Chiba, Hideharu Nakajima, and Toyomi Meguro. 2023. Empirical Analysis of Training Strategies of Transformer-Based Japanese Chit-Chat Systems. In 2022 IEEE Spoken Language Technology Workshop, pages 685–691. [find](https://scholar.google.com/scholar?q=Hiroaki+Sugiyama%2C+Masahiro+Mizukami%2C+Tsunehiro+Arimoto%2C+Hiromi+Narimatsu%2C+Yuya+Chiba%2C+Hideharu+Nakajima%2C+and+Toyomi+Meguro%2E+2023%2E+Empirical+Analysis+of+Training+Strategies+of+Transformer%2DBased+Japanes) 14. Kazuma Takaoka, Sorami Hisamoto, Noriko Kawahara, Miho Sakamoto, Yoshitaka Uchida, and Yuji Matsumoto. 2018. Sudachi: A Japanese Tokenizer for Business. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation. [find](https://scholar.google.com/scholar?q=Kazuma+Takaoka%2C+Sorami+Hisamoto%2C+Noriko+Kawahara%2C+Miho+Sakamoto%2C+Yoshitaka+Uchida%2C+and+Yuji+Matsumoto%2E+2018%2E+Sudachi%3A+A+Japanese+Tokenizer+for+Business%2E+In+Proceedings+of+the+Eleventh+International+Co) 15. Benjamin Warner, Antoine Chaffin, Benjamin Clavié, Orion Weller, Oskar Hallström, Said Taghadouini, Alexis Gallagher, Raja Biswas, Faisal Ladhak, Tom Aarsen, Griffin Thomas Adams, Jeremy Howard, and Iacopo Poli. 2025. Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory E… [find](https://scholar.google.com/scholar?q=Benjamin+Warner%2C+Antoine+Chaffin%2C+Benjamin+Clavi%C3%A9%2C+Orion+Weller%2C+Oskar+Hallstr%C3%B6m%2C+Said+Taghadouini%2C+Alexis+Gallagher%2C+Raja+Biswas%2C+Faisal+Ladhak%2C+Tom+Aarsen%2C+Griffin+Thomas+Adams%2C+Jeremy+Howard%2C+and+I) [our card](https://md.research.elamur.ai/free/papers/warner-2024-smarter-better-faster-longer-modern/) 16. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, and 3 others. 2020. Transf… [find](https://scholar.google.com/scholar?q=Thomas+Wolf%2C+Lysandre+Debut%2C+Victor+Sanh%2C+Julien+Chaumond%2C+Clement+Delangue%2C+Anthony+Moi%2C+Pier%2Dric+Cistac%2C+Tim+Rault%2C+Remi+Louf%2C+Morgan+Funtowicz%2C+Joe+Davison%2C+Sam+Shleifer%2C+Patrick+von+Platen%2C+Clara+) [our card](https://md.research.elamur.ai/free/papers/wolf-2020-transformers-state-art-natural-language/) 17. Kevin Yang and Dan Klein. 2021. FUDGE: Controlled Text Generation with Future Discriminators. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3511–3535. [find](https://scholar.google.com/scholar?q=Kevin+Yang+and+Dan+Klein%2E+2021%2E+FUDGE%3A+Controlled+Text+Generation+with+Future+Discriminators%2E+In+Proceedings+of+the+2021+Conference+of+the+North+American+Chapter+of+the+Association+for+Computational+L) [our card](https://md.research.elamur.ai/free/papers/yang-2021-fudge-controlled-text-generation-future/)