Papers by Yusuke Miyao

A Multi-Perspective Analysis of Memorization in Large Language Models (2024.emnlp-main)

Challenge:	Combinatory Categorial Grammar (CCG) is a lexicalized grammar formalism that can capture both syntactic and semantic information.
Approach:	They propose an algorithm to convert UD treebanks to CCG treebank and propose future extensions.
Outcome:	The proposed algorithm performs lexical, sentential, and syntactic rule coverage analysis, as well as CCG parsing experiments.

Copied to clipboard

Bowen Chen, Namgi Han, Yusuke Miyao

Learning to Select, Track, and Generate for Data-to-Text (P19-1)

Challenge:	Large Language Models (LLMs) can generate the same sequences contained in the pre-train corpus, known as memorization.
Approach:	They analyze the relationship between memorization and outputs from Large Language Models (LLMs) they show a sudden drop and increase in the frequency of input tokens when generating memorized/unmemorized sequences .
Outcome:	The proposed model can generate the same sequences contained in the pre-train corpus, and it can predict unmemorized tokens.

Copied to clipboard

Hayate Iso, Yui Uehara, Tatsuya Ishigaki, Hiroshi Noji, Eiji Aramaki, Ichiro Kobayashi, Yusuke Miyao, Naoaki Okazaki, Hiroya Takamura

Text Generation

Challenge:	Existing models often refer to the same data record multiple times.
Approach:	They propose a data-to-text generation model with two modules, one for tracking and the other for text generation.
Outcome:	The proposed model outperforms existing models even without writer information in all evaluation metrics and contributes to content planning and surface realization.

Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality (2025.emnlp-main)

Copied to clipboard

Yuto Harada, Yusuke Yamauchi, Yusuke Oda, Yohei Oseki, Yusuke Miyao, Yu Takagi

Rethinking Offensive Text Detection as a Multi-Hop Reasoning Problem (2022.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have greatly improved natural language understanding and generation.
Approach:	They train a wide range of base models on a variety of datasets including code generation, mathematical reasoning, and general-domain tasks.
Outcome:	The results show that training–task synergies persist across all models while others vary substantially, emphasizing the importance of model-specific strategies.

Copied to clipboard

Qiang Zhang, Jason Naradowsky, Yusuke Miyao

Analyzing Word Embedding Through Structural Equation Modeling (2020.lrec-1)

Challenge:	Existing methods of offensive text detection perform poorly when asked to detect implicitly offensive statements . a dataset based on SLIGHT provides a framework for implicit offensive text identification .
Approach:	They propose a dataset to support the task of implicit offensive text detection in dialogues . they show that reasoning is crucial for understanding this broader class of offensive utterances - SLIGHT .
Outcome:	The proposed model achieves 11% accuracy in implicit offensive text detection tasks . the proposed model can be used to identify toxic speech in specific domains .

Copied to clipboard

Namgi Han, Katsuhiko Hayashi, Yusuke Miyao

An empirical analysis of existing systems and datasets toward general simple question answering (2020.coling-main)

Challenge:	Existing studies have shown that word embedding improves accuracy on NLP tasks.
Approach:	They propose a causal diagram based on the evaluation results of word embeddings using partial least squares path modeling.
Outcome:	The proposed model proves that word embedding contributes to solving downstream tasks.

Copied to clipboard

Namgi Han, Goran Topic, Hiroshi Noji, Hiroya Takamura, Yusuke Miyao

Robustness in NLP

Challenge:	evaluators of simple factoid question answering using different datasets are not able to solve SimpleQuestions.
Approach:	They evaluate the progress of the field toward solving simple factoid questions over a knowledge base.
Outcome:	The proposed model is nearly solved on the most popular dataset, but not on the robustness of existing systems.

Transferability of Syntax-Aware Graph Neural Networks in Zero-Shot Cross-Lingual Semantic Role Labeling (2024.findings-emnlp)

Copied to clipboard

Rachel Devianti, Yusuke Miyao

Inducing Temporal Relations from Time Anchor Annotation (N18-1)

Challenge:	Existing studies in cross-lingual semantic role labeling (SRL) lack a comprehensive analysis of their network selection.
Approach:	They compare the transferability of graph neural network-based models with universal dependency trees to English and 23 target languages.
Outcome:	The proposed models perform better in resource-poor languages than in resource rich ones.

Copied to clipboard

Fei Cheng, Yusuke Miyao

Challenge:	Existing methods for judging temporal relations are limited to “salient” event pairs or on pairs in a fixed window of sentences.
Approach:	They propose a new method to obtain temporal relations from absolute time value (a.k.a. time anchors) they start from time anchor for events and time expressions and induced temporal relation annotations automatically .
Outcome:	The proposed method shows that it requires less annotation effort and induces inter-sentence relations easily.

Ask an Expert: Leveraging Language Models to Improve Strategic Reasoning in Goal-Oriented Dialogue Models (2023.findings-acl)

Copied to clipboard

Qiang Zhang, Jason Naradowsky, Yusuke Miyao

Mind the Gap Between Conversations for Improved Long-Term Dialogue Generation (2023.findings-emnlp)

Challenge:	Existing dialogue models may encounter scenarios which are not well-represented in the training data and produce unnatural, inappropriate, or unhelpful responses.
Approach:	They propose a framework in which a model is trained with access to an "expert" they propose to optimize the model to selectively utilize (or ignore) advice given context and dialogue history.
Outcome:	The proposed framework improves quality across all expert sizes and with fewer parameters than the dialogue model itself.

Copied to clipboard

Qiang Zhang, Jason Naradowsky, Yusuke Miyao

StoryER: Automatic Story Evaluation via Ranking, Rating and Reasoning (2022.emnlp-main)

Challenge:	a gap between conversations can be weeks, months or years, and dialogue systems which do not explicitly model time may generate unnatural responses.
Approach:	They propose to model the passage of time between conversations by exposing time information to a multi-session dialogue dataset and comparing different representations of time and event progress.
Outcome:	The proposed model is based on a real-time dataset showing that it can predict topics and information gained from conversations over a long time span.

Copied to clipboard

Hong Chen, Duc Vo, Hiroya Takamura, Yusuke Miyao, Hideki Nakayama

Universal Dependencies Version 2 for Japanese (L18-1)

Challenge:	Existing automatic story evaluation methods place a premium on story lexical level coherence, deviating from human preference.
Approach:	They propose a novel Story Evaluation method that mimics human preference when judging a story . the model is based on a well-annotated dataset and a longformer-encoder-decoder .
Outcome:	The proposed method is applicable to machine-generated and human-written stories.

Copied to clipboard

Masayuki Asahara, Hiroshi Kanayama, Takaaki Tanaka, Yusuke Miyao, Sumire Uematsu, Shinsuke Mori, Yuji Matsumoto, Mai Omura, Yugo Murawaki

Does My Rebuttal Matter? Insights from a Major NLP Conference (N19-1)

Challenge:	UD Japanese resources are built on automatic conversion from several treebanks.
Approach:	They propose to port the word delimitation, POS, and syntactic relations of existing treebanks to UD Japanese . they discuss the issues of the UD scheme found through porting of the Japanese language .
Outcome:	The proposed UD Japanese resources are based on automatic conversion from treebanks.

Copied to clipboard

Yang Gao, Steffen Eger, Ilia Kuznetsov, Iryna Gurevych, Yusuke Miyao

The Impact of Language on Arithmetic Proficiency: A Multilingual Investigation with Cross-Agent Checking Computation (2024.naacl-short)

Challenge:	Peer review is a core element of the scientific process, but few studies have evaluated its properties empirically.
Approach:	They propose to use peer review to assess the effectiveness of rebuttal phase in NLP conferences.
Outcome:	The proposed task predicts after-rebuttal scores from initial reviews and author responses.

Copied to clipboard

Chung-Chi Chen, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao

Multilinguality

Challenge:	Large language models (LLMs) have garnered significant attention over the past year . previous studies have evaluated LLMs' performance in solving math word problems, but there is little discussion on whether they comprehend the operations they generate.
Approach:	They challenge the notion that arithmetic is language-independent and compare models with cross-agent collaborations to find significant limitations in their performance.
Outcome:	The proposed model outperforms collaborative approaches in basic arithmetic tasks.

Integrating Headedness Information into an Auto-generated Multilingual CCGbank for Improved Semantic Interpretation (2024.lrec-main)

Copied to clipboard

Tu-Anh Tran, Yusuke Miyao

Language Model Based Unsupervised Dependency Parsing with Conditional Mutual Information and Grammatical Constraints (2024.naacl-long)

Challenge:	Combinatory Categorial Grammar is a grammar formalism that provides a transparent interface between syntax and semantics.
Approach:	They propose an algorithm that adds semantic representations to existing CCG derivations by combining them with predefined combinatory rules.
Outcome:	The proposed method produces bare CCG derivations without any accompanying semantic representations and limits its general applicability.

Copied to clipboard

Junjie Chen, Xiangheng He, Yusuke Miyao

An Empirical Investigation of Error Types in Vietnamese Parsing (C18-1)

Challenge:	Existing methods for unsupervised dependency parsing use difficult to interpret dependence scores.
Approach:	They propose to use Conditional Mutual Information (CMI) to measure bi-lexical dependence and incorporate grammatical constraints into unsupervised parsing.
Outcome:	The proposed model outperforms state-of-the-art models and grammar-based models in five languages and eight datasets.

Copied to clipboard

Quy Nguyen, Yusuke Miyao, Hiroshi Noji, Nhung Nguyen

Fiction-Writing Mode: An Effective Control for Human-Machine Collaborative Writing (2023.eacl-main)

Challenge:	Syntactic parsing improves the quality of natural language processing tasks.
Approach:	They evaluated Vietnamese Treebank model to find most suitable parsing method . they found that Vietnamese parsers produced limited training data and POS errors .
Outcome:	The proposed method improves the parsing quality in Vietnamese . the results highlight three possible sources of parser errors .

Copied to clipboard

Wenjie Zhong, Jason Naradowsky, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao

Text Generation

Challenge:	Large-scale pre-trained language models (PLMs) have demonstrated an exceptional aptitude for generating text with an exceptional degree of fluency and structure.
Approach:	They propose to integrate writing skills curricula into human-machine collaborative writing scenarios by adding writing modes as a control for text generation models.
Outcome:	The proposed model can be used to generate narrative fiction with a high level of accuracy and similarity with the professionally written target story.

GADFA: Generator-Assisted Decision-Focused Approach for Opinion Expressing Timing Identification (2025.coling-main)

Copied to clipboard

Chung-Chi Chen, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao, Hsin-Hsi Chen

Challenge:	Existing models generate text on demand, but in real-life situations, individuals do not continuously generate text or voice opinions.
Approach:	They propose a novel task to identify news-triggered opinion expressing timing by using a dataset generated by professional stock analysts.
Outcome:	The proposed model can generate opinion on stock analysts' actions and improves performance in various opinion understanding tasks.

Who Said What: Formalization and Benchmarks for the Task of Quote Attribution (2024.lrec-main)

Copied to clipboard

Wenjie Zhong, Jason Naradowsky, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao

Challenge:	Existing methods for quote attribution are poorly understood, despite advances in research . previous approaches have used hand-crafted features to identify speaker names .
Approach:	They formalize the task of quote attribution and establish a basis for comparison . they compare CEQA and ChatGPT models on available datasets in both English and Chinese .
Outcome:	The proposed model outperforms all supervised methods on English and Chinese datasets.

Collection and Analysis of Travel Agency Task Dialogues with Age-Diverse Speakers (2022.lrec-1)

Copied to clipboard

Michimasa Inaba, Yuya Chiba, Ryuichiro Higashinaka, Kazunori Komatani, Yusuke Miyao, Takayuki Nagai

How Much Do Large Language Models Know about Human Motion? A Case Study in 3D Avatar Control (2025.findings-emnlp)

Challenge:	Using deep neural networks, task-oriented dialogue systems can be used to generate an appropriate response to users' inputs.
Approach:	They collected a multimodal dialogue corpus with a wide range of speaker ages and set up a dialogue task based on travel . results suggest adult speakers have more independent opinions, older speakers express opinions more frequently compared with other age groups, and operators expressed a smile more frequently to minor speakers.
Outcome:	The results show that adult speakers have more independent opinions, the older speakers express their opinions more frequently compared with other age groups, and the operators expressed a smile more frequently to the minor speakers.

Copied to clipboard

Kunhang Li, Jason Naradowsky, Yansong Feng, Yusuke Miyao

The Imperfective Paradox in Large Language Models (2026.acl-long)

Challenge:	a new study explores the human motion knowledge of Large Language Models (LLMs) using 3D avatar control.
Approach:	They use 20 representative motion instructions to interpolate LLMs into avatar animations . they find they are strong at interpreting high-level body movements but struggle with precise body part positioning .
Outcome:	The proposed model is strong at interpreting high-level body movements but struggles with precise body part positioning.

Copied to clipboard

Bolei Ma, Yusuke Miyao

Introducing Spatial Information and a Novel Evaluation Scheme for Open-Domain Live Commentary Generation (2024.findings-emnlp)

Challenge:	Existing models rely on surface-level probabilistic heuristics to grasp compositional semantics of events . authors: current open-weight models operate as predictive narrative engines rather than faithful reasoners .
Approach:	They propose a diagnostic dataset to probe the imperfective paradox . they uncover a pervasive Teleological Bias in open-weight models .
Outcome:	The proposed dataset reveals a pervasive Teleological Bias in open-weight models . the findings suggest that these models operate as predictive narrative engines rather than faithful reasoners .

Copied to clipboard

Erica Shimomoto, Edison Marrese-Taylor, Ichiro Kobayashi, Hiroya Takamura, Yusuke Miyao

Multimodality

Challenge:	Compared to domain-specific work in this task, this task proved particularly challenging due to the absence of domain- specific features.
Approach:	They propose an utterance generation model with a novel spatial graph that integrates spatial information to deal with the open-domain characteristics of the commentaries and significantly improves performance.
Outcome:	The proposed model significantly improves performance in the open-domain live commentary generation task.

Evaluating Intention Detection Capability of Large Language Models in Persuasive Dialogues (2024.acl-long)

Copied to clipboard

Hiromasa Sakurai, Yusuke Miyao

Modeling Syntactic-Semantic Dependency Correlations in Semantic Role Labeling Using Mixture Models (2022.acl-long)

Challenge:	Existing studies measure the intention detection capability of machine learning models without considering the conversational history.
Approach:	They modified existing persuasive conversation datasets and created a dataset using a multiple-choice paradigm to evaluate LLMs' intention detection capability.
Outcome:	The proposed model can detect speakers' intentions well in persuasive multi-turn dialogs using the largest available Large Language Models (LLMs).

Copied to clipboard

Junjie Chen, Xiangheng He, Yusuke Miyao

How a Bilingual LM Becomes Bilingual: Tracing Internal Representations with Sparse Autoencoders (2025.findings-emnlp)

Challenge:	Existing methods for SRL identify semantic dependencies that specify the semantic role of arguments in relation to predicates.
Approach:	They propose a mixture model-based end-to-end method to model syntactic-semantic dependency correlation in Semantic Role Labeling.
Outcome:	The proposed method improves performance in English, German, and Spanish . it achieves small but statistically significant improvement over baseline methods .

Copied to clipboard

Tatsuro Inaba, Go Kamoda, Kentaro Inui, Masaru Isonuma, Yusuke Miyao, Yohei Oseki, Yu Takagi, Benjamin Heinzerling

A Comprehensive Evaluation of Inductive Reasoning Capabilities and Problem Solving in Large Language Models (2024.findings-eacl)

Challenge:	Using sparse autoencoders, we explore how bilingual language models develop complex internal representations.
Approach:	They employ sparse autoencoders to analyze bilingual language models' internal representations.
Outcome:	The proposed method integrates decomposed representations from a fully trained model into a mid-training model.

Copied to clipboard

Chen Bowen, Rune Sætre, Yusuke Miyao

Syntactic and Semantic Uniformity for Semantic Parsing and Task-Oriented Dialogue Systems (2022.findings-emnlp)

Challenge:	Inductive reasoning is fundamental to both human and artificial intelligence.
Approach:	They evaluated the inductive reasoning abilities of current Large Language Models (LLMs) and their performance on symbolic tasks.
Outcome:	The proposed models fail on symbolic tasks and show that chain-of-thought prompts help them by decomposing the problem-solving process, but the LLMs learn limitedly.

Copied to clipboard

Bowen Chen, Yusuke Miyao

Natural Language Interfaces

Challenge:	Existing approaches to model natural language use pre-trained language models, but little attention has been paid to the representation of machine-readable formats.
Approach:	They propose a data representation framework for semantic parsing and task-oriented dialogue systems . they define a meta grammar for syntactically uniform representations and translate semantically equivalent functions into a uniform vocabulary.
Outcome:	The proposed representation improves accuracy and allows for transfer learning across datasets.

Learning with Contrastive Examples for Data-to-Text Generation (2020.coling-main)

Copied to clipboard

Yui Uehara, Tatsuya Ishigaki, Kasumi Aoki, Hiroshi Noji, Keiichi Goshima, Ichiro Kobayashi, Hiroya Takamura, Yusuke Miyao

Text Generation

Challenge:	Existing models for data-to-text generation generate fluent but sometimes incorrect sentences . Existing studies show that using contrastive examples improves the ability of generating sentences with better lexical choice without degrading the fluency.
Approach:	They propose to use models trained on incorrect sentences and learning methods that exploit contrastive examples to reduce such errors.
Outcome:	The proposed models generate fluent sentences but often have problematic ones in terms of correctness.

A Statistical and Multi-Perspective Revisiting of the Membership Inference Attack in Large Language Models (2025.acl-long)

Copied to clipboard

Bowen Chen, Namgi Han, Yusuke Miyao

Open-domain Video Commentary Generation (2022.emnlp-main)

Challenge:	Membership Inference Attack (MIA) is a method that differentiates trained (member) and untrained (non-member) data.
Approach:	They used thousands of experiments to examine membership inference attacks from different settings and then revisited them with thousands of different methods.
Outcome:	The proposed methods outperform baselines in the study and improve with model size and varies with domains.

Copied to clipboard

Edison Marrese-Taylor, Yumi Hamazono, Tatsuya Ishigaki, Goran Topić, Yusuke Miyao, Ichiro Kobayashi, Hiroya Takamura

Visual Data in NLP

Challenge:	Existing approaches to generate live commentary on specific domains have been limited.
Approach:	They propose to generate live commentary from transcribed videos in an open-domain setting . they propose to use well-known neural architectures to build models based on transcriptions .
Outcome:	The proposed model is based on well-known neural architectures and based off existing models.

Towards Parameter-Efficient Integration of Pre-Trained Language Models In Temporal Video Grounding (2023.findings-acl)

Copied to clipboard

Erica Kido Shimomoto, Edison Marrese-Taylor, Hiroya Takamura, Ichiro Kobayashi, Hideki Nakayama, Yusuke Miyao

Universal Dependencies for Amharic (L18-1)

Challenge:	Recent studies have improved query inputs with pre-trained language models, but the effects of this integration are unclear.
Approach:	They propose to integrate query sentences with pre-trained language models to train TVG models.
Outcome:	The proposed model integrates query sentences with pre-trained language models at cost of more expensive training.

Copied to clipboard

Binyam Ephrem Seyoum, Yusuke Miyao, Baye Yimam Mekonnen

Unsupervised Parsing by Searching for Frequent Word Sequences among Sentences with Equivalent Predicate-Argument Structures (2024.findings-acl)

Challenge:	Amharic is a morphologically rich language with a dependency relation between orthographic words and lexical categories.
Approach:	They propose to create an Amharic Dependency Treebank by POS tagging, morphological information and dependency relations.
Outcome:	The proposed treebanks are based on 1,096 sentences and are able to parse Amharic.

Copied to clipboard

Junjie Chen, Xiangheng He, Danushka Bollegala, Yusuke Miyao

Improving Numeracy by Input Reframing and Quantitative Pre-Finetuning Task (2023.findings-eacl)

Challenge:	Unsupervised constituency parsing focuses on identifying word sequences that form a syntactic unit (i.e., constituents) in target sentences.
Approach:	They propose a frequency-based parser that computes the span-overlap score as the word sequence’s frequency in the PAS-equivalent sentence set and identifies the constituent structure by finding a constituent tree with the maximum span- overlap score.
Outcome:	The proposed method outperforms existing unsupervised parsers in eight out of ten languages and is more accurate than previous methods.

Copied to clipboard

Chung-Chi Chen, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao