KFKotaro Funakoshi

Kotaro Funakoshi

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Kotaro Funakoshi has a total of 89 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    SignHAI 2026: Moving Towards Inclusive Sign Language Dialogue and Turn-taking

    by Koji Inoue, Kotaro Funakoshi, Takao Obi, Kazuhiro Nakadai, Yutaka Osugi, Mayumi Bono

  • Open Access

    Information extraction for materials science using large language models: A survey of recent advances and new frontiers

    by Yutong Duan, Ying Zhang, Runjia Yu, Dongyuan Li, Satoshi Kosugi, Manabu Okumura, Kotaro Funakoshi


  • Open Access

    Unlocking High-Frequency Temporal Visual Cues for Personality Traits Recognition

    by Yusong Wang, Keyu Mao, Minghao Shao, Zheyuan Gu, Qingsong Zhao, Kotaro Funakoshi

    Personality traits recognition (PTR) plays a pivotal role in developing intuitive human-computer interaction systems. Visual-based PTR methods have achieved notable success by leveraging dynamic cues such as facial expressions and body movements. However, they typically rely on sparse frame sampling to maintain computational efficiency. This strategy inherently discards high-frequency temporal visual cues such as micro-expressions and instantaneous gaze shifts, which are important for accurate personality traits recognition. To address this limitation, we propose FrePTR, a framework that captures and integrates high-frequency temporal visual cues for visual-based PTR. Unlike existing motion- or event-based representations, which mainly quantify geometric displacement or generic intensity variation, FrePTR introduces a retina-inspired dual-pathway extractor to model high-frequency cues for PTR. Motivated by the observation that personality impressions from visual behavior are formed through the eye, whose retina captures rapid temporal transients, the extractor decomposes high-frequency PTR cues into two complementary types: cone-inspired chromatic physiological transients and rod-inspired achromatic motion dynamics. Since the high-frequency and low-frequency streams originate from different signal sources, they exhibit heterogeneous feature distributions and reliability patterns, which makes feature-level fusion difficult. We therefore develop a heteroscedastic uncertainty-guided late fusion mechanism for reliability-aware integration at the decision level. Extensive experiments on the ChaLearn First Impressions and ELEA datasets show that FrePTR outperforms advanced visual-only methods and remains competitive with multimodal methods, indicating the benefit of incorporating high-frequency temporal visual cues into PTR.

  • Open Access

    E²MP: Emotion-Enhanced Modality Fusion With Contrastive Learning for Multi-Modal Personality Traits Recognition

    by Yusong Wang, Dongyuan Li, Kotaro Funakoshi, Manabu Okumura

    Multi-modal personality traits recognition is crucial for advancing human-computer interaction, yet current methods struggle with extracting distinguishable features, mitigating inter-modal noise, and aligning heterogeneous modalities, thereby limiting accuracy. To address these challenges, we propose an emotion-enhanced multi-modal fusion model leveraging supervised contrastive learning for personality traits recognition. Specifically, we use supervised contrastive learning to learn highly distinguishable representations of modalities. Capitalizing on the intrinsic link between emotions and personality, we introduce a novel emotion-enhanced multi-modal fusion mechanism. This mechanism strategically guides the fusion process, reducing inter-modal noise and sharpening feature alignment across modalities. Each mechanism provides flexible designs for diverse personality traits recognition scenarios, optimizing both feature extraction and fusion for superior accuracy. In addition, we explore the interrelations among cultural contexts, emotions, and personality traits. Extensive experiments on three diverse datasets validate the superior performance and generalization of our proposed method.

  • Open Access

    Live Football Commentary System Providing Background Information

    by Yuichiro Mori, Chikara Tanaka, Aru Maekawa, Satoshi Kosugi, Tatsuya Ishigaki, Kotaro Funakoshi, Hiroya Takamura, Manabu Okumura

    Yuichiro Mori, Chikara Tanaka, Aru Maekawa, Satoshi Kosugi, Tatsuya Ishigaki, Kotaro Funakoshi, Hiroya Takamura, Manabu Okumura. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations). 2025.

  • Open Access

    Can Respiration Make Spoken Interactions Better?

    by Takao Obi, Kotaro Funakoshi

    With the growing use of agents such as chatbots in daily life, enhancing human-agent spoken interactions has become a critical area of research. This study explores the potential of respiratory information to enhance these interactions. We implemented two functions: speech collision avoidance based on speech onset prediction using human respiratory information, and pseudo-respiration presentation to humans by the vertical movements of a robot. We developed a conversational robot incorporating these functions and conducted experiments with 26 human participants. The experimental results confirmed these functions contributed to the smoothness of interactions.

  • FINE-LMT: Fine-Grained Feature Learning for Multi-modal Machine Translation

    by Yusong Wang, Ying Zhang, Dongyuan Li, Jialun Shen, Yicheng Xu, Mingkun Xu, Kotaro Funakoshi, Manabu Okumura

  • Open Access

    Active Learning with Task Adaptation Pre-training for Speech Emotion Recognition

    by Dongyuan Li, Ying Zhang, Yusong Wang, Funakoshi Kataro, Manabu Okumura

    Speech emotion recognition (SER) has garnered increasing attention due to its wide range of applications in various fields, including human-machine interaction, virtual assistants, and mental health assistance. However, existing SER methods often overlook the information gap between the pre-training speech recognition task and the downstream SER task, resulting in sub-optimal performance. Moreover, current methods require much time for fine-tuning on each specific speech dataset, such as IEMOCAP, which limits their effectiveness in real-world scenarios with large-scale noisy data. To address these issues, we propose an active learning (AL)-based fine-tuning framework for SER, called \textsc{After}, that leverages task adaptation pre-training (TAPT) and AL methods to enhance performance and efficiency. Specifically, we first use TAPT to minimize the information gap between the pre-training speech recognition task and the downstream speech emotion recognition task. Then, AL methods are employed to iteratively select a subset of the most informative and diverse samples for fine-tuning, thereby reducing time consumption. Experiments demonstrate that our proposed method \textsc{After}, using only 20\% of samples, improves accuracy by 8.45\% and reduces time consumption by 79\%. The additional extension of \textsc{After} and ablation studies further confirm its effectiveness and applicability to various real-world scenarios. Our source code is available on Github for reproducibility. (https://github.com/Clearloveyuan/AFTER).

  • Open Access

    DiLM: Distilling Dataset into Language Model for Text-level Dataset Distillation

    by Aru Maekawa, Satoshi Kosugi, Kotaro Funakoshi, Manabu Okumura

    Dataset distillation aims to compress a training dataset by creating a small number of informative synthetic samples such that neural networks trained on them perform as well as those trained on the original training dataset. Current text dataset distillation methods create each synthetic sample as a sequence of word embeddings instead of a text to apply gradient-based optimization; however, such embedding-level distilled datasets cannot be used for training other models whose word embedding weights are different from the model used for distillation. To address this issue, we propose a novel text dataset distillation approach, called Distilling dataset into Language Model (DiLM), which trains a language model to generate informative synthetic training samples as text data, instead of directly optimizing synthetic samples. We evaluated DiLM on various text classification datasets and showed that distilled synthetic datasets from DiLM outperform those from current coreset selection methods. DiLM achieved remarkable generalization performance in training different types of models and in-context learning of large language models. Our code will be available at https://github.com/arumaekawa/DiLM.