Makoto P. Kato

Benjamin Clavié, Sean Lee, Aamir Shakir et al.

We propose Latent Terms, a method revealing that models trained for dense retrieval, whether single- or multi-vector, learn representations that can trivially be decomposed into retrieval-ready sparse features. When trained on frozen retrievers, Sparse Autoencoders without any retrieval-specific adjustments extract a latent vocabulary with approximately Zipfian collection statistics, directly suitable for classical sparse retrieval scoring via BM25. This approach enables sparse retrieval while requiring no learned expansion objective or sparse retrieval supervision whatsoever, and can be readily applied to any dense retriever. Latent Terms is able to match or outperform single-vector scoring methods from its own base model as well as comparable SPLADE variants. In addition, it substantially outperforms its base model on LIMIT, a task specifically designed to highlight the failures of single-vector retrieval. Overall, our results highlight that neural retrievers contain more expressive and indexable structure than their default scoring functions expose, but that other methods can nonetheless be leveraged.

Benjamin Clavié, Atoof Shakir, Jonah Turner et al.

Multimodal Information Retrieval has made significant progress in recent years, leveraging the increasingly strong multimodal abilities of deep pre-trained models to represent information across modalities. Music Information Retrieval (MIR), in particular, has considerably increased in quality, with neural representations of music even making its way into everyday life products. However, there is a lack of high-quality benchmarks for evaluating music retrieval performance. To address this issue, we introduce \textbf{IncompeBench}, a carefully annotated benchmark comprising $1,574$ permissively licensed, high-quality music snippets, $500$ diverse queries, and over $125,000$ individual relevance judgements. These annotations were created through the use of a multi-stage pipeline, resulting in high agreement between human annotators and the generated data. The resulting datasets are publicly available at https://huggingface.co/datasets/mixedbread-ai/incompebench-strict and https://huggingface.co/datasets/mixedbread-ai/incompebench-lenient with the prompts available at https://github.com/mixedbread-ai/incompebench-programs.

Haruki Fujimaki, Makoto P. Kato

This study addresses the challenge of improving dense retrieval performance for queries containing numerical conditions, such as ``companies with more than one billion dollars in R&D expenditure.'' Although recent research has shown that standard models struggle with numeric information in domains such as finance, e-commerce, and medicine, existing solutions typically decompose queries into textual and numerical components and score them separately. These approaches modify late-interaction retrieval models such as ColBERT and introduce challenges in deployment, latency, and maintainability. To overcome these limitations, we propose NumColBERT, an inference-time non-intrusive method that enhances numerically conditioned retrieval while preserving the original late-interaction mechanism. Because NumColBERT retains the standard ColBERT indexing and MaxSim scoring pipeline, existing optimizations and ecosystem components can be reused directly, facilitating practical deployment. NumColBERT introduces a Numerical Gating Mechanism and a Numerical Contrastive Learning objective to enable numerical conditions to contribute more effectively within standard ColBERT scoring. The gating mechanism amplifies tokens carrying critical numerical constraints while suppressing context-neutral numerical mentions, and the contrastive objective shapes the embedding space to reflect numerical magnitudes, units, and conditions. Experimental results show that NumColBERT substantially outperforms standard fine-tuning baselines and achieves accuracy comparable to or better than prior approaches relying on separate textual and numerical scoring. These findings demonstrate the feasibility of numerically conditioned retrieval with a non-intrusive inference pipeline and present a maintainable solution for real-world deployment.

Koji Nishikawa, Makoto P. Kato

Scientific reading is an active process that frequently requires consulting external resources, but manual keyword searching interrupts the reading flow and imposes a high cognitive load. Existing proactive information retrieval systems often suffer from context ambiguity, as they rely solely on on-screen text and ignore the reader's specific background and intent. In this demonstration, we present H-MAPS (Hierarchical Memory-Augmented Proactive Search Assistant), a proactive literature exploration assistant that resolves this ambiguity by leveraging a three-layered hierarchical memory. Triggered by implicit reading behaviors, H-MAPS articulates the user's latent information needs into explicit natural language questions and performs neural retrieval entirely on the local device to ensure privacy. We demonstrate H-MAPS using a scenario where two researchers, specializing in NLP and HCI, read the same paper. In response, the system generates profile-specific questions and retrieves distinct literature tailored to each user.

Benjamin Clavié, Sean Lee, Rikiya Takehi et al.

Multi-vector dense retrieval methods like ColBERT systematically use a single-layer linear projection to reduce the dimensionality of individual vectors. In this study, we explore the implications of the MaxSim operator on the gradient flows of the training of multi-vector models and show that such a simple linear projection has inherent, if non-critical, limitations in this setting. We then discuss the theoretical improvements that could result from replacing this single-layer projection with well-studied alternative feedforward linear networks (FFN), such as deeper, non-linear FFN blocks, GLU blocks, and skip-connections, could alleviate these limitations. Through the design and systematic evaluation of alternate projection blocks, we show that better-designed final projections positively impact the downstream performance of ColBERT models. We highlight that many projection variants outperform the original linear projections, with the best-performing variants increasing average performance on a range of retrieval benchmarks across domains by over 2 NDCG@10 points. We then conduct further exploration on the individual parameters of these projections block in order to understand what drives this empirical performance, highlighting the particular importance of upscaled intermediate projections and residual connections. As part of these ablation studies, we show that numerous suboptimal projection variants still outperform the traditional single-layer projection across multiple benchmarks, confirming our hypothesis. Finally, we observe that this effect is consistent across random seeds, further confirming that replacing the linear layer of ColBERT models is a robust, drop-in upgrade.

Kojiro Iizuka, Hajime Morita, Makoto P. Kato

This study presents a theoretical analysis on the efficiency of interleaving, an efficient online evaluation method for rankings. Although interleaving has already been applied to production systems, the source of its high efficiency has not been clarified in the literature. Therefore, this study presents a theoretical analysis on the efficiency of interleaving methods. We begin by designing a simple interleaving method similar to ordinary interleaving methods. Then, we explore a condition under which the interleaving method is more efficient than A/B testing and find that this is the case when users leave the ranking depending on the item's relevance, a typical assumption made in click models. Finally, we perform experiments based on numerical analysis and user simulation, demonstrating that the theoretical results are consistent with the empirical results.

Junjie H. Xu, Kohei Shinden, Makoto P. Kato

This paper addresses the problem of generating table captions for scholarly documents, which often require additional information outside the table. To this end, we propose a method of retrieving relevant sentences from the paper body, and feeding the table content as well as the retrieved sentences into pre-trained language models (e.g. T5 and GPT-2) for generating table captions. The contributions of this paper are: (1) discussion on the challenges in table captioning for scholarly documents; (2) development of a dataset DocBank-TB, which is publicly available; and (3) comparison of caption generation methods for scholarly documents with different strategies to retrieve relevant sentences from the paper body. Our experimental results showed that T5 is the better generation model for this task, as it outperformed GPT-2 in BLEU and METEOR implying that the generated text are clearer and more precise. Moreover, inputting relevant sentences matching the row header or whole table is effective.

Bei Liu, Jianlong Fu, Makoto P. Kato et al.

Automatic generation of natural language from images has attracted extensive attention. In this paper, we take one step further to investigate generation of poetic language (with multiple lines) to an image for automatic poetry creation. This task involves multiple challenges, including discovering poetic clues from the image (e.g., hope from green), and generating poems to satisfy both relevance to the image and poeticness in language level. To solve the above challenges, we formulate the task of poem generation into two correlated sub-tasks by multi-adversarial training via policy gradient, through which the cross-modal relevance and poetic language style can be ensured. To extract poetic clues from images, we propose to learn a deep coupled visual-poetic embedding, in which the poetic representation from objects, sentiments and scenes in an image can be jointly learned. Two discriminative networks are further introduced to guide the poem generation, including a multi-modal discriminator and a poem-style discriminator. To facilitate the research, we have released two poem datasets by human annotators with two distinct properties: 1) the first human annotated image-to-poem pair dataset (with 8,292 pairs in total), and 2) to-date the largest public English poem corpus dataset (with 92,265 different poems in total). Extensive experiments are conducted with 8K images, among which 1.5K image are randomly picked for evaluation. Both objective and subjective evaluations show the superior performances against the state-of-the-art methods for poem generation from images. Turing test carried out with over 500 human subjects, among which 30 evaluators are poetry experts, demonstrates the effectiveness of our approach.