#Japan #Tokyo #Natural Language Processing #Sherpa Company #NLP2026

Sherpa's Presentation at NLP2026

Sherpa & Company, headquartered in Shinagawa, Tokyo, is set to make waves at the upcoming 32nd Annual Conference of the Society for Natural Language Processing (NLP2026), taking place from March 9 to March 13, 2026. The company will unveil two groundbreaking research papers focused on enhancing natural language processing capabilities.

Background

In recent years, the landscape of corporate information disclosure has evolved significantly. Amid stricter regulations and rising demands from investors and partners, companies are now disclosing a diverse range of information, not just financial data, but also encompassing environmental initiatives, human capital, and governance. This trend has led to an explosive increase in sustainability-related disclosures, much of which include textual data and visuals available in PDF formats.

As organizations work to present this complex information, the challenge of accurately extracting and interpreting critical details from unstructured documents has grown. The need for advanced AI technologies that combine natural language processing with the ability to integrate visual information such as graphs and figures has never been more urgent.

To apply these technologies effectively in real-world business scenarios, establishing an objective framework for comparison and validation based on actual business documents is essential. However, resources for validating Japanese business documents remain limited. The two studies presented by Sherpa aim to address this gap, forming a foundation for evaluating comprehension of Japanese documents from both specialized and general perspectives.

Paper Summaries

Paper 1

• - Date: March 10 (Tuesday) 11:15-12:45 / Session C: "Multimodal / Speech Language Processing" (C2-24)
• - Title: "Omni-JDocVQA: Constructing a Benchmark for Japanese Visual Document Understanding with Diverse Document Types"
• - Authors: Reo Kajikawa, Kota Nakayama, Yusuke Oda, Shunsuke Kanda, Koichi Akabe, Takashi Ninomiya, Naoya Okazaki
• - Summary: This paper introduces a new benchmark, "Omni-JDocVQA," characterized by a wide array of document type labels and realistic questions that do not depend on document content, aimed at enhancing Japanese visual document comprehension.

Paper 2

• - Date: March 10 (Tuesday) 16:55-18:25 / Session Q: "Language Resources, Annotation, and Evaluation" (Q4-10)
• - Title: "ESG-QA: Constructing a Multimodal Question Answering Benchmark for Japanese ESG Documents"
• - Authors: Ryoaki Sata, Koichi Akabe, Shunsuke Kanda, Yusuke Oda
• - Summary: Based on insights from industry experts, this benchmark evaluates the understanding capability of Japanese ESG documents, including unconventional layouts, through three tasks: "page search," "answer generation," and "region detection."

NLP2026 Conference Details

• - Date: March 9 (Monday) to March 13 (Friday), 2026
• - Venue: Light Cube Utsunomiya, 1-20 Miyamirai, Utsunomiya City, Tochigi Prefecture
• - Organizer: The Association for Natural Language Processing
• - Website: NLP2026

(Pre-registration required via the website to participate.)

Sherpa operates under the vision of achieving a world where profit and sustainability intertwine. The company is committed to unlocking the potential of non-financial information and enhancing research and development in AI technologies.

Company Overview

• - Name: Sherpa & Company, Inc.
• - CEO: Jun Sugimoto
• - Location: 1F Terrace Site Gotanda, 3-6-32 Nishi-Gotanda, Shinagawa, Tokyo
• - Established: September 2019
• - Capital: 100 million JPY
• - Business: Development and provision of a cloud-based platform for sustainability information disclosure, operation of "ESG Journal Japan," a media outlet focused on ESG and sustainability, and consulting services through LEV (ESG Advisory) by sustainability experts.

• - Corporate Website: Sherpa Company
• - Recruitment Site: Careers
• - Research and Development: R&D