1. Language and Robotics: Toward Building Robots Coexisting with Human Society Using Language Interface

Yutaka Nakamura1, Shuhei Kurita2, and Koichiro Yoshino1

1Guardian Robot Project (GRP), Information R&D and Strategy Headquarters, RIKEN

2Center for Advanced Intelligence Project (AIP), RIKEN

Robots are one of the archetypes of AI systems we imagine, and the realization of such robots operating in the real world with language interfaces has long been a dream of us.

This introductory tutorial aims to help researchers who will start language and robotics, (LangRobo) research in the future by summarizing three points: awareness of the community’s issues, recent approaches for these issues, and remaining problems.

We arrange this tutorial involving not only NLP researchers but also robotics researchers in order to raise issues that are relevant to actual robotics problems.

There are several difficulties in connecting NLP and robotics, but the following three are particularly problematic:

– The great difference in granularity between language and robot behavior.

– Robotics tasks involving real-world control often do not allow for language ambiguity.

– Language expressions themselves are often ambiguous and require background knowledge or commonsense reasoning to understand them correctly.

Many recent works have suggested that deep learning or LLMs can provide solutions.

This tutorial summarizes the recent approaches to the language and robotics problem using such learning-based approaches.

The goal of this tutorial is to share the discussion on how these problems can be solved in the future.

2. Current Status of NLP in South East Asia with Insights from Multilingualism and Language Diversity

Alham Fikri Aji1, Jessica Zosa Forde2, Alyssa Marie Loo2, Lintang Sutawika3, Skyler Wang4,5, Genta Indra Winata6, Zheng-Xin Yong2, Ruochen Zhang2, A. Seza Doğruöz7, Yin Lin Tan8,9, and Jan Christian Blaise Cruz10

1MBZUAI, 2Brown University, 3EleutherAI , 4UC Berkeley, 5Meta AI , 6Bloomberg , 7Universiteit Gent, 8Stanford University, 9National University of Singapore , 10Samsung R&D Institute Philippines

South East Asia (SEA) is a region with immense cultural and linguistic diversity—a melting pot of cultures, religions, and languages, home to over 1000 languages. In addition, multilingualism (i.e., speaking more than one language or dialect) is widely practiced on a daily basis. Despite the variety of languages, there is relatively less research on natural language processing (NLP) of SEA languages and their users in the area compared to languages in other regions. The scarcity of available datasets for the region’s languages presents a challenge for developing NLP technology for SEA languages. Other challenges include the complexity of language use in the region, such as code-switching, and the inaccessibility of language technology to certain groups of SEA researchers due to constraints on computing resources. This tutorial will present an overview of language issues in the SEA region, link multilingualism and computational sociolinguistics with historical and societal perspectives, and provide a summary of existing datasets for computational linguistics research, language models and NLP systems, and evaluation benchmarks. The tutorial will also characterize the existing research ecosystem in SEA, including community-based initiatives working on SEA languages and opportunities for developing NLP technologies for SEA languages.

3. Practical Tools from Domain Adaptation for Designing Inclusive, Equitable, and Robust Generative AI

Anthony Sicilia and Malihe Alikhani

Khoury College of Computer Science, Northeastern University, Boston, MA

Generative language technologies have become integral to everyday communication, shaping social interactions and informing critical decision-making processes in areas such as recruitment, healthcare, and education. However, they often struggle to grasp the “long tail” of data distributions — concepts less frequently observed during training — which could have significant repercussions. These models may marginalize underrepresented groups by failing to comprehend preferred communication styles, such as code-switching, or perpetuating societal biases like gender bias. Sectors like healthcare, education, and law, requiring personalization and exhibiting nuanced linguistic features, are also particularly affected when pre-trained models misconstrue or overlook “long tail” data concepts. While methods like distillation of smaller language models, active learning, and other bias mitigation strategies can augment traditional training techniques, a careful statistical analysis is essential for their effective application. This tutorial offers a comprehensive examination of how to develop equitable, robust, and inclusive language technologies using statistical tools from Domain Adaptation (DA) that catalyze positive social change. We will delve into strategies for bias mitigation, explore how to measure bias, and examine open problems in creating culturally-grounded and inclusive language technologies. Accompanying materials including code notebooks, python packages, and coursework will be provided.

4. Editing Large Language Models

Ningyu Zhang1, Yunzhi Yao1, and Shumin Deng2

1Zhejiang University, China, 2National University of Singapore, Singapore

Even with their impressive abilities, Large Language Models (LLMs) such as ChatGPT are not immune to issues of factual or logically consistent. Concretely, the key concern is how to seamlessly update those LLMs to correct mistakes without resorting to an exhaustive retraining or continuous training procedure, both of which can demand significant computational resources and time. Thus, the capability to edit LLMs offers an efficient solution to alter a model’s behavior, notably within a distinct area of interest, without negatively impacting its performance on other tasks. Through this tutorial, we strive to acquaint interested NLP researchers with recent and emerging techniques for editing LLMs. Specifically, we aim to present a systematic and current overview of cutting-edge methods, supplemented with practical tools, and unveil new research opportunities for our audience. All resources can be found at https://github.com/zjunlp/ModelEditingPapers.

5. Learning WHO Saying WHAT to WHOM in Multi-Party Conversations

Jia-Chen Gu1, Zhuosheng Zhang2, and Zhen-Hua Ling1

1University of Science and Technology of China

2Shanghai Jiao Tong University

Multi-party conversations (MPC) are a more practical and challenging scenario involving more than two interlocutors. This research topic has drawn significant attention from both academia and industry, and it is nowadays counted as one of the most promising research areas in the field of dialogue systems. In general, MPC algorithms aim at addressing the issues of Who saying What to Whom, specifically, who speaks, say what, and address whom. The complicated interactions between interlocutors, between utterances, and between interlocutors and utterances develop many variant tasks of MPC worth investigation. In this tutorial, we present a comprehensive survey of recent advances in MPC. In particular, we summarize recent advances on the research of MPC modeling which is categorized by Who saying What to Whom. Finally, we highlight the challenges which are not yet well addressed in MPC and present future research directions.

6. Developing State-Of-The-Art Massively Multilingual Machine Translation Systems for Related Languages

Tutors and Affiliations: Jay Gala (AI4Bharat, India), Pranjal A. Chitale (IIT Madras, India), and Raj Dabre (NICT, Japan)

The race for developing state-of-the-art (SOTA) machine translation (MT) systems often gives the impression that this can only be done by large organizations, most of which do not fully open-source their systems. In this tutorial, we dispel this myth by generalizing our experiences in developing SOTA MT systems for related languages. We cover topics ranging from (a) the history of MT systems for related languages, (b) curating high-quality datasets, manually created as well as mined, (c) creating domain-diverse benchmarks, (d) compact but high-quality open-source MT systems that surpass other systems despite being an order of magnitude smaller in terms of parameters and computational costs than massively multilingual generic systems and (e) robust automatic and human evaluation. We hope that our tutorial encourages other groups, regardless of scale, to engage in focussed efforts on related languages or language groups to develop open-source, high-quality MT systems.