1 Origin

Reviewed the author's article on language planning published 24 years ago, and introduced the concept and development of language planning. The article points out that the definition of language planning is constantly changing with the development of society and the deepening of research, but its basic characteristic is restrictivism. Language planning has the characteristics of sociality, authority, communicative purpose, long-term, practicality, and data-driven. The author also mentioned two main aspects of language planning: status planning and ontology planning. In addition, the author emphasizes the importance of language and writing resource planning and proposes the necessity of planning language and writing resources in the era of big data. The article illustrates the important role of text data mining in new knowledge discovery through examples, and calls for attention and research on language and text resource planning.

The impact of the scale of language resources on machine translation

With the multilingualization of the Internet and the arrival of the era of big data, language data resources have become the key to machine translation. In 2019, the top ten languages used on the Internet include English, Chinese, Spanish, etc. The global data volume is growing at an annual rate of about 40%, and is expected to reach 2142ZB by 2035. The data volume generated by China accounts for about one fifth of the world's total. The number of Internet users has increased from 22.5 million in 2000 to nearly 1.1 billion in 2022, and the Internet penetration rate has significantly increased. The demand for machine translation has increased, but China's translation capabilities are insufficient. Although the translation market is large, domestic translation companies can only digest about 10%. The quality of machine translation is closely related to the scale of language data resources, and neural machine translation particularly relies on large-scale data resources. Mainstream languages have good neural machine translation results due to abundant data resources, but non mainstream languages and professional fields have difficulty ensuring translation quality due to a lack of data resources. Chinese ethnic minority languages and professional translation fields also face the problem of insufficient data resources. Therefore, strengthening the acquisition, utilization, and management of language data resources, conducting research on language and text resource planning, is crucial for improving the quality of machine translation.

The impact of the scale of 3 language resources on large language models

The Large Language Model (LLM) is a language model constructed by deep neural networks, containing billions of parameters. Based on rich language data resources, it is trained using self supervised learning methods. Since 2018, multiple companies and research institutions have released various large language models, such as BERT, GPT, etc., which have performed well in natural language processing (NLP) tasks. The scale of large language models and the size of language data resources have a significant impact on the effectiveness of natural language processing. N-gram provides theoretical support for large language models, and higher-order N-gram can improve system quality. The large language models launched by companies such as Google and OpenAI, such as ChatGPT and GPT-4, demonstrate powerful natural language understanding and generation capabilities.

Four Issues in Language and Text Resource Planning in the Age of Big Data

In the era of big data, language and text resource planning faces four main issues: firstly, it is necessary to strengthen the alignment of big language models to ensure that their outputs and behaviors are in line with human expectations and values, in order to address potential ethical risks and threats. Secondly, the construction of Chinese language resources needs to be strengthened to break down data silos and enhance the performance of Chinese in large language models. Thirdly, the intellectual property protection of language data resources needs to be strengthened to promote the orderly circulation and effective utilization of resources. Finally, the construction of resources for low resource languages should also be given attention to support the development of large language models for these languages. In addition, a series of governance measures and initiatives for large language models have been proposed both domestically and internationally, demonstrating widespread attention to the use and management of language and textual data resources. The introduction of relevant regulations and initiatives, such as the Administrative Provisions on the Deep Integration of Internet Information Services, the Interim Measures for the Management of Generative AI Services, and the Global AI Governance Initiative, has provided guidance and framework for the governance of the big language model. Meanwhile, the governance of big language models has become a global consensus, emphasizing the importance of value alignment while pursuing knowledge and capability enhancement.