1 Introduction

Corpus plays a crucial role in language research, and its scale and retrieval tools have undergone significant development. From the initial 1 million word count to the current tens of billions of word count, search tools have evolved from specialized software such as Xaira to general-purpose software such as AntConc, WordSmith Tools, LancsBox, and then to web-based fourth generation platforms such as CQPweb and Sketch Engine. The functions of these tools continue to enrich, including indexed rows, word lists, multi word sequence lists, keyword lists, word frequency distribution maps, word clouds, and vocabulary networks, but the core idea is always based on frequency. The definition of form symbols can be flexible, including combinations of words and part of speech codes to more accurately count word frequency. Corpus tools provide various statistical lists, such as word lists, word element lists, multi word sequence lists, collocation word lists, and topic word lists, which reflect the degree of prominence of language units from different perspectives. Although corpus research has been in-depth in terms of morphology and syntax, there is relatively little research on semantics, indicating that there is still room for further research.

Semantic research related to corpus 2

Deeply explored the progress of semantic research in natural language processing and corpus linguistics, with a particular focus on the challenges and algorithm research of word sense disambiguation. Research has shown that accurate recognition of polysemous words is crucial for natural language understanding, and existing semantic annotation corpora such as SemCor and SenseDefs provide training and validation data for word sense disambiguation algorithms. Meanwhile, the chapter also discusses the importance of semantic annotation and how it enhances the performance of tasks such as part of speech tagging, dependency grammar tagging, and natural language inference. In addition, the chapter also points out the shortcomings of existing semantic annotation tools and dictionaries in terms of semantic usage information, such as the problem with the distribution map of semantic usage in dictionaries. Finally, the chapter introduces the concept of a dictionary based semantic table, explores the potential of large language models in generating semantic tables, and proposes three specific research questions aimed at addressing the limitations of existing methods and improving the accuracy and practicality of polysemous word recognition.

The form of the three item table

The construction form of the semantic table was discussed, and it was pointed out that in corpus retrieval tools, word lists are treated as strings, and word frequencies of the same string are merged and counted. When generating a part of speech code table, add part of speech information after each word to distinguish the frequency of different parts of speech. Wmatrix distinguishes the usage of different semantic fields by adding semantic classification labels. The author proposes that the meaning table should be presented in the form of "token_1emma-POS2senseNo.", which includes the glyph, prefix, part of speech code, and meaning item number. This form helps to merge and statistically analyze inflectional changes in word forms. The author did not use a complex set of word class labels, but instead used simplified word class labels consistent with the dictionary to avoid affecting the frequency statistics of semantic terms. The semantic table based on the Oxford Advanced Learner's Dictionary (10th edition) should include corresponding part of speech labels. The corpus annotated with semantic terms can be used to calculate the frequency of usage of each word's semantic terms in different parts of speech, forming a semantic table. If you focus on the semantic level, you can remove the glyph and only retain the form of 'lemma-POS2enseNo.'. The sorting of the semantic table can be adjusted according to the needs to meet different research requirements. The author also mentioned that the excellent performance of large language models in semantic understanding makes them potential for semantic annotation to handle large-scale corpora.

Semantic annotation assisted by four major language models

The ChatGPT 4.0 language model performs well in semantic annotation tasks, outperforming version 3.5. Research on the use of the OALD10 dictionary's semantic library, testing through the Brown corpus, and comparing manual judgment with model judgment. When selecting polysemous words, assign part of speech codes to the Brown corpus and filter out words with a meaning count greater than 1. Manual judgment adopts a two person mode to ensure objectivity. Model judgment relies on high-quality prompt words, and ChatGPT returns results in Python dictionary format. The comparison of results shows that ChatGPT has a higher consistency and acceptability rate, especially for actual words, but a lower consistency rate for conjunctions and prepositions. The low accuracy of article judgment may be related to insufficient representation in the model. Overall, ChatGPT has practical value and high accuracy in dictionary based semantic judgment.

Application of the 5-point table

The semantic table has important value in semantic research, as it can reveal the temporal changes and spatial differences in vocabulary meanings. It has applications in areas such as dictionary compilation, language teaching, and second language acquisition. By comparing the semantic tables of different corpora, a thematic semantic table can be generated to reflect the salience of semantic items, which is used for vocabulary research and discourse analysis. Taking the noun 'screen' as an example, by analyzing the distribution of meanings in corpora from different periods, it is found that its semantic changes are related to the increase in electronic device popularity and contextual changes.

6 Conclusion

We studied the generation of semantic tables based on dictionaries and improved the accuracy and generalization ability of semantic judgment with the assistance of large language models. The semantic table has practical value in multiple fields, especially in the study of semantic changes. Despite limitations in testing scale and models, with technological advancements, the application prospects of semantic tables are vast.