1 Introduction

This section introduces the universality of German compound word formation and its low frequency in corpora, points out the limitations of traditional frequency-based vocabulary difficulty assessment methods in handling German compound words, and emphasizes the importance of developing automated methods for identifying and splitting German compounds. It also explains that the development of word vector technology provides new tools for compound word identification and splitting, and proposes a strategy for compound word identification and splitting based on word vector technology, aiming to optimize vocabulary difficulty assessment models and accumulate experience in computational processing of complex lexical structures.

2 Theoretical Basis and Research Status

First, the theoretical basis and research status of word vectors are introduced, noting that word vectors are mathematical representations extracted from large-scale corpora based on the distributional hypothesis, capable of reflecting the semantic distribution of words. They transform word meanings into vector forms through co-occurrence statistical patterns, enabling quantitative modeling of semantic and syntactic features and allowing algebraic operations to analyze semantic relationships. The development of word vector technology, from Mikolov's word2vec model to Devlin's BERT model, has progressed from static to dynamic, enhancing sensitivity to context and providing new methods for natural language processing. However, its application in morphology remains limited, particularly in the identification and splitting of German compound words, where research is scarce. Subsequently, the research status of German compound words is elaborated, explaining that languages generate new words by combining a finite set of morphemes to control vocabulary size. Compound words serve as tools for information compression. German compounds consist of a head constituent and a non-head constituent, with the head determining the compound's word class, inflection type, and contextual environment. The non-head constituent is located at the left periphery, and there may be linking elements between them. Although these linking elements lack independent semantics, they can affect automatic segmentation and morpheme boundary identification. Constructing a German compound word identification system requires consideration of their morphological behavior and distribution patterns.

3 Research Design

This chapter details the research design for the automated identification and splitting of German compound words based on word vector technology. It begins by introducing the research platform and configuration, selecting the open-source model de_core_news_lg from the spaCy platform. This model is trained on multiple German corpora, possesses 500k word vectors, offers broad coverage and good annotation accuracy, though it has the issue of assigning vector representations to a small number of meaningless letter combinations. The experimental environment includes the Windows 11 operating system, Python 3.8.0 programming language, and corresponding hardware configurations. It then discusses the calculation and comparative analysis of word vector similarities between compounds and their non-head and head constituents. Based on traditional linguistic perspectives, it hypothesizes that the word vector similarity between a compound and its head constituent is higher than that with its non-head constituent, and this hypothesis is validated through calculations with ten example words, providing quantitative evidence for compound structure analysis. Additionally, background noise is calculated, with values determined via the Monte Carlo algorithm to eliminate its interference in experimental data. Finally, the identification threshold is established; after multiple tests, the threshold is set at 0.2 to ensure consistency, repeatability, and objectivity in the identification process, and to quantify the proportion of semantic contribution of the head constituent to the compound word.

4 Research Results

This chapter elaborates on the methods and practical results of automated identification and splitting of German compound words based on word vector technology. First, using German compound nouns as examples, it introduces the methods for identifying and splitting compound words, including steps such as word vector existence checks, extraction and similarity calculation of candidate head constituents, optimal suffix screening and misjudgment prevention, and constraints based on morphological rules. Then, by constructing a lexicon using vocabulary lists from the first and second volumes of "New Classic German" for experiments, the results are analyzed, concluding that the method exhibits high accuracy and reliability in the automated identification and splitting of compound nouns, while also noting the limitations of the experimental dataset. Finally, the reasons for incorrect identifications and splits are reflected upon, including erroneous training, semantic weight biases of polysemous words, and conflicts in identifying derivational prefixes and linking elements, with corresponding suggestions for improvement proposed.

5 Conclusion

This section summarizes the research on automated identification and splitting of German compound words based on word vector technology. The proposed method effectively addresses the limitations of traditional approaches and fills an academic gap. Through semantic similarity, similarity thresholds, and constraints of morphological rules, high-accuracy identification and splitting are achieved. The algorithm demonstrates high computational efficiency, applicable to long text segment analysis, and can be adapted to the splitting and identification of compound words of other parts of speech. The research findings can be widely applied in teaching practices, such as vocabulary difficulty labeling and word family statistical analysis in digital textbook development, and also offer insights for studies on German derivational words and separable verbs.