The NP de VP construction in Modern Chinese, a grammatical pattern where a noun phrase (NP) is followed by the possessive particle *de* and a verb phrase (VP), has been a subject of ongoing linguistic inquiry. Previous research has focused on the word classes within the VP, the semantic relationship between the NP and the VP, and the lexical items that can occupy either slot. However, this study distinguishes itself by focusing on the typical meanings of the construction as a whole and analyzing the meaning patterns of lexical items in both the NP and VP slots. The research utilizes a construction grammar framework, defining construction as a form-meaning pair where aspects of the form or meaning are not predictable from the components (Goldberg, 1995). The study employs large-scale corpus analysis to identify statistically significant cooccurrences and meaning patterns objectively, addressing a gap in previous research which often relied on smaller datasets and subjective interpretations. The research questions are: 1) What are the typical meanings of the NP de VP construction? 2) What are the meaning patterns of the VP in the NP de VP construction? 3) What are the meaning patterns of the NP in the NP de VP construction?

Prior studies on the NP de VP construction have explored various aspects, including the word class of the VP (nominalized or verbal), the head of the NP de VP construction, and the semantic relationship between the NP and VP (agent or patient). There are differing views on whether the VP is nominalized or remains verbal, with arguments supporting both perspectives. Regarding the semantic relationship, studies have shown the NP can function as both agent and patient, although the patient role is more frequent. Previous research also investigated the meaning patterns of the VP and NP slots but lacked a statistically robust corpus-based approach to determine the typical and most frequent meanings. Zhan (1998) and Wang (2002) suggested that VPs with weak "action" and strong "event" are more likely to appear in this construction. Shen and Wang (2000) posited that NPs must possess a sense of prominence (informativity and accessibility) to fill the NP slot. However, these studies lacked the quantitative rigor and large corpus data to establish objectively the typical meanings and meaning patterns.

This research employed the Beijing Language and Culture University Corpus (BCC), a large corpus of approximately 9.5 billion tokens covering diverse genres. Data were collected using the search query "../n的../v", focusing on 2-character nouns (NP) and 2-character verbs (VP) to improve the precision of the analysis and focus on higher-frequency, more typical instances. Two primary analytical approaches were used: 1. **Covarying Collexeme Analysis:** This statistical method identifies the association strength between lexical items in the NP and VP slots. It measures how often specific NP-VP pairs co-occur in the construction compared to what would be expected by chance. The Fisher-Yates exact test was used to assess statistical significance, with association strengths above 1.301 (p < 0.05), 2 (p < 0.01), and 3 (p < 0.001) indicating significance. Infinite values indicated extremely high association strengths. This step helped identify significantly attracted lexical items. 2. **Hierarchical Cluster Analysis:** This method grouped the significantly attracted lexical items (NPs and VPs) based on their semantic similarity. The clustering relied on covarying collexemes identified through the first step, meaning the clusters of VPs were informed by their frequent co-occurrence with specific NPs, and vice versa. Ward's method was used for hierarchical clustering, producing dendrograms that visually represented the clusters and their relationships. This analysis provided an objective and nuanced understanding of the semantic patterns of the construction. The R programming language was utilized for both statistical analyses. The selection of these methods aims to identify typical meanings objectively, avoiding subjective interpretations based solely on researcher intuition or smaller datasets.

The covarying collexeme analysis identified 515 significantly attracted instances of the NP de VP construction, with 71 showing infinite association strength, indicating extremely strong co-occurrence. Analysis of these high-association instances revealed three typical meaning pairings: 1. **Regulations - Implementation:** NPs denoting regulations (e.g., *zhengce* 'policy', *zhidu* 'regulation') frequently co-occurred with VPs denoting implementation (e.g., *zhiding* 'enact', *guanche* 'implement'). 2. **Systems - Establishment:** NPs representing systems (e.g., *tizhi* 'regulation', *tixi* 'system') often paired with VPs indicating establishment (e.g., *jianli* 'establish', *sheli* 'set up'). 3. **Results - Achievement:** NPs expressing results (e.g., *mubiao* 'target', *chengji* 'results') frequently appeared with VPs conveying achievement (e.g., *qude* 'achieve', *shixian* 'realize'). Hierarchical cluster analysis of the significantly attracted VPs identified six main meaning patterns: "cognition", "augmentation", "implementation", "achievement", "establishment", and "report". The "augmentation" cluster, for example, included verbs like *kuoda* 'expand', *zengjia* 'augment', and *zengqiang* 'enhance', reflecting a shared semantic feature of increasing degree or quantity. Similarly, the NP cluster analysis, considering their co-occurring VPs, revealed meaning patterns of "internal traits", "medical names", "regulations", "results", "systems", and "business". These findings demonstrated a complex interplay between the semantic roles of the NPs and VPs in shaping the overall meaning of the NP de VP construction. The dendrograms (Figures 1 and 2) visually illustrate these clusters and their hierarchical relationships.

The findings of this study partially align with, but also extend, previous research. While Zhan (1998) suggested that only VPs with weak "action" and strong "event" were suitable, this research identified VPs across different action-event combinations, with a higher proportion of weak action/strong event verbs. This discrepancy likely stems from differences in data sources (corpus data versus invented examples) and the focus on statistically significant instances in this study. The study's findings also support and expand upon Shen and Wang's (2000) work on the prominence of NPs, but it provides a more objective and detailed understanding of the semantic patterns through cluster analysis. The wide range of meaning patterns observed (both for NPs and VPs) demonstrates the principle of linguistic meaning conservation, where the variety of lexical items reflects a broader range of information conveyed through the construction. The high degree of compositionality, prototypicality, and conventionality in the construction was also explored through the lens of construction grammar, suggesting that the meaning of the construction is largely transparent and predictable based on the combined meaning of its constituent parts. The identified meaning patterns can also inform the design of pedagogical materials for both native and non-native learners of Chinese.

This research provides a detailed and nuanced analysis of the NP de VP construction in Modern Chinese using covarying collexeme analysis and hierarchical cluster analysis. The study scientifically identifies typical instances of the construction and objectively reveals the semantic patterns of lexical items in both the NP and VP slots. The findings have implications for linguistic theory and language pedagogy, demonstrating the intricate relationship between form and meaning in this prevalent construction. Future research could explore variations in meaning patterns across different genres and text types, further refining our understanding of this significant construction.

The study's findings are based on a large corpus but do not account for genre-specific differences. While the focus on two-character NPs and VPs improved analysis precision, some valid instances might have been excluded. The hierarchical clustering process, while objective, still relies on underlying semantic assumptions. Further investigation into the impact of modifying elements and contextual factors on meaning patterns could enhance the comprehensiveness of future studies.