TL;DR
This paper introduces a novel approach to compare simplified and traditional Chinese by analyzing word co-occurrence networks using complex network theory. The findings reveal that both forms exhibit small-world and scale-free network properties, with traditional Chinese networks being denser. Crucially, motif analysis indicates semantic consistency between the two systems, despite differences in vocabulary and structure.
What Problem Does This Paper Address?
Chinese is written in two primary forms: simplified (used in mainland China and Singapore) and traditional (used in Hong Kong, Macao, and Taiwan). While these forms share a common origin, they have diverged over the past half-century due to independent development, leading to differences in character sets, orthography, vocabulary, and semantics. These differences can create communication barriers. Existing research primarily focuses on character or lexical-level comparisons, failing to capture the broader, global linguistic differences. This paper aims to bridge this gap by employing complex network analysis to provide a holistic understanding of the distinctions and similarities between simplified and traditional Chinese.
Methods
The core methodology involves constructing and analyzing word co-occurrence networks.
- Word Co-occurrence Network Construction: Words from selected news corpora are treated as nodes. An edge connects two words if they appear within a certain proximity (distance not greater than 2). This process is repeated for both simplified and traditional Chinese texts, creating distinct networks.
- Complex Network Analysis: Several standard complex network metrics are applied:
- Network Statistics: Average path length and clustering coefficient are calculated to characterize the overall network structure.
- Small-World and Scale-Free Properties: The networks are assessed to determine if they exhibit small-world (short average path length, high clustering coefficient) and scale-free (power-law degree distribution) characteristics.
- Kernel Lexicon Extraction: Words are ranked by their degree (frequency of co-occurrence), and a "kernel lexicon" is formed by selecting words above a certain degree threshold (identified by a crossover point in the cumulative degree distribution). The size and composition of these lexicons are compared.
- Motif Analysis: Subgraphs (motifs) of a specific size (4 nodes in this case) are counted and compared across the simplified and traditional Chinese networks. This method is used to infer semantic consistency.
Datasets
The study utilizes texts from the Chinese GigaWord Third Edition (LDC2007T38).
- Simplified Chinese texts are sourced from the "Xinhua News Agency" (XIN).
- Traditional Chinese texts are sourced from the "Central News Agency" (CNA).
The experiments were conducted using three different text scales, increasing from 7 million words to 10 million and then 15 million words for both XIN and CNA corpora.
Key Insights and Ideas
- Global Linguistic Structure: The paper moves beyond character-level analysis to explore the underlying structural differences in language use through word co-occurrence.
- Network Properties as Linguistic Signatures: The application of complex network theory reveals that language structure can be characterized by network metrics, offering a quantitative way to compare linguistic systems.
- Semantic Consistency Despite Structural Differences: The finding that motif analysis shows no significant difference between the networks is particularly important, suggesting that despite structural variations, the core semantic relationships between words remain consistent.
Key Results
- Small-World and Scale-Free Networks: Both simplified and traditional Chinese word co-occurrence networks exhibit small-world and scale-free properties, consistent with findings in other languages.
- Denser Traditional Chinese Networks: For comparable corpus sizes, traditional Chinese networks tend to have more nodes and edges, leading to higher average degrees. This is attributed to the one-to-many mapping of simplified characters to traditional characters (e.g., one simplified character representing multiple traditional characters or words).
- Larger Kernel Lexicons in Traditional Chinese: Traditional Chinese kernel lexicons are generally larger than their simplified counterparts. This is explained by the retention of more ancient Chinese words and fewer "weak verbs" in traditional Chinese, leading to a richer vocabulary base.
- Semantic Consistency: Motif analysis indicates that the semantic consistency between simplified and traditional Chinese is high, suggesting that the core meaning and relationships of words are preserved across both systems.
Limitations and Future Work
- Corpus Bias: The study relies on news corpora. While extensive, these may not fully represent the entirety of written Chinese. Future work could explore other genres (e.g., literature, academic texts) to provide a more comprehensive comparison.
- One-to-Many Mapping Explanation: While the paper suggests the one-to-many mapping as a reason for denser traditional Chinese networks, further investigation into the specific linguistic mechanisms driving this could be beneficial.
- Motif Analysis Depth: While motif analysis suggests semantic consistency, exploring more complex motifs or different types of semantic networks could offer deeper insights into nuanced semantic differences.
- Evolutionary Dynamics: The paper focuses on a static comparison. Future research could investigate the evolutionary dynamics of these networks over time to understand how simplified and traditional Chinese continue to diverge or converge.
Summary updated