@article{4778,
  author = {Pit Pichappan},
  title = {Vocabulary Distribution, Lexical Diversity, and Statistical Properties of the Fact Verification Corpus},
  journal = {International Journal of Computational Linguistics Research},
  year = {2026},
  volume = {17},
  number = {3},
  doi = {https://doi.org/10.6025/ijclr/2026/17/3/172-185},
  url = {https://www.dline.info/jcl/fulltext/v17n3/jclv17n3_3.pdf},
  abstract = {The proliferation of digital misinformation necessitates robust automated fact verification systems, which
rely heavily on high quality benchmark datasets. However, the underlying linguistic and statistical properties
of these corpora often remain underexplored. This study presents a comprehensive corpus-linguistic
analysis of a fact verification dataset, examining its vocabulary distribution, lexical diversity, and adherence
to natural language statistical regularities. Employing a multi metric framework, we evaluated lexical
richness using indices such as the Shannon and Simpson Diversity Indices, Type Token Ratio, MTLD, and HDD.
Furthermore, we investigated corpus level statistical properties through Zipf's and Heap's laws. Results
indicate substantial lexical heterogeneity, with high diversity indices confirming a rich vocabulary inventory
suitable for complex semantic reasoning. The rank frequency distribution conforms to Zipf's Law ( =
0.482), revealing a characteristic long tail pattern where domain specific terminology and functional words
dominate, alongside numerous infrequent but semantically critical entities. Additionally, Heap's Law analysis
( = 0.637) demonstrates sublinear vocabulary growth, indicating lexical saturation typical of specialized
technical corpora. These findings validate the dataset's linguistic coherence and statistical suitability
for downstream natural language processing applications, including transformer based language modeling,
information retrieval, and retrieval augmented generation. Ultimately, this research establishes a strong
empirical foundation for optimizing data preprocessing, model architecture selection, and the development
of reliable, evidence aware fact checking systems},
}