Archives for Technical Sciences | 2026
Authors: Tadjieva M.; Matniyazova Z.; Djumayeva Z.; Umarkhujaeva K.; Khoshimkhujaeva M.; Ankabayeva M.; Yusupov O.
DOI: 10.70102/afts.2026.1835.049
Journal: Archives for Technical Sciences
Year: 2026
Publisher: Technical institute of Bijeljina
Document Type: Article
Open Access: All Open Access; Gold Open Access
Cited by: 0
Manual comparative methods have long been the main source for reconstructing Proto-Indo-European (PIE) dialects, with their weaknesses including fragmentary corpora, interpretive bias, and a lack of direct textual evidence. This paper introduces a probabilistic semantic reconstruction model that combines computational comparative linguistics and deep neural archiving to learn and reconstruct the dialectal variations that have been lost in PIE. A multilingual dataset of 12 Indo-European language branches and 18,742 cognate sets, with phonological, morphological, and semantic feature embeddings, was compiled and entered. An inverse phylogenetic inference model based on Bayesian inference and a transformer-based deep neural network trained on 4.6 million aligned lexical tokens was used to predict proto-forms and semantic shifts. When tested against known scholarly reconstructions, the proposed model achieved 86.3% accuracy in phonological reconstruction and 0.81 semantic consistency (cosine similarity metric). Cross-validation indicated a 14.7% decrease in reconstruction variance compared to traditional rule-based methods. Probabilistic confidence intervals (95% CI) also showed consistent predictions for high-frequency lexical roots, with posterior probabilities greater than 0.90 for the reconstructed forms (63%). Moreover, statistically significant divergence patterns (p < 0.01) were observed in the dialectal clustering analysis and were consistent with established Indo-European subgroup stratifications. The results show that probabilistic modelling with deep neural semantic archiving can significantly improve the reliability and interpretability of reconstruction. This framework offers a computational approach to historical linguistics that can be scaled and replicated. Also, it provides a new quantitative understanding of the evolution of proto-languages and dialect differentiation within the Indo-European family. © 2026, Technical institute of Bijeljina. All rights reserved.
computational comparative linguistics; deep neural language models; dialectal evolution modelling; historical linguistics digitisation; probabilistic phylogenetic modelling; proto-indo-european reconstruction; semantic embedding analysis