The Samskriti Gap: Why Deep Learning Models Struggle with Vedic Sandhi Rules

Standard NLP architectures consistently misparse the compound vowel transformations in Ṛgveda Hymn 10.129 because they treat sandhi as a phonetic error rather than a syntactic marker. A new benchmark dataset reveals that models trained on machine-translated texts achieve less than 60% accuracy on basic prākṛta inflections.

Full article content coming soon.