Scholarlink

In the globalization era, preserving and digitizing cultural heritage is crucial. Cross-cultural language text-to-text generation systems face challenges like script differences, phonetic nuances, and domain-specific terminology. This paper presents SanskritiGenAI, a dynamic framework for cross-cultural text generation, aiding in preserving diverse linguistic and cultural artifacts. SanskritiGenAI uses a novel transformer architecture with a model dimension of 256, processing batches of 64 sequences. The feed-forward network (FFN) in each transformer layer has a hidden size of 1024, and 3 encoding and decoding layers with unique performer attention layers which have 8 attention heads, reducing computational complexity from quadratic to linear with respect to sequence length. The model is trained on 30 lac words for each of 21 languages and tested on historical manuscripts, literary works, and folk narratives. The results show the model's effectiveness in transliterating and transcribing texts with minimal loss, achieving the best accuracy of 98.5% compared to AIforBharat, SMT, and OpenNMT. SanskritiGenAI promises applications in digital libraries, language learning platforms, and cultural preservation initiatives, offering a valuable tool for bridging linguistic gaps and ensuring the longevity of cultural expressions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Cross-Cultural Multilingual Text-to-Text Generation Framework Using Novel SanskritiGenAI Model for Preserving and Digitizing Cultural Heritage

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jamuna S. Murthy,
  • K. Dhanashekar,
  • G. M. Siddesh

摘要

In the globalization era, preserving and digitizing cultural heritage is crucial. Cross-cultural language text-to-text generation systems face challenges like script differences, phonetic nuances, and domain-specific terminology. This paper presents SanskritiGenAI, a dynamic framework for cross-cultural text generation, aiding in preserving diverse linguistic and cultural artifacts. SanskritiGenAI uses a novel transformer architecture with a model dimension of 256, processing batches of 64 sequences. The feed-forward network (FFN) in each transformer layer has a hidden size of 1024, and 3 encoding and decoding layers with unique performer attention layers which have 8 attention heads, reducing computational complexity from quadratic to linear with respect to sequence length. The model is trained on 30 lac words for each of 21 languages and tested on historical manuscripts, literary works, and folk narratives. The results show the model's effectiveness in transliterating and transcribing texts with minimal loss, achieving the best accuracy of 98.5% compared to AIforBharat, SMT, and OpenNMT. SanskritiGenAI promises applications in digital libraries, language learning platforms, and cultural preservation initiatives, offering a valuable tool for bridging linguistic gaps and ensuring the longevity of cultural expressions.