Large language models (LLMs) process text through subword tokenizers rather than directly reading characters or words. Because these tokenizers are trained predominantly on English-centric corpora, they introduce a systematic and often overlooked disadvantage for many non-English languages. In this work, we quantify this tokenizer tax for Indian languages using the FLORES-200 parallel corpus, measuring tokenization fertility across six widely used tokenizers and fourteen languages. Under cl100k_base (used by GPT-3.5 and GPT-4), Indian languages experience an average 8.0x tokenization tax relative to English, reaching 13.0x for Malayalam, reducing the effective context window to as little as 12% of that available to English users for equivalent semantic content. We identify the primary mechanism behind this disparity: failed byte-pair merges that leave text fragmented into single-byte tokens, with merge failure strongly correlating with tokenizer tax (Pearson r = 0.89). We further show that this phenomenon is not an inherent property of Indic scripts but a consequence of tokenizer design. Multilingual tokenizers such as XLM-R and OpenAI's o200k_base reduce the average Indic tokenizer tax by 73%, demonstrating that the disparity is largely remediable. Beyond token statistics, we quantify a practical consequence by showing that, under fixed context budgets, Indian-language documents preserve substantially less original content than equivalent English documents. Finally, we examine the relationship between tokenizer fertility and reading comprehension performance on the Belebele benchmark, finding that the apparent correlation is largely explained by language resource availability rather than tokenizer behavior alone.
In this work, we introduce Indic DiarBench, a speaker diarization and ASR benchmark dataset spanning all 22 scheduled languages of India. This corpus comprises approximately 108 hours of natural multi-speaker audio from near-field meetings, far-field recordings, and in-the-wild audios. All annotations are human-corrected with time-aligned speaker attributed transcriptions. The dataset captures conversational nuance prevalent in Indian speech, such as English code-mixing, dialectal variation, and frequent speaker overlap. To establish a baseline for joint ASR and diarization capabilities we evaluate leading systems including commercial speech APIs and multimodal large language models. Indic DiarBench is released as an open-access resource to advance inclusive, multilingual speech technology research for Indian languages.
Sign language is a primary mode of communication for the global deaf and hard-of-hearing community, yet automated tools that recognize sign gestures from video and translate them into natural language text remain limited, particularly for low-resource Indian languages. We present a two-stage deep learning pipeline that (i) classifies short sign language video clips into English word labels using a fine-tuned VideoMAE video transformer, and (ii) translates the predicted English label into Hindi, Telugu, and Bengali using Meta AI's No Language Left Behind (NLLB-200) multilingual translation model. The classification model is fine-tuned on a 13-class subset of the AI4Bharat Indian Sign Language video corpus from IIT Madras, processing 16-frame clips sampled uniformly from each video at 224 x 224 resolution. Under a small-scale academic setting (13 classes, 197 clips, 80-20 split), the fine-tuned model reaches 99% training accuracy and 78% validation accuracy after 15 epochs. We provide a per-class breakdown via a confusion matrix and classification report, identify the dominant failure modes (confusable adjective pairs such as ugly, deaf, blind, hat, and dress), and describe a Streamlit-based inference demo that takes a user-uploaded video and returns the predicted English label alongside its Hindi, Telugu, and Bengali translations. We discuss the scope, limitations (small label set, isolated-word rather than continuous signing, single-signer style sensitivity, ambiguity of single-word machine translation), and directions for future work, including expanding to sentence-level generation and a larger vocabulary. Code is released to support reproducibility.