In the context of pretraining of Large Language Models (LLMs), synthetic data has emerged as an alternative for generating high-quality pretraining data at scale. This is particularly beneficial in low resource language settings where the benefits of the recent LLMs have been unevenly distributed across languages. In this work, we present a systematic study on the generation and evaluation of synthetic multilingual pretraining data for Indic languages, where we construct a large scale synthetic dataset BhashaKritika, comprising 540B tokens using 5 different techniques for 10 languages. We explore the impact of grounding generation in documents, personas and topics. We analyze how language choice, both in the prompt instructions and document grounding affects data quality and we compare translations of English content with native generation in Indic languages. In order to support scalable and language-sensitive evaluation, we introduce a modular quality evaluation pipeline that integrates script and language detection, metadata consistency checks, n-gram repetition analysis, and perplexity-based filtering using KenLM models. Our framework enables robust quality control across diverse scripts and linguistic contexts. Empirical results through model runs reveal key trade-offs in generation strategies and highlight best practices for constructing effective multilingual corpora. This work contributes practical insights for advanced pretraining recipes in low-resource and script-diverse settings, particularly in the Indian context.

Tokenizers play a crucial role in determining the performance, training efficiency, and the inference cost of Large Language Models (LLMs).Designing effective tokenizers for multilingual LLMs is particularly challenging due to diverse scripts and rich morphological variation.While subword methods like Byte Pair Encoding (BPE) are widely adopted, their effectiveness in multilingual settings remains underexplored.We present MUTANT, a recipe for building multilingual tokenizers, with careful vocabulary and training data design, language-aware pre-tokenization, and subword and multiword aware training.We also introduce MUTANT-Indic, a tokenizer for Indiaspecific multilingual LLMs, that produces linguistically coherent tokens and achieves stateof-the-art performance.Evaluated across English, 22 Indian languages and code data, our tokenizer improves the average fertility score by 39.5% over LLaMA4 and by 18% over Sutra (the current best).This translates to 44% improvement in inference throughput over LLaMA4 while maintaining comparable performance on English and Indic benchmarks.We present detailed ablations across tokenizer training data size, vocabulary size, merging techniques, and pre-tokenization strategies, demonstrating the robustness of our design choices.

Debrup Das, Debopriyo Banerjee, Somak Aditya, Ashish Kulkarni. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

We present a multi-speaker Japanese audiobook text-to-speech (TTS) system that leverages multimodal context information of preceding acoustic context and bilateral textual context to improve the prosody of synthetic speech. Previous work either uses unilateral or single-modality context, which does not fully represent the context information. The proposed method uses an acoustic context encoder and a textual context encoder to aggregate context information and feeds it to the TTS model, which enables the model to predict context-dependent prosody. We conducted comprehensive objective and subjective evaluations on a multi-speaker Japanese audiobook dataset. Experimental results demonstrate that the proposed method significantly outperforms two previous works. Additionally, we present insights about the different choices of context - modalities, lateral information and length - for audiobook TTS that have never been discussed in the literature before.

Robotic manipulators are increasingly being used to perform additive manufacturing. The accuracy of a built part is dependent on the trajectory execution error of the manipulator. For articulated manipulators, the trajectory execution error and achievable build accuracy vary considerably over the workspace. Therefore, the build accuracy depends on where the part is placed in the manipulator workspace. If the part is small compared to the manipulator workspace, its placement can be optimized to improve the accuracy. This paper provides experimental evidence that the placement of the parts changes its build accuracy. We model the trajectory execution error of the manipulator for additive manufacturing. We validate these errors by comparing the predicted errors with the experimental errors. Finally, we present an algorithm to optimize the part placement for improving built part accuracy during robot-based additive manufacturing.

Multi-label classification is crucial to several practical applications including document categorization, video tagging, targeted advertising etc. Training a multi-label classifier requires a large amount of labeled data which is often unavailable or scarce. Labeled data is then acquired by consulting multiple labelers---both human and machine. Inspired by ensemble methods, our premise is that labels inferred with high consensus among labelers, might be closer to the ground truth. We propose strategies based on interaction and active learning to obtain higher quality labels that potentially lead to greater consensus. We propose a novel formulation that aims to collectively optimize the cost of labeling, labeler reliability, label-label correlation and inter-labeler consensus. Evaluation on data labeled by multiple labelers (both human and machine) shows that our consensus output is closer to the ground truth when compared to the "majority" baseline. We present illustrative cases where it even improves over the existing ground truth. We also present active learning strategies to leverage our consensus model in interactive learning settings. Experiments on several real-world datasets (publicly available) demonstrate the efficacy of our approach in achieving promising classification results with fewer labeled data.

Translation systems are known to benefit from the availability of a bilingual lexicon for a domain of interest. A system, aiming to build such a lexicon from source language corpus, often requires human assistance and is confronted by conflicting requirements of minimizing human translation effort while improving the translation quality. We present an approach that exploits redundancy in the source corpus and extracts recurring patterns which are: frequent, syntactically well-formed, and provide maximum corpus coverage. The patterns generalize over phrases and word types and our approach finds a succinct set of good patterns with high coverage. Our interactive system leverages these patterns in multiple iterations of translation and post-editing, thereby progressively generating a high quality bilingual lexicon.

Within the broad area of information extraction, we study the problem of effective dictionary curation in an enterprise setting. Equipped with an ontology, representative of the domain of an enterprise, our approach populates the attributes of leaf nodes of the ontology with instances extracted from the enterprise corpus. For an attribute of interest, given a few seed examples or indicative features for the attribute, we first obtain a ranked list of 'list pages' potentially containing additional dictionary terms. Our ranking model ranks pages from the enterprise corpus based on their 'list' content using several visual and lexical features. We gather users' judgement of the result pages and the model continuously learns from this feedback. We compare different techniques of dictionary curation using rule based extractors and visual features of pages. Based on rule writing exercise, we show the benefit of dictionaries for leaf node attributes, in writing rule based extractors for higher level nodes in an ontology. We have implemented a dictionary curation system based on these ideas. Experimental analysis using academic domain ontology and universities corpora, reveal (in the context of enterprise analytics) (i) the merit of dictionary support in rule based information extraction (ii) the viability and effectiveness of an interactive approach for dictionary creation.