arXiv:2609. 21362v1 Announce Type: new Abstract: Conventional tokenizers represent text as characters or statistically derived subwords, overlooking the internal phonological structure of syllables and often requiring large vocabularies.
The paper presents an end‑to‑end sequence‑to‑sequence approach for correcting Tamil spelling and grammar errors, leveraging progressively fine‑tuned transformer models (mT5‑small and mBART‑50). Using a synthetic corpus of 657,720 noisy‑clean sentence pairs across ten error categories, the authors introduce a four‑stage training schedule that targets surface noise, contextual grammar, single‑site sandhi, and multi‑site cross‑word sandhi. The best model, mBART‑50 v5, achieves 69.3% exact‑match accuracy on a balanced diagnostic set, with notable gains in sandhi (87.5%) and subject‑verb agreement (43.5%) accuracy, while also revealing a precision‑recall trade‑off for sandhi corrections.
The paper introduces SinLlama, the first decoder‑based open‑source large language model with explicit support for Sinhala. By extending Llama‑3‑8B, adding Sinhala‑specific tokenizer vocabulary, and performing continual pre‑training on a cleaned 10‑million‑token Sinhala corpus, the authors created a model that surpasses both the base and instruction‑fine‑tuned variants of Llama‑3‑8B on three text classification tasks. This work addresses the underrepresentation of low‑resource languages in open‑source LLMs.
arXiv:2606. 15144v1 Announce Type: cross Abstract: Large language models (LLMs) process text as sequences of subword tokens, which can obscure the character-level and morphological structure that underlies word formation.
arXiv:2608. 01281v1 Announce Type: cross Abstract: Phoneme-based multilingual automatic speech recognition (ASR) can share acoustic evidence across languages more directly than language-specific subword modeling.
arXiv:2606. 18717v1 Announce Type: cross Abstract: Turkish is agglutinative: meaning is carried by morphemes, yet the subword tokenizers that drive modern language models split words by corpus statistics, fragmenting semantically loaded suffixes and -- in the case of WordPiece and rule-based analyzers -- failing to decode their output back to the original text.
arXiv:2609.22138v1 Announce Type: cross Abstract: Hidden-state hallucination probing - training a linear classifier on an LLM's internal activations to detect whether a generated answer is faithful t...
arXiv:2608. 03494v1 Announce Type: cross Abstract: Vocabulary extension is an efficient way to adapt pretrained large language models (LLMs) to new languages, but the initialization of newly added token embeddings can strongly affect continued pre-training (CPT) efficiency.
arXiv:2609.24310v1 Announce Type: cross Abstract: Text-to-speech models for Bantu tonal languages are challenged by a tonal system that is rooted in both the lexis (i.e., the inventory of words, stem...
T‑SANDHI is a lightweight Taiwanese Hokkien ASR model that addresses tone sandhi by decoupling surface acoustics from lexical intent on a frozen Whisper backbone. It uses a lexicon‑guided multi‑task learning framework with text‑derived pseudo labels and a hybrid injection module that dynamically gates independent citation and sandhi phonetic streams. Experiments on the TAT‑MOE corpus and two blind test sets show that this explicit disentanglement resolves tonal mapping confusion and outperforms baselines while keeping parameter usage low.
arXiv:2607. 23319v1 Announce Type: cross Abstract: Standard subword tokenization algorithms such as Byte-Pair Encoding (BPE) and SentencePiece are trained predominantly on modern language corpora and produce inefficient segmentations when applied to classical Indian languages.
The paper introduces a cross-dialect Named Entity Recognition (NER) framework for Bangla, leveraging the ANCHOLIK-NER dataset that covers five major regional dialects. Using a Leave-One-Dialect-Out Cross-Validation strategy, eight transformer-based models were evaluated, with Multilingual-E5 Large achieving the best performance (F1 up to 97.26% on Mymensingh, 82.38% on Chattogram). Local Interpretable Model-agnostic Explanations (LIME) revealed that the models rely mainly on the surface form of entity words rather than surrounding context, suggesting a direction for future improvement.