Attapol Rutherford

Peerawat Chomphooyod, Jian Gang Ngui, Yosephine Susanto et al.

Frontier LLMs perform well in Western contexts, but remain poorly tested on underrepresented cultures such as those in Southeast Asia (SEA). Existing NLI benchmarks are largely Western-centric, translation-derived, or monolingual, limiting their ability to measure culturally grounded reasoning. We introduce SEA-NLI, a native, culturally grounded NLI benchmark covering eight SEA countries in English and native regional languages, verified by native speakers. Across 17 encoder and decoder models, we observe a low performance from all models, especially for knowledge-intensive categories such as Languages and Science and Technology. Our analysis shows that failure cases mainly stem from missing SEA cultural knowledge: SEA-adapted models and culture-aware prompting improve performance, while CoT prompting offers limited gains.

Panyut Sriwirote, Jalinee Thapiang, Vasan Timtong et al.

While WangchanBERTa has become the de facto standard in transformer-based Thai language modeling, it still has shortcomings in regard to the understanding of foreign words, most notably English words, which are often borrowed without orthographic assimilation into Thai in many contexts. We identify the lack of foreign vocabulary in WangchanBERTa's tokenizer as the main source of these shortcomings. We then expand WangchanBERTa's vocabulary via vocabulary transfer from XLM-R's pretrained tokenizer and pretrain a new model using the expanded tokenizer, starting from WangchanBERTa's checkpoint, on a new dataset that is larger than the one used to train WangchanBERTa. Our results show that our new pretrained model, PhayaThaiBERT, outperforms WangchanBERTa in many downstream tasks and datasets.

Jullajak Karnjanaekarin, Pontakorn Trakuekul, Narongkorn Panitsrisit et al.

We present JaiTTS-v1.0, a state-of-the-art Thai voice cloning text-to-speech model built through continual training on a large Thai-centric speech corpus. The model architecture is adapted from VoxCPM, a tokenizer-free autoregressive TTS model. JaiTTS-v1.0 directly processes numerals and Thai-English code-switching, which is very common in realistic settings, without explicit text normalization. We test the models on short-duration speech generation and long-duration speech generation, which reflects many real-world use cases. JaiTTS-v1.0 achieves a state-of-the-art CER of 1.94\%, surpassing the human ground truth of 1.98% for short-duration tasks while performing on par with human ground truth for long-duration tasks. In human judgment evaluations, our model wins 283 of 400 pairwise comparisons against commercial flagships, with only 58 losses.

Pontakorn Trakuekul, Attapol T. Rutherford, Jullajak Karnjanaekarin et al.

We introduce OpenJAI-v1.0, an open-source large language model for Thai and English, developed from the Qwen3-14B model. Our work focuses on boosting performance on practical tasks through carefully curated data across three key use cases: instruction following, long-context understanding, and tool use. Evaluation results show that OpenJAI-v1.0 improves on the capabilities of its base model and outperforms other leading open-source Thai models on a diverse suite of benchmarks, while avoiding catastrophic forgetting. OpenJAI-v1.0 is publicly released as another alternative NLP resource for the Thai AI community.

Attapol T. Rutherford, Jullajak Karnjanaekarin, Narongkorn Panitsrisit et al.

This technical report introduces JAI-1, a Thai-centric language model with 75B parameters. Recent Thai models have primarily relied on existing open-source models, applying additional training without structural modifications to specialize in Thai. However, this approach risks eroding pre-existing knowledge in the model's parameter space during the injection of Thai-specific information, as optimized parameters for general tasks may conflict with new linguistic requirements. In contrast, JAI-1 adopts an upscaling strategy: starting from a smaller, high-performing English open-source LLM, we expanded its parameter space and utilized the newly allocated capacity to systematically integrate Thai-language knowledge. This methodology not only preserves the original model's general intelligence but also establishes a unique architecture distinct from other open-source models, enabling scalable future enhancements. During pre-training, JAI-1 was exposed to 1.5T tokens, including over 300B Thai language tokens. This was followed by post-training stages -- supervised fine-tuning and alignment tuning -- using more than 600K instruction-based examples. The final model demonstrated superior performance compared to Typhoon2-70B on Thai-centric benchmarks (IFEval-TH, MT-Bench-TH, and JAI-Hall-Bench), validating the efficacy of its upscaling and knowledge-integration framework.

Pontakorn Trakuekul, Wei Qi Leong, Charin Polpanumas et al.

While coreference resolution is a well-established research area in Natural Language Processing (NLP), research focusing on Thai language remains limited due to the lack of large annotated corpora. In this work, we introduce ThaiCoref, a dataset for Thai coreference resolution. Our dataset comprises 777,271 tokens, 44,082 mentions and 10,429 entities across four text genres: university essays, newspapers, speeches, and Wikipedia. Our annotation scheme is built upon the OntoNotes benchmark with adjustments to address Thai-specific phenomena. Utilizing ThaiCoref, we train models employing a multilingual encoder and cross-lingual transfer techniques, achieving a best F1 score of 67.88\% on the test set. Error analysis reveals challenges posed by Thai's unique linguistic features. To benefit the NLP community, we make the dataset and the model publicly available at http://www.github.com/nlp-chula/thai-coref .

Napat Laosaengpha, Thanit Tativannarat, Attapol Rutherford et al.

Understanding the textual components of resumes and job postings is critical for improving job-matching accuracy and optimizing job search systems in online recruitment platforms. However, existing works primarily focus on analyzing individual components within this information, requiring multiple specialized tools to analyze each aspect. Such disjointed methods could potentially hinder overall generalizability in recruitment-related text processing. Therefore, we propose a unified sentence encoder that utilized multi-task dual-encoder framework for jointly learning multiple component into the unified sentence encoder. The results show that our method outperforms other state-of-the-art models, despite its smaller model size. Moreover, we propose a novel metric, Language Bias Kullback-Leibler Divergence (LBKL), to evaluate language bias in the encoder, demonstrating significant bias reduction and superior cross-lingual performance.

Panyut Sriwirote, Wei Qi Leong, Charin Polpanumas et al.

Automatic dependency parsing of Thai sentences has been underexplored, as evidenced by the lack of large Thai dependency treebanks with complete dependency structures and the lack of a published systematic evaluation of state-of-the-art models, especially transformer-based parsers. In this work, we address these problems by introducing Thai Universal Dependency Treebank (TUD), a new largest Thai treebank consisting of 3,627 trees annotated in accordance with the Universal Dependencies (UD) framework. We then benchmark dependency parsing models that incorporate pretrained transformers as encoders and train them on Thai-PUD and our TUD. The evaluation results show that most of our models can outperform other models reported in previous papers and provide insight into the optimal choices of components to include in Thai dependency parsers. The new treebank and every model's full prediction generated in our experiment are made available on a GitHub repository for further study.

Davor Lauc, Attapol Rutherford, Weerin Wongwarawipatr

This study introduces AyutthayaAlpha, an advanced transformer-based machine learning model designed for the transliteration of Thai proper names into Latin script. Our system achieves state-of-the-art performance with 82.32% first-token accuracy and 95.24% first-three-token accuracy, while maintaining a low character error rate of 0.0047. The complexity of Thai phonology, including tonal features and vowel length distinctions, presents significant challenges for accurate transliteration, which we address through a novel two-model approach: AyutthayaAlpha-Small, based on the ByT5 architecture, and AyutthayaAlpha-VerySmall, a computationally efficient variant that unexpectedly outperforms its larger counterpart. Our research combines linguistic rules with deep learning, training on a carefully curated dataset of 1.2 million Thai-Latin name pairs, augmented through strategic upsampling to 2.7 million examples. Extensive evaluations against existing transliteration methods and human expert benchmarks demonstrate that AyutthayaAlpha not only achieves superior accuracy but also effectively captures personal and cultural preferences in name romanization. The system's practical applications extend to cross-lingual information retrieval, international data standardization, and identity verification systems, with particular relevance for government databases, academic institutions, and global business operations. This work represents a significant advance in bridging linguistic gaps between Thai and Latin scripts, while respecting the cultural and personal dimensions of name transliteration.

Napat Laosaengpha, Thanit Tativannarat, Chawan Piansaddhayanon et al.

Learning job title representation is a vital process for developing automatic human resource tools. To do so, existing methods primarily rely on learning the title representation through skills extracted from the job description, neglecting the rich and diverse content within. Thus, we propose an alternative framework for learning job titles through their respective job description (JD) and utilize a Job Description Aggregator component to handle the lengthy description and bidirectional contrastive loss to account for the bidirectional relationship between the job title and its description. We evaluated the performance of our method on both in-domain and out-of-domain settings, achieving a superior performance over the skill-based approach.

Jin Cheevaprawatdomrong, Alexandra Schofield, Attapol T. Rutherford

Traditionally, Latent Dirichlet Allocation (LDA) ingests words in a collection of documents to discover their latent topics using word-document co-occurrences. However, it is unclear how to achieve the best results for languages without marked word boundaries such as Chinese and Thai. Here, we explore the use of Pearson's chi-squared test, t-statistics, and Word Pair Encoding (WPE) to produce tokens as input to the LDA model. The Chi-squared, t, and WPE tokenizers are trained on Wikipedia text to look for words that should be grouped together, such as compound nouns, proper nouns, and complex event verbs. We propose a new metric for measuring the clustering quality in settings where the vocabularies of the models differ. Based on this metric and other established metrics, we show that topics trained with merged tokens result in topic keys that are clearer, more coherent, and more effective at distinguishing topics than those unmerged models.

Lalita Lowphansirikul, Charin Polpanumas, Attapol T. Rutherford et al.

The primary objective of our work is to build a large-scale English-Thai dataset for machine translation. We construct an English-Thai machine translation dataset with over 1 million segment pairs, curated from various sources, namely news, Wikipedia articles, SMS messages, task-based dialogs, web-crawled data and government documents. Methodology for gathering data, building parallel texts and removing noisy sentence pairs are presented in a reproducible manner. We train machine translation models based on this dataset. Our models' performance are comparable to that of Google Translation API (as of May 2020) for Thai-English and outperform Google when the Open Parallel Corpus (OPUS) is included in the training data for both Thai-English and English-Thai translation. The dataset, pre-trained models, and source code to reproduce our work are available for public use.

Pattarawat Chormai, Ponrawee Prasertsom, Attapol Rutherford

Word segmentation is a fundamental pre-processing step for Thai Natural Language Processing. The current off-the-shelf solutions are not benchmarked consistently, so it is difficult to compare their trade-offs. We conducted a speed and accuracy comparison of the popular systems on three different domains and found that the state-of-the-art deep learning system is slow and moreover does not use sub-word structures to guide the model. Here, we propose a fast and accurate neural Thai Word Segmenter that uses dilated CNN filters to capture the environment of each character and uses syllable embeddings as features. Our system runs at least 5.6x faster and outperforms the previous state-of-the-art system on some domains. In addition, we develop the first ML-based Thai orthographical syllable segmenter, which yields syllable embeddings to be used as features by the word segmenter.

Attapol T. Rutherford, Vera Demberg, Nianwen Xue

Inferring implicit discourse relations in natural language text is the most difficult subtask in discourse parsing. Surface features achieve good performance, but they are not readily applicable to other languages without semantic lexicons. Previous neural models require parses, surface features, or a small label set to work well. Here, we propose neural network models that are based on feedforward and long-short term memory architecture without any surface features. To our surprise, our best configured feedforward architecture outperforms LSTM-based model in most cases despite thorough tuning. Under various fine-grained label sets and a cross-linguistic setting, our feedforward models perform consistently better or at least just as well as systems that require hand-crafted surface features. Our models present the first neural Chinese discourse parser in the style of Chinese Discourse Treebank, showing that our results hold cross-linguistically.