Jan Christian Blaise Cruz

Zheng-Xin Yong, Ruochen Zhang, Jessica Zosa Forde et al. · cmu, meta-ai

While code-mixing is a common linguistic practice in many parts of the world, collecting high-quality and low-cost code-mixed data remains a challenge for natural language processing (NLP) research. The recent proliferation of Large Language Models (LLMs) compels one to ask: how capable are these systems in generating code-mixed data? In this paper, we explore prompting multilingual LLMs in a zero-shot manner to generate code-mixed data for seven languages in South East Asia (SEA), namely Indonesian, Malay, Chinese, Tagalog, Vietnamese, Tamil, and Singlish. We find that publicly available multilingual instruction-tuned models such as BLOOMZ and Flan-T5-XXL are incapable of producing texts with phrases or clauses from different languages. ChatGPT exhibits inconsistent capabilities in generating code-mixed texts, wherein its performance varies depending on the prompt template and language pairing. For instance, ChatGPT generates fluent and natural Singlish texts (an English-based creole spoken in Singapore), but for English-Tamil language pair, the system mostly produces grammatically incorrect or semantically meaningless utterances. Furthermore, it may erroneously introduce languages not specified in the prompt. Based on our investigation, existing multilingual LLMs exhibit a wide range of proficiency in code-mixed data generation for SEA languages. As such, we advise against using LLMs in this context without extensive human checks.

Jan Christian Blaise Cruz

In this paper, we describe the constrained MT systems submitted by Samsung R&D Institute Philippines to the WMT 2023 General Translation Task for two directions: en$\rightarrow$he and he$\rightarrow$en. Our systems comprise of Transformer-based sequence-to-sequence models that are trained with a mix of best practices: comprehensive data preprocessing pipelines, synthetic backtranslated data, and the use of noisy channel reranking during online decoding. Our models perform comparably to, and sometimes outperform, strong baseline unconstrained systems such as mBART50 M2M and NLLB 200 MoE despite having significantly fewer parameters on two public benchmarks: FLORES-200 and NTREX-128.

Gabriel Louis Tan, Adrian Paule Ty, Schuyler Ng et al.

Response generation is a task in natural language processing (NLP) where a model is trained to respond to human statements. Conversational response generators take this one step further with the ability to respond within the context of previous responses. While there are existing techniques for training such models, they all require an abundance of conversational data which are not always available for low-resource languages. In this research, we make three contributions. First, we released the first Filipino conversational dataset collected from a popular Philippine online forum, which we named the PEx Conversations Dataset. Second, we introduce a data augmentation (DA) methodology for Filipino data by employing a Tagalog RoBERTa model to increase the size of the existing corpora. Lastly, we published the first Filipino conversational response generator capable of generating responses related to the previous 3 responses. With the supplementary synthetic data, we were able to improve the performance of the response generator by up to 12.2% in BERTScore, 10.7% in perplexity, and 11.7% in content word usage as compared to training with zero synthetic data.

Jan Christian Blaise Cruz, Alham Fikri Aji

Benchmarks and leaderboards are how NLP most often communicates progress, but in the LLM era they are increasingly easy to misread. Scores can reflect benchmark-chasing, hidden evaluation choices, or accidental exposure to test content -- not just broad capability. Closed benchmarks delay some of these issues, but reduce transparency and make it harder for the community to learn from results. We argue for a complementary practice: an Olympiad-style evaluation event where problems are sealed until evaluation, submissions are frozen in advance, and all entries run through one standardized harness. After scoring, the full task set and evaluation code are released so results can be reproduced and audited. This design aims to make strong performance harder to ``manufacture'' and easier to trust.

Genta Indra Winata, David Anugraha, Emmy Liu et al. · amazon-science

High-quality datasets are fundamental to training and evaluating machine learning models, yet their creation-especially with accurate human annotations-remains a significant challenge. Many dataset paper submissions lack originality, diversity, or rigorous quality control, and these shortcomings are often overlooked during peer review. Submissions also frequently omit essential details about dataset construction and properties. While existing tools such as datasheets aim to promote transparency, they are largely descriptive and do not provide standardized, measurable methods for evaluating data quality. Similarly, metadata requirements at conferences promote accountability but are inconsistently enforced. To address these limitations, this position paper advocates for the integration of systematic, rubric-based evaluation metrics into the dataset review process-particularly as submission volumes continue to grow. We also explore scalable, cost-effective methods for synthetic data generation, including dedicated tools and LLM-as-a-judge approaches, to support more efficient evaluation. As a call to action, we introduce DataRubrics, a structured framework for assessing the quality of both human- and model-generated datasets. Leveraging recent advances in LLM-based evaluation, DataRubrics offers a reproducible, scalable, and actionable solution for dataset quality assessment, enabling both authors and reviewers to uphold higher standards in data-centric research. We also release code to support reproducibility of LLM-based evaluations at https://github.com/datarubrics/datarubrics.

Genta Indra Winata, Frederikus Hudi, Patrick Amadeus Irawan et al.

Vision Language Models (VLMs) often struggle with culture-specific knowledge, particularly in languages other than English and in underrepresented cultural contexts. To evaluate their understanding of such knowledge, we introduce WorldCuisines, a massive-scale benchmark for multilingual and multicultural, visually grounded language understanding. This benchmark includes a visual question answering (VQA) dataset with text-image pairs across 30 languages and dialects, spanning 9 language families and featuring over 1 million data points, making it the largest multicultural VQA benchmark to date. It includes tasks for identifying dish names and their origins. We provide evaluation datasets in two sizes (12k and 60k instances) alongside a training dataset (1 million instances). Our findings show that while VLMs perform better with correct location context, they struggle with adversarial contexts and predicting specific regional cuisines and languages. To support future research, we release a knowledge base with annotated food entries and images along with the VQA data.

Jan Christian Blaise Cruz, Alham Fikri Aji

In this paper, we propose the use of simple knowledge distillation to produce smaller and more efficient single-language transformers from Massively Multilingual Transformers (MMTs) to alleviate tradeoffs associated with the use of such in low-resource settings. Using Tagalog as a case study, we show that these smaller single-language models perform on-par with strong baselines in a variety of benchmark tasks in a much more efficient manner. Furthermore, we investigate additional steps during the distillation process that improves the soft-supervision of the target language, and provide a number of analyses and ablations to show the efficacy of the proposed method.

Samuel Cahyawijaya, Ruochen Zhang, Holy Lovenia et al.

Multilingual large language models (LLMs) have gained prominence, but concerns arise regarding their reliability beyond English. This study addresses the gap in cross-lingual semantic evaluation by introducing a novel benchmark for cross-lingual sense disambiguation, StingrayBench. In this paper, we demonstrate using false friends -- words that are orthographically similar but have completely different meanings in two languages -- as a possible approach to pinpoint the limitation of cross-lingual sense disambiguation in LLMs. We collect false friends in four language pairs, namely Indonesian-Malay, Indonesian-Tagalog, Chinese-Japanese, and English-German; and challenge LLMs to distinguish the use of them in context. In our analysis of various models, we observe they tend to be biased toward higher-resource languages. We also propose new metrics for quantifying the cross-lingual sense bias and comprehension based on our benchmark. Our work contributes to developing more diverse and inclusive language modeling, promoting fairer access for the wider multilingual community.

Emilio Villa-Cueva, Sholpan Bolatzhanova, Diana Turmakhan et al.

Translating cultural content poses challenges for machine translation systems due to the differences in conceptualizations between cultures, where language alone may fail to convey sufficient context to capture region-specific meanings. In this work, we investigate whether images can act as cultural context in multimodal translation. We introduce CaMMT, a human-curated benchmark of over 5,800 triples of images along with parallel captions in English and regional languages. Using this dataset, we evaluate five Vision Language Models (VLMs) in text-only and text+image settings. Through automatic and human evaluations, we find that visual context generally improves translation quality, especially in handling Culturally-Specific Items (CSIs), disambiguation, and correct gender marking. By releasing CaMMT, our objective is to support broader efforts to build and evaluate multimodal translation systems that are better aligned with cultural nuance and regional variations.

Emilio Villa-Cueva, S M Masrur Ahmed, Rendi Chevi et al.

Understanding Theory of Mind is essential for building socially intelligent multimodal agents capable of perceiving and interpreting human behavior. We introduce MoMentS (Multimodal Mental States), a comprehensive benchmark designed to assess the ToM capabilities of multimodal large language models (LLMs) through realistic, narrative-rich scenarios presented in short films. MoMentS includes over 2,300 multiple-choice questions spanning seven distinct ToM categories. The benchmark features long video context windows and realistic social interactions that provide deeper insight into characters' mental states. We evaluate several MLLMs and find that although vision generally improves performance, models still struggle to integrate it effectively. For audio, models that process dialogues as audio do not consistently outperform transcript-based inputs. Our findings highlight the need to improve multimodal integration and point to open challenges that must be addressed to advance AI's social understanding.

Holy Lovenia, Rahmad Mahendra, Salsabil Maulana Akbar et al.

Southeast Asia (SEA) is a region rich in linguistic diversity and cultural variety, with over 1,300 indigenous languages and a population of 671 million people. However, prevailing AI models suffer from a significant lack of representation of texts, images, and audio datasets from SEA, compromising the quality of AI models for SEA languages. Evaluating models for SEA languages is challenging due to the scarcity of high-quality datasets, compounded by the dominance of English training data, raising concerns about potential cultural misrepresentation. To address these challenges, we introduce SEACrowd, a collaborative initiative that consolidates a comprehensive resource hub that fills the resource gap by providing standardized corpora in nearly 1,000 SEA languages across three modalities. Through our SEACrowd benchmarks, we assess the quality of AI models on 36 indigenous languages across 13 tasks, offering valuable insights into the current AI landscape in SEA. Furthermore, we propose strategies to facilitate greater AI advancements, maximizing potential utility and resource equity for the future of AI in SEA.

Ruochen Zhang, Samuel Cahyawijaya, Jan Christian Blaise Cruz et al.

Multilingual Large Language Models (LLMs) have recently shown great capabilities in a wide range of tasks, exhibiting state-of-the-art performance through zero-shot or few-shot prompting methods. While there have been extensive studies on their abilities in monolingual tasks, the investigation of their potential in the context of code-switching (CSW), the practice of alternating languages within an utterance, remains relatively uncharted. In this paper, we provide a comprehensive empirical analysis of various multilingual LLMs, benchmarking their performance across four tasks: sentiment analysis, machine translation, summarization and word-level language identification. Our results indicate that despite multilingual LLMs exhibiting promising outcomes in certain tasks using zero or few-shot prompting, they still underperform in comparison to fine-tuned models of much smaller scales. We argue that current "multilingualism" in LLMs does not inherently imply proficiency with code-switching texts, calling for future research to bridge this discrepancy.

Lintang Sutawika, Jan Christian Blaise Cruz

In this paper, we describe the submission of the joint Samsung Research Philippines-Konvergen AI team for the WMT'21 Large Scale Multilingual Translation Task - Small Track 2. We submit a standard Seq2Seq Transformer model to the shared task without any training or architecture tricks, relying mainly on the strength of our data preprocessing techniques to boost performance. Our final submission model scored 22.92 average BLEU on the FLORES-101 devtest set, and scored 22.97 average BLEU on the contest's hidden test set, ranking us sixth overall. Despite using only a standard Transformer, our model ranked first in Indonesian to Javanese, showing that data preprocessing matters equally, if not more, than cutting edge model architectures and training techniques.

Jan Christian Blaise Cruz, Charibeth Cheng

In this paper, we improve on existing language resources for the low-resource Filipino language in two ways. First, we outline the construction of the TLUnified dataset, a large-scale pretraining corpus that serves as an improvement over smaller existing pretraining datasets for the language in terms of scale and topic variety. Second, we pretrain new Transformer language models following the RoBERTa pretraining technique to supplant existing models trained with small corpora. Our new RoBERTa models show significant improvements over existing Filipino models in three benchmark datasets with an average gain of 4.47% test accuracy across the three classification tasks of varying difficulty.

Jan Christian Blaise Cruz, Jose Kristian Resabal, James Lin et al.

Transformers represent the state-of-the-art in Natural Language Processing (NLP) in recent years, proving effective even in tasks done in low-resource languages. While pretrained transformers for these languages can be made, it is challenging to measure their true performance and capacity due to the lack of hard benchmark datasets, as well as the difficulty and cost of producing them. In this paper, we present three contributions: First, we propose a methodology for automatically producing Natural Language Inference (NLI) benchmark datasets for low-resource languages using published news articles. Through this, we create and release NewsPH-NLI, the first sentence entailment benchmark dataset in the low-resource Filipino language. Second, we produce new pretrained transformers based on the ELECTRA technique to further alleviate the resource scarcity in Filipino, benchmarking them on our dataset against other commonly-used transfer learning techniques. Lastly, we perform analyses on transfer learning techniques to shed light on their true performance when operating in low-data domains through the use of degradation tests.

Jan Christian Blaise Cruz, Charibeth Cheng

While transformer-based finetuning techniques have proven effective in tasks that involve low-resource, low-data environments, a lack of properly established baselines and benchmark datasets make it hard to compare different approaches that are aimed at tackling the low-resource setting. In this work, we provide three contributions. First, we introduce two previously unreleased datasets as benchmark datasets for text classification and low-resource multilabel text classification for the low-resource language Filipino. Second, we pretrain better BERT and DistilBERT models for use within the Filipino setting. Third, we introduce a simple degradation test that benchmarks a model's resistance to performance degradation as the number of training samples are reduced. We analyze our pretrained model's degradation speeds and look towards the use of this method for comparing models aimed at operating within the low-resource setting. We release all our models and datasets for the research community to use.

Luis Enrico Lopez, Diane Kathryn Cruz, Jan Christian Blaise Cruz et al.

Question generation (QG) is a natural language generation task where a model is trained to ask questions corresponding to some input text. Most recent approaches frame QG as a sequence-to-sequence problem and rely on additional features and mechanisms to increase performance; however, these often increase model complexity, and can rely on auxiliary data unavailable in practical use. A single Transformer-based unidirectional language model leveraging transfer learning can be used to produce high quality questions while disposing of additional task-specific complexity. Our QG model, finetuned from GPT-2 Small, outperforms several paragraph-level QG baselines on the SQuAD dataset by 0.95 METEOR points. Human evaluators rated questions as easy to answer, relevant to their context paragraph, and corresponding well to natural human speech. Also introduced is a new set of baseline scores on the RACE dataset, which has not previously been used for QG tasks. Further experimentation with varying model capacities and datasets with non-identification type questions is recommended in order to further verify the robustness of pretrained Transformer-based LMs as question generators.

Jan Christian Blaise Cruz, Julianne Agatha Tan, Charibeth Cheng

The use of the internet as a fast medium of spreading fake news reinforces the need for computational tools that combat it. Techniques that train fake news classifiers exist, but they all assume an abundance of resources including large labeled datasets and expert-curated corpora, which low-resource languages may not have. In this work, we make two main contributions: First, we alleviate resource scarcity by constructing the first expertly-curated benchmark dataset for fake news detection in Filipino, which we call "Fake News Filipino." Second, we benchmark Transfer Learning (TL) techniques and show that they can be used to train robust fake news classifiers from little data, achieving 91% accuracy on our fake news dataset, reducing the error by 14% compared to established few-shot baselines. Furthermore, lifting ideas from multitask learning, we show that augmenting transformer-based transfer techniques with auxiliary language modeling losses improves their performance by adapting to writing style. Using this, we improve TL performance by 4-6%, achieving an accuracy of 96% on our best model. Lastly, we show that our method generalizes well to different types of news articles, including political news, entertainment news, and opinion articles.

Jan Christian Blaise Cruz, Charibeth Cheng

Unlike mainstream languages (such as English and French), low-resource languages often suffer from a lack of expert-annotated corpora and benchmark resources that make it hard to apply state-of-the-art techniques directly. In this paper, we alleviate this scarcity problem for the low-resourced Filipino language in two ways. First, we introduce a new benchmark language modeling dataset in Filipino which we call WikiText-TL-39. Second, we show that language model finetuning techniques such as BERT and ULMFiT can be used to consistently train robust classifiers in low-resource settings, experiencing at most a 0.0782 increase in validation error when the number of training examples is decreased from 10K to 1K while finetuning using a privately-held sentiment dataset.