KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/kmitl/1

Browse

Search Results

Now showing 1 - 9 of 9

Some of the metrics are blocked by your
consent settings
Enhancing Burmese News Classification with Kolmogorov-Arnold Network Head Fine-Tuning
(2025-01-01)
Aung, Thura
;
Kyaw, Eaint Kay Khaing
;
Thu, Ye Kyaw
;
Oo, Thazin Myint
;
Supnithi, Thepchai
In low-resource languages like Burmese, classification tasks often fine-tune only the final classification layer, keeping pre-trained encoder weights frozen. While Multi-Layer Perceptrons (MLPs) are commonly used, their fixed non-linearity can limit expressiveness and increase computational cost. This work explores Kolmogorov-Arnold Networks (KANs) as alternative classification heads, evaluating Fourierbased FourierKAN, Spline-based EfficientKAN, and Grid-based FasterKAN-across diverse embeddings including TF-IDF, fastText, and multilingual transformers (mBERT, Distil-mBERT). Experimental results show that KAN-based heads are competitive with or superior to MLPs. EfficientKAN with fast-Text achieved the highest F1-score (0.928), while FasterKAN offered the best trade-off between speed and accuracy. On transformer embeddings, EfficientKAN matched or slightly outperformed MLPs with mBERT (0.917 F1). These findings highlight KANs as expressive, efficient alternatives to MLPs for lowresource language classification.
Some of the metrics are blocked by your
consent settings
DataDecon: Data Cleansing Tools for Large Language Model with Efficient Decontamination Techniques
(2024-01-01)
Yuenyong, Sumeth
;
Buppodom, Norapat
;
Sangkaew, Koravich
;
Boonmeeprakob, Konthee
;
Boonkwan, Prachya
Large language models (LLMs) play an important role in modern NLP technology as they are versatile for a wide array of NLP tasks. However, constructing an LLM is challenging due to concealed construction pipelines, the lack of cleansed datasets, and hyperparameter settings, making it almost irreproducible. This paper presents an efficient pipeline for constructing an LLM tailored to a low-to-medium-sourced language with a high level of data contamination and tools to cleanse the dataset. Following our pipeline, we constructed OpenThaiGPT, an LLM for Thai, with only open-sourced datasets such as CC100, OSCAR, and mC4, and achieved the state-of-the-art accuracies on our downstream tasks. Here, we disclosed the data statistics and all hyperparameter settings for reproducibility.
Some of the metrics are blocked by your
consent settings
Improving neural machine translation with POS-tag features for low-resource language pairs
(2022-08-01)
Hlaing, Zar Zar
;
Thu, Ye Kyaw
;
Supnithi, Thepchai
;
Netisopakul, Ponrudee
Integrating linguistic features has been widely utilized in statistical machine translation (SMT) systems, resulting in improved translation quality. However, for low-resource languages such as Thai and Myanmar, the integration of linguistic features in neural machine translation (NMT) systems has yet to be implemented. In this study, we propose transformer-based NMT models (transformer, multi-source transformer, and shared-multi-source transformer models) using linguistic features for two-way translation of Thai-to-Myanmar, Myanmar-to-English, and Thai-to-English. Linguistic features such as part-of-speech (POS) tags or universal part-of-speech (UPOS) tags are added to each word on either the source or target side, or both the source and target sides, and the proposed models are conducted. The multi-source transformer and shared-multi-source transformer models take two inputs (i.e., string data and string data with POS tags) and produce string data or string data with POS tags. A transformer model that utilizes only word vectors was used as the first baseline model for comparison with the proposed models. The second baseline model, an Edit-Based Transformer with Repositioning (EDITOR) model, was also used to compare with our proposed models in addition to the baseline transformer model. The findings of the experiments show that adding linguistic features to the transformer-based models enhances the performance of a neural machine translation in low-resource language pairs. Moreover, the best translation results were yielded using shared-multi-source transformer models with linguistic features resulting in more significant Bilingual Evaluation Understudy (BLEU) scores and character n-gram F-score (chrF) scores than the baseline transformer and EDITOR models.
Some of the metrics are blocked by your
consent settings
Graph-based Dependency Parser Building for Myanmar Language
(2022-01-01)
Hlaing, Zar Zar
;
Thu, Ye Kyaw
;
Supnithi, Thepchai
;
Netisopakul, Ponrudee
Examining the relationships between words in a sentence to determine its grammatical structure is known as dependency parsing (DP). Based on this, a sentence is broken down into several components. The process is based on the concept that every linguistic component of a sentence has a direct relationship to one another. These relationships are called dependencies. Dependency parsing is one of the key steps in natural language processing (NLP) for several text mining approaches. As the dominant formalism for dependency parsing in recent years, Universal Dependencies (UD) have emerged. The various UD corpus and dependency parsers are publicly accessible for resource-rich languages. However, there are no publicly available resources for dependency parsing, especially for the low-resource language, Myanmar. Thus, we manually extended the existing small Myanmar UD corpus (i.e., myPOS UD corpus) as myPOS version 3.0 UD corpus to publish the extended Myanmar UD corpus as the publicly available resource. To evaluate the effects of the extended UD corpus versus the original UD corpus, we utilized the graph-based neural dependency parsing models, namely, jPTDP (joint POS tagging and dependency parsing) and UniParse (universal graph-based parsing), and the evaluation scores are measured in terms of unlabeled and labeled attachment scores: (UAS) and (LAS). We compared the accuracies of graph-based neural models based on the original and extended UD corpora. The experimental results showed that, compared to the original myPOS UD corpus, the extended myPOS version 3.0 UD corpus enhanced the accuracy of dependency parsing models.
Some of the metrics are blocked by your
consent settings
NECTEC’s Participation in WAT-2021
(2021-01-01)
Hlaing, Zar Zar
;
Thu, Ye Kyaw
;
Oo, Thazin Myint
;
San, Mya Ei
;
Usanavasin, Sasiporn
In this paper, we report the experimental results of Machine Translation models conducted by a NECTEC team (Team-ID: NECTEC) for the WAT-2021 Myanmar-English translation task (Nakazawa et al., 2021). Basically, our models are based on neural methods for both directions of English-Myanmar and Myanmar-English language pairs. Most of the existing Neural Machine Translation (NMT) models mainly focus on the conversion of sequential data and do not directly use syntactic information. However, we conduct multi-source neural machine translation (NMT) models using the multilingual corpora such as string data corpus, tree data corpus, or POS-tagged data corpus. The multi-source translation is an approach to exploit multiple inputs (e.g. in two different formats) to increase translation accuracy. The RNN-based encoder-decoder model with attention mechanism and transformer architectures have been carried out for our experiment. The experimental results showed that the proposed models of RNN-based architecture outperform the baseline model for the English-to-Myanmar translation task, and the multi-source and sharedmulti-source transformer models yield better translation results than the baseline.
Some of the metrics are blocked by your
consent settings
Myanmar POS Resource Extension Effects on Automatic Tagging Methods
(2020-11-18)
Hlaing, Zar Zar
;
Thu, Ye Kyaw
;
Wai, Myat Myo Nwe
;
Supnithi, Thepchai
;
Netisopakul, Ponrudee
Part-of-speech (POS) tagging is the process of assigning the part-of-speech tag or other lexical class marker to each word in a sentence. It is also one of the most important steps in Natural Language Processing (NLP) task pipeline. There are several research works in Myanmar POS tagging implemented with different approaches. However, there is only one publicly available tagged corpus named myPOS corpus. The size of this corpus is only 11 thousand sentences. It is not enough to train downstream NLP tasks, such as machine learning. For this reason, we manually extended the original myPOS corpus as myPOS version 2.0 and the size of the extended corpus becomes approximately triple size of the original myPOS corpus. To evaluate the effects of the extended corpus versus the original corpus, the accuracies of four supervised tagging algorithms, namely, Conditional Random Fields (CRFs), Hidden Markov Model (HMM), Ripple Down Rules based (RDR), and neural sequence labeling approach of Conditional Random Fields (NCRF++) are compared. The results showed that the extended myPOS version 2.0 improved the accuracies of automatic POS tagging methods compared with the original myPOS.
Some of the metrics are blocked by your
consent settings
Increasing SMT and NMT Performance by Corpus Extension with Free Online Machine Translation Services
(2020-11-04)
Hlaing, Zar Zar
;
Thu, Ye Kyaw
;
Supnithi, Thepchai
;
Netisopakul, Ponrudee
In machine translation, parallel corpora of source-Target language pair are essential to improve the performance of the translation. However, the existing parallel corpora for the low resource language is not sufficient to improve the quality of the translation. In this paper, we explore the role of corpus extension by using the three freely available online machine translation services; 'Google Translate', 'SYSTRAN Translate' and 'Yandex Translate' for English and Thai language pair. We compare three statistical and neural machine translation performances between the original ASEAN-MT corpus, and their extended version, which double the original size of the ASEAN-MT. The results showed that, for SMT models, extended Thai corpus can help improve the translation performance for th-en translation up to 2.6% and the extended English corpus can do so significantly for en-Th translation up to 4.2%. While for the NMT model, the extended Thai corpus can improve the translation performance up to 5.5%.
Some of the metrics are blocked by your
consent settings
Statistical level checker with personalised English passage suggestion
(2011-12-01)
Chai, Wasan Na
;
Ruangrajitpakorn, Taneth
;
Hiransakolwong, Nualsawat
;
Supnithi, Thepchai
In this paper, a system to classify a readability level of English reading passage and to match student personal interest is purposed. Student model is applied to collect student information for selecting their preferable passage topic. Statistical passage level checker is implemented to match student readability level with passage difficulty by using neural network. Three linguistic features, syllable, vocabulary and sentence complexity, are chosen to distinguish a difficulty difference among passage level. The best accuracy gained by the system is 86.25% and the constantly reliable feature for this task is a sentence complexity of the passage.
Some of the metrics are blocked by your
consent settings
English-Thai example-based machine translation using n-gram model
(2006-01-01)
Kritsuthikul, Nattapol
;
Thammano, Arit
;
Supnithi, Thepchai
The necessity on exchanging information among countries become a major task in information based society. Machine translation is an application that enables users to communicate each other without language barrier problem. With the great support on computer's efficiency, corpus-based technology becomes a fundamental concept for developing software based on a large amount of data. We introduce the first example-based English to Thai machine translation using n-gram model and implemented the system. Some advantages and disadvantages of this method are discussed. © 2006 IEEE.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results