Hasil Pencarian  ::  Simpan CSV :: Kembali

Abstrak :
This article brings together two ostensibly separate subjects: language empowerment and the Covid-19 pandemic. It argues that knowledge of local languages can help disseminate health-related information on a regional level. This addresses two problems simultaneously: the problems raised by the intelligibilty of governmental healthcare protocols and the functions of the use of local languages. The article is a case study presenting a number of interventions in the languages of East Kalimantan and can be seen as an inclusive, grassroots example of health communication. The study was initially a modest attempt to generate on-theground examples of health information in the dominant languages of the region of East Kalimantan. These studies demonstrate that the local communities of these languages are very enthusiastic about getting involved in the interventions. They also reveal that communication using IT and social media is thriving. One of our observations was that information about this pandemic tends to be understood only by highly educated urban people. This happens because it is conveyed by the government in standard Indonesian, which includes many foreign loanwords. The application of local languages is not just using local language vocabulary, it is instead a trigger to revive the collective memory of disaster management based on local culture. In this case, local languages are recognized and considered useful in helping to break the chain of virus transmission to free Indonesia of the Covid-19 outbreak. There were a number of unexpected developments. We found support for the intervention being rolled out on a national level by Special Task Force for Covid-19 under National Disaster Management Agency (Badan Nasional Penanggulangan Bencana/BNPB). We also welcomed the online publication by the National Agency for Language Development and Cultivation (Badan Bahasa) of the Handbook for managing behavior about health protocols in seventy-seven local languages. The main thrust of this article should therefore be of interest to anyone working to empower local languages and language vitality.

Abstrak :
Neural machine translation (NMT) untuk bahasa daerah yang low resource di Indonesia menghadapi tantangan yang signifikan, meliputi kurangnya tolok ukur dasar yang representatif dan ketersediaan data yang terbatas. Penelitian ini mengatasi masalah tersebut dengan cara mengembangkan sebuah tolok ukur dasar yang bersifat replicable untuk empat bahasa daerah di Indonesia yang sering digunakan menggunakan sumber daya komputasi terbatas pada dataset FLORES-200. Penelitian ini mengadakan penyelidikan sistematis dan pemeriksaan menyeluruh terhadap berbagai pendekatan dan paradigma untuk melatih model NMT pada konteks sumber daya komputasi terbatas yang pertama. Tolok ukur ini, dilatih menggunakan sumber daya komputasi dan data pelatihan terbatas, mencapai performa yang kompetitif serta mampu melewati performa GPT-3.5-turbo yang telah di zero-shot untuk berbagai arah translasi dari bahasa Indonesia ke bahasa daerah yang low resource. Penelitian ini berkontribusi kepada kemajuan bidang NMT untuk bahasa-bahasa low resource di Indonesia dan membuka jalan untuk penelitian kedepannya sekaligus mengeksplorasi limitasi GPT-3.5-turbo dalam melakukan translasi bahasa daerah yang low resource. Akhirnya, penelitian ini menunjukkan bahwa melatih model XLM menggunakan data sintetis hasil code-switch memiliki performa translasi diatas pendekatan pelatihan penuh dan pelatihan model XLM dengan data monolingual saja. ......Neural machine translation (NMT) for low-resource local languages in Indonesia faces significant challenges, including the lack of a representative benchmark and limited data availability. This study addresses these challenges by establishing a replicable benchmark for four frequently spoken Indonesian local languages using limited computing resources on the FLORES-200 dataset. This study conduct the first systematic and thorough examination of various approaches and paradigms for NMT models in low-resource language settings. The benchmark, trained with limited computing power and training data, achieves competitive performance and surpass zero-shot GPT-3.5-turbo in multiple translation directions from Indonesian to low-resource local languages. This work contributes to the advancement of NMT for low-resource Indonesian languages and pave ways for future studies while exploring the limit of GPT-3.5-turbo in translating low-resource local languages. This study shows that training XLM models using synthetic data through code-switching increases translation performance of NMT models down the line compared to just training NMT models from scratch or training XLM models with only monolingual data.

Abstrak :
Pengembangan NLP di Indonesia terbilang lambat, terutama penelitian terkait bahasa daerah Indonesia. Alasannya adalah sumber data bahasa daerah tidak terdokumentasikan dengan baik sehingga sumber daya NLP yang ditemukan juga sedikit. Penelitian ini membahas metode ekstraksi kamus-kamus bahasa daerah di Indonesia untuk menghasilkan suatu sumber daya NLP yang dapat dibaca oleh mesin. Tahap penelitian dimulai dari pengumpulan data kamus, perancangan dan eksperimen metode ekstraksi, serta evaluasi hasil ekstraksi. Hasil penelitian berupa korpus paralel, leksikon bilingual, dan pasangan kata dasar-kata berimbuhan dalam format CSV dari beberapa kamus dwibahasa di Indonesia. Beberapa bahasa di antaranya adalah bahasa Minangkabau, Sunda, Mooi, Jambi, Bugis, Bali, dan Aceh. Perancangan metode ekstraksi berfokus pada kamus Minangkabau yang kemudian dilakukan eksperimen pada kamus-kamus bahasa daerah lainnya. Evaluasi dilakukan terhadap hasil ekstraksi kamus Minangkabau dengan melakukan anotasi data. Perhitungan akurasi dilakukan terhadap penempatan kelompok kata dari hasil anotasi. Hasil perhitungan menunjukkan 99% hasil ekstraksi sudah tepat untuk penentuan kelompok kata pada leksikon bilingual dan 88% untuk korpus paralel. Tim peneliti menemukan bahwa struktur dalam kamus bahasa daerah Indonesia sangat beragam, sehingga menuntut perlakuan yang berbeda pada setiap kamus, seperti perihal penomoran halaman. Selain itu, tim peneliti menemukan banyak kamus bahasa daerah Indonesia dengan kualitas yang kurang baik. Kualitas yang kurang baik ditunjukan dengan banyaknya kesalahan baca akibat noise yang terdapat pada tampilan berkas kamus. ......The development of NLP in Indonesia is relatively slow, especially for Indonesian local languages. Indonesian local language data sources are not well-documented so that there are only few NLP resources found. This study discusses the extraction method of Indonesian local language dictionaries to produce a machine-readable NLP resource. Starting from collecting dictionary data, designing and experimentation of the extraction method, and evaluating the extraction results. The extraction results are parallel corpus, bilingual lexicon, and words’ morphological form in CSV format from several Indonesian Local Language bilingual dictionaries that are Baso Minangkabau, Sundanese, Moi, Jambinese, Buginese, Balinese, and Acehnese. The designed method is also applied to some other local language dictionaries. Data annotation has been done to evaluate the extraction results so that we can calculate its accuracy of word classification for parallel corpus and bilingual lexicon. Extraction method design focuses on the Minangkabau dictionary which is then applied to other dictionaries. Data annotation has been done to evaluate the extraction results.The evaluation results show that 99% of the extraction results are correct for word classifying in the bilingual lexicon and 88% correct for parallel corpus. We found that the structure of dictionaries varies, so it requires different approaches for each dictionary, for example regarding page numbering. We also found many dictionaries with poor quality. The poor quality is indicated by the number of reading errors due to noise contained in the original dictionary file.

Abstrak :
Pengembangan NLP di Indonesia terbilang lambat, terutama penelitian terkait bahasa daerah Indonesia. Alasannya adalah sumber data bahasa daerah tidak terdokumentasikan dengan baik sehingga sumber daya NLP yang ditemukan juga sedikit. Penelitian ini membahas metode ekstraksi kamus-kamus bahasa daerah di Indonesia untuk menghasilkan suatu sumber daya NLP yang dapat dibaca oleh mesin. Tahap penelitian dimulai dari pengumpulan data kamus, perancangan dan eksperimen metode ekstraksi, serta evaluasi hasil ekstraksi. Hasil penelitian berupa korpus paralel, leksikon bilingual, dan pasangan kata dasar-kata berimbuhan dalam format CSV dari beberapa kamus dwibahasa di Indonesia. Beberapa bahasa di antaranya adalah bahasa Minangkabau, Sunda, Mooi, Jambi, Bugis, Bali, dan Aceh. Perancangan metode ekstraksi berfokus pada kamus Minangkabau yang kemudian dilakukan eksperimen pada kamus-kamus bahasa daerah lainnya. Evaluasi dilakukan terhadap hasil ekstraksi kamus Minangkabau dengan melakukan anotasi data. Perhitungan akurasi dilakukan terhadap penempatan kelompok kata dari hasil anotasi. Hasil perhitungan menunjukkan 99% hasil ekstraksi sudah tepat untuk penentuan kelompok kata pada leksikon bilingual dan 88% untuk korpus paralel. Tim peneliti menemukan bahwa struktur dalam kamus bahasa daerah Indonesia sangat beragam, sehingga menuntut perlakuan yang berbeda pada setiap kamus, seperti perihal penomoran halaman. Selain itu, tim peneliti menemukan banyak kamus bahasa daerah Indonesia dengan kualitas yang kurang baik. Kualitas yang kurang baik ditunjukan dengan banyaknya kesalahan baca akibat noise yang terdapat pada tampilan berkas kamus. ......The development of NLP in Indonesia is relatively slow, especially for Indonesian local languages. Indonesian local language data sources are not well-documented so that there are only few NLP resources found. This study discusses the extraction method of Indonesian local language dictionaries to produce a machine-readable NLP resource. Starting from collecting dictionary data, designing and experimentation of the extraction method, and evaluating the extraction results. The extraction results are parallel corpus, bilingual lexicon, and words’ morphological form in CSV format from several Indonesian Local Language bilingual dictionaries that are Baso Minangkabau, Sundanese, Moi, Jambinese, Buginese, Balinese, and Acehnese. The designed method is also applied to some other local language dictionaries. Data annotation has been done to evaluate the extraction results so that we can calculate its accuracy of word classification for parallel corpus and bilingual lexicon. Extraction method design focuses on the Minangkabau dictionary which is then applied to other dictionaries. Data annotation has been done to evaluate the extraction results.The evaluation results show that 99% of the extraction results are correct for word classifying in the bilingual lexicon and 88% correct for parallel corpus. We found that the structure of dictionaries varies, so it requires different approaches for each dictionary, for example regarding page numbering. We also found many dictionaries with poor quality. The poor quality is indicated by the number of reading errors due to noise contained in the original dictionary file.

Abstrak :
ABSTRAK
Place names play a vital role in human society. Names exist in all languages and place names are an indispensible part of international communication. This has been acknowledged by the establishment of the United Nations Group of Experts on Geographical Names (UNGEGN). One of UNGEGN?s tasks is to coordinate international efforts on the proper use of place names. Indonesia supports this effort and through its National Geospatial Agency (BIG). Place names are also of interest as an object of study in themselves. Academic studies into place names are found in linguistics, onomastics, philosophy and a number of other academic disciplines. This article looks at these two dimensions of place names, standardization efforts under the auspices of international and national bodies, and academic studies of names, with particular reference to the situation in Indonesia.