File size: 13,468 Bytes
a05fb91 698c91d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 |
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Vietnamese NLP Tasks – Benchmark Overview</title>
<meta name="viewport" content="width=device-width,initial-scale=1">
<style>
body { font-family: 'Segoe UI', Arial, sans-serif; background: #f7faff; color: #263347; margin: 0; }
.container { max-width: 1000px; margin: 36px auto; background: #fff; padding: 28px 20px 38px 20px; border-radius: 14px; box-shadow: 0 2px 14px #0002;}
h1, h2, h3, h4 { margin-top: 1.7em; margin-bottom: 0.5em; }
h1 { color: #12469b; font-size: 2.2em; margin-top: 0; }
h2 { color: #188754; border-left: 6px solid #a3d7ff; padding-left: 12px;}
h3, h4 { color: #1a324b;}
table { width: 100%; border-collapse: collapse; margin: 1.1em 0 1.8em 0;}
th, td { padding: 8px 10px; border-bottom: 1px solid #eee;}
th { background: #eaf3ff; }
tr:hover { background: #f6fbff;}
a { color: #2369cb; text-decoration: none;}
a:hover { text-decoration: underline;}
ul, ol { margin-top: 0.5em; margin-bottom: 1.1em; }
.dataset { background: #f1f8fc; border-left: 5px solid #97d3f6; padding: 8px 18px; margin: 12px 0 18px 0;}
.note { color: #555; background: #f6fcf7; border-left: 5px solid #7de59d; padding: 6px 16px; margin: 14px 0 20px 0;}
.icon { font-size: 1.1em; margin-right: 6px;}
.footer { text-align: center; font-size: 0.96em; color: #999; margin-top: 36px; }
@media (max-width: 700px) {
.container { padding: 6px; }
table, th, td { font-size: 14px;}
}
</style>
</head>
<body>
<div class="container">
<h1>🇻🇳 Vietnamese NLP Tasks <span style="font-size:0.8em; color:#555;">— Benchmark & SOTA Overview</span></h1>
<div style="margin-bottom:1.2em; color:#537fc2;">
<span class="icon">📈</span>
<b>This page tracks major Vietnamese NLP datasets and models for <u>Dependency Parsing</u>, <u>Intent Detection</u>, <u>Machine Translation</u>, <u>NER</u>, <u>POS Tagging</u>, <u>Semantic Parsing</u>, and <u>Word Segmentation</u>.</b>
</div>
<!-- DEPENDENCY PARSING -->
<h2>Dependency Parsing</h2>
<div class="dataset">
<span class="icon">🗂️</span>
<b>VnDT v1.1/v1.0</b>: Benchmark treebank >10K sentences. <br>
<b>Test:</b> 1,020 (v1.1), Dev: 200, Rest: Train.
</div>
<h3>VnDT v1.1</h3>
<table>
<tr>
<th>Model</th>
<th>LAS</th>
<th>UAS</th>
<th>Paper</th>
<th>Code</th>
</tr>
<tr>
<td>PhoNLP (2021)</td><td>79.11</td><td>85.47</td>
<td><a href="https://aclanthology.org/2021.naacl-demos.1.pdf">PhoNLP</a></td>
<td><a href="https://github.com/VinAIResearch/PhoNLP">Official</a></td>
</tr>
<tr>
<td>PhoBERT-base (2020)</td><td>78.77</td><td>85.22</td>
<td><a href="https://arxiv.org/abs/2003.00744">PhoBERT</a></td>
<td><a href="https://github.com/VinAIResearch/PhoBERT">Official</a></td>
</tr>
<tr>
<td>Biaffine (2017)</td><td>74.99</td><td>81.19</td>
<td><a href="https://arxiv.org/abs/1611.01734">Biaffine Parsing</a></td>
<td></td>
</tr>
<tr>
<td>VnCoreNLP (2018)</td><td>71.38</td><td>77.35</td>
<td><a href="http://aclweb.org/anthology/N18-5012">VnCoreNLP</a></td>
<td><a href="https://github.com/vncorenlp/VnCoreNLP">Official</a></td>
</tr>
</table>
<h3>VnDT v1.0 (Gold POS)</h3>
<table>
<tr>
<th>Model</th>
<th>LAS</th>
<th>UAS</th>
<th>Paper</th>
<th>Code</th>
</tr>
<tr>
<td>VnCoreNLP (2018)</td><td>73.39</td><td>79.02</td>
<td><a href="http://aclweb.org/anthology/N18-5012">VnCoreNLP</a></td>
<td><a href="https://github.com/vncorenlp/VnCoreNLP">Official</a></td>
</tr>
<tr>
<td>BIST BiLSTM graph (2016)</td><td>73.17</td><td>79.39</td>
<td><a href="https://aclweb.org/anthology/Q16-1023">BIST Parser</a></td>
<td><a href="https://github.com/elikip/bist-parser/tree/master/bmstparser/src">Official</a></td>
</tr>
<tr>
<td>MSTparser (2006)</td><td>70.29</td><td>76.47</td>
<td><a href="http://www.aclweb.org/anthology/P05-1012">MSTparser</a></td>
<td></td>
</tr>
</table>
<!-- INTENT DETECTION -->
<h2>Intent Detection & Slot Filling</h2>
<div class="dataset">
<span class="icon">🛫</span>
<b>PhoATIS Dataset</b> (flight booking domain): Train: 4,478, Dev: 500, Test: 893
</div>
<table>
<tr>
<th>Model</th><th>Intent Acc.</th><th>Slot F1</th><th>Sent. Acc.</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>JointIDSF (2021)</td><td>97.62</td><td>94.98</td><td>86.25</td>
<td><a href="https://arxiv.org/abs/2104.02021">JointIDSF</a></td>
<td><a href="https://github.com/VinAIResearch/JointIDSF">Official</a></td>
</tr>
<tr>
<td>JointBERT+PhoBERT</td><td>97.40</td><td>94.75</td><td>85.55</td>
<td><a href="https://arxiv.org/abs/2104.02021">JointIDSF</a></td>
<td><a href="https://github.com/VinAIResearch/JointIDSF">Official</a></td>
</tr>
</table>
<!-- MACHINE TRANSLATION -->
<h2>Machine Translation</h2>
<div class="dataset">
<span class="icon">🌐</span>
<b>PhoMT Dataset</b>: 3.02M sentence pairs | 6 domains (TED, WikiHow, MediaWiki, OpenSubtitles, News, Blog)
</div>
<table>
<tr>
<th>Model</th><th>EN→VI (BLEU)</th><th>VI→EN (BLEU)</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>mBART (2020)</td><td>43.46</td><td>39.78</td>
<td><a href="https://arxiv.org/abs/2001.08210">mBART</a></td>
<td><a href="https://github.com/pytorch/fairseq/tree/main/examples/mbart">Link</a></td>
</tr>
<tr>
<td>Transformer-big</td><td>42.94</td><td>37.83</td>
<td><a href="https://arxiv.org/abs/1706.03762">Transformer</a></td>
<td><a href="https://github.com/pytorch/fairseq/tree/main/examples/translation">Link</a></td>
</tr>
</table>
<div class="dataset">
<span class="icon">📋</span>
<b>IWSLT2015</b>: 150K sentence pairs (EN↔VI) | <a href="https://github.com/tensorflow/nmt">Data & Scripts</a>
</div>
<table>
<tr>
<th>Model</th><th>BLEU</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>Nguyen & Salazar (2019)</td><td>32.8</td>
<td><a href="https://arxiv.org/abs/1910.05895">Transformers w/o Tears</a></td>
<td><a href="https://github.com/tnq177/transformers_without_tears">Official</a></td>
</tr>
<tr>
<td>Provilkov et al. (2019)</td><td>33.27 (uncased)</td>
<td><a href="https://arxiv.org/abs/1910.13267">BPE-Dropout</a></td>
<td></td>
</tr>
<tr>
<td>Xu et al. (2019)</td><td>31.4</td>
<td><a href="https://papers.nips.cc/paper/8689-understanding-and-improving-layer-normalization.pdf">Layer Norm</a></td>
<td><a href="https://github.com/lancopku/AdaNorm">Official</a></td>
</tr>
<tr>
<td>Transformer (2017)</td><td>28.9</td>
<td><a href="http://papers.nips.cc/paper/7181-attention-is-all-you-need">Transformer</a></td>
<td><a href="https://github.com/duyvuleo/Transformer-DyNet">Link</a></td>
</tr>
</table>
<!-- NER -->
<h2>Named Entity Recognition (NER)</h2>
<div class="dataset">
<span class="icon">🩺</span>
<b>PhoNER_COVID19</b>: 10 types, 34,984 entities, 10,027 sentences
</div>
<table>
<tr>
<th>Model</th><th>F1</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>PhoBERT-large</td><td>94.5</td>
<td><a href="https://arxiv.org/abs/2003.00744">PhoBERT</a></td>
<td><a href="https://github.com/VinAIResearch/PhoBERT">Official</a></td>
</tr>
<tr>
<td>XLM-R-large</td><td>93.8</td>
<td><a href="https://aclanthology.org/2020.acl-main.747/">XLM-R</a></td>
<td><a href="https://github.com/facebookresearch/XLM">Official</a></td>
</tr>
<tr>
<td>BiLSTM-CRF + CNN-char</td><td>91.0</td>
<td><a href="http://www.aclweb.org/anthology/P16-1101">BiLSTM-CRF</a></td>
<td><a href="https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/">Link</a></td>
</tr>
</table>
<div class="dataset">
<span class="icon">📄</span>
<b>VLSP 2016 NER</b>: 16,861 train/dev, 2,831 test sentences.
</div>
<table>
<tr>
<th>Model</th><th>F1</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>PhoBERT-large</td><td>94.7</td>
<td><a href="https://arxiv.org/abs/2003.00744">PhoBERT</a></td>
<td><a href="https://github.com/VinAIResearch/PhoBERT">Official</a></td>
</tr>
<tr>
<td>PhoNLP</td><td>94.41</td>
<td><a href="https://aclanthology.org/2021.naacl-demos.1.pdf">PhoNLP</a></td>
<td><a href="https://github.com/VinAIResearch/PhoNLP">Official</a></td>
</tr>
<tr>
<td>vELECTRA</td><td>94.07</td>
<td><a href="https://arxiv.org/abs/2006.15994">vELECTRA</a></td>
<td><a href="https://github.com/fpt-corp/viBERT">Official</a></td>
</tr>
<tr>
<td>VnCoreNLP</td><td>91.30</td>
<td><a href="http://aclweb.org/anthology/N18-5012">VnCoreNLP</a></td>
<td><a href="https://github.com/vncorenlp/VnCoreNLP">Official</a></td>
</tr>
</table>
<!-- PART OF SPEECH -->
<h2>Part-of-Speech Tagging</h2>
<div class="dataset">
<span class="icon">🔤</span>
<b>VLSP 2013</b>: 27,870 train/dev, 2,120 test
</div>
<table>
<tr>
<th>Model</th><th>Accuracy</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>PhoBERT-large</td><td>96.8</td>
<td><a href="https://arxiv.org/abs/2003.00744">PhoBERT</a></td>
<td><a href="https://github.com/VinAIResearch/PhoBERT">Official</a></td>
</tr>
<tr>
<td>vELECTRA</td><td>96.77</td>
<td><a href="https://arxiv.org/abs/2006.15994">vELECTRA</a></td>
<td><a href="https://github.com/fpt-corp/viBERT">Official</a></td>
</tr>
<tr>
<td>PhoNLP</td><td>96.76</td>
<td><a href="https://aclanthology.org/2021.naacl-demos.1.pdf">PhoNLP</a></td>
<td><a href="https://github.com/VinAIResearch/PhoNLP">Official</a></td>
</tr>
<tr>
<td>PhoBERT-base</td><td>96.7</td>
<td><a href="https://arxiv.org/abs/2003.00744">PhoBERT</a></td>
<td><a href="https://github.com/VinAIResearch/PhoBERT">Official</a></td>
</tr>
<tr>
<td>VnCoreNLP-VnMarMoT</td><td>95.88</td>
<td><a href="http://aclweb.org/anthology/U17-1013">VnMarMoT</a></td>
<td><a href="https://github.com/datquocnguyen/vnmarmot">Official</a></td>
</tr>
<tr>
<td>BiLSTM-CRF + CNN-char</td><td>95.40</td>
<td><a href="http://www.aclweb.org/anthology/P16-1101">BiLSTM-CRF</a></td>
<td><a href="https://github.com/XuezheMax/LasagneNLP">Official</a></td>
</tr>
<tr>
<td>RDRPOSTagger</td><td>95.11</td>
<td><a href="http://www.aclweb.org/anthology/E14-2005">RDRPOSTagger</a></td>
<td><a href="https://github.com/datquocnguyen/rdrpostagger">Official</a></td>
</tr>
</table>
<!-- SEMANTIC PARSING -->
<h2>Semantic Parsing</h2>
<div class="dataset">
<span class="icon">🗃️</span>
<b>ViText2SQL</b>: 10K question/SQL pairs, the first public Text-to-SQL dataset for Vietnamese.
</div>
<table>
<tr>
<th>Model</th><th>Exact Match Acc.</th><th>Paper</th><th>Code</th><th>Note</th>
</tr>
<tr>
<td>IRNet (2019)</td><td>53.2</td>
<td><a href="https://aclanthology.org/2020.findings-emnlp.364/">ViText2SQL</a></td>
<td><a href="https://github.com/microsoft/IRNet">Link</a></td>
<td>Using PhoBERT encoder</td>
</tr>
<tr>
<td>EditSQL (2019)</td><td>52.6</td>
<td><a href="https://aclanthology.org/2020.findings-emnlp.364/">ViText2SQL</a></td>
<td><a href="https://github.com/ryanzhumich/editsql">Link</a></td>
<td>Using PhoBERT encoder</td>
</tr>
</table>
<!-- WORD SEGMENTATION -->
<h2>Word Segmentation</h2>
<div class="dataset">
<span class="icon">✂️</span>
<b>VLSP 2013</b>: 75k train, 2,120 test sentences (manually word-segmented)
</div>
<table>
<tr>
<th>Model</th><th>F1</th><th>Paper</th><th>Code</th>
</tr>
<tr>
<td>UITws-v1 (2019)</td><td>98.06</td>
<td><a href="https://arxiv.org/abs/2006.07804">UITws-v1</a></td>
<td><a href="https://github.com/ngannlt/UITws-v1">Official</a></td>
</tr>
<tr>
<td>VnCoreNLP-RDRsegmenter (2018)</td><td>97.90</td>
<td><a href="http://www.lrec-conf.org/proceedings/lrec2018/pdf/55.pdf">VnCoreNLP</a></td>
<td><a href="https://github.com/datquocnguyen/RDRsegmenter">Official</a></td>
</tr>
<tr>
<td>UETsegmenter (2016)</td><td>97.87</td>
<td><a href="http://doi.org/10.1109/RIVF.2016.7800279">UETsegmenter</a></td>
<td><a href="https://github.com/phongnt570/UETsegmenter">Official</a></td>
</tr>
<tr>
<td>vnTokenizer (2008)</td><td>97.33</td>
<td><a href="https://link.springer.com/chapter/10.1007/978-3-540-88282-4_23">vnTokenizer</a></td>
<td></td>
</tr>
<tr>
<td>JVnSegmenter (2006)</td><td>97.06</td>
<td><a href="http://www.aclweb.org/anthology/Y06-1028">JVnSegmenter</a></td>
<td></td>
</tr>
</table>
<div class="footer">
NLP Progress – Benchmarks collected by the open-source community.<br>
<span style="color:#ccc;">Style inspired by <a href="https://github.com/sebastianruder/NLP-progress" target="_blank">NLP-progress</a></span>
</div>
</div>
</body>
</html>
|