modify English task pages

yiqihuang · yiqihuang · commit 2083f6e25c5d · 2020-07-29T15:01:14.000-07:00
diff --git a/docs/co-reference_resolution.md b/docs/co-reference_resolution.md
@@ -46,8 +46,8 @@ Scoring code: https://github.com/conll/reference-coreference-scorers
 
 |  System | Average F1 of MUC, B-cubed, CEAF |
 | --- | --- |
-|  [Kong & Jian (2019)](https://www.ijcai.org/Proceedings/2019/700) | 63.85 |
 |  [Clark & Manning (2016b)](https://nlp.stanford.edu/static/pubs/clark2016deep.pdf) | 63.88 |
+|  [Kong & Jian (2019)](https://www.ijcai.org/Proceedings/2019/700) | 63.85 |
 |  [Clark & Manning (2016a)](https://nlp.stanford.edu/static/pubs/clark2016improving.pdf) | 63.66 |
 
 ### Resources
diff --git a/docs/entity_linking.md b/docs/entity_linking.md
@@ -57,7 +57,7 @@ NERC F-score
 | --- | --- | --- | --- |
 | [Sil et al (2018)](https://arxiv.org/abs/1712.01813) | 84.4 | | |
 | [Pan et al (2020)](https://www.aclweb.org/anthology/D19-6107.pdf) | 84.2 | | |
-| [Pan et al (2020)](https://www.aclweb.org/anthology/D19-6107.pdf) | 81.2 (unsup)| | |
+| [Pan et al (2020)](https://www.aclweb.org/anthology/D19-6107.pdf) | 81.2 (unsupervised)| | |
 | Best anonymous system in shared task writeup | 76.9 | 76.2 | 67.8 |
 
 ### Resources
diff --git a/docs/language_modeling.md b/docs/language_modeling.md
@@ -77,17 +77,17 @@ These numbers are not comparable, given different training conditions.
 |  [Huang et al, 2010 [GW v2]](http://www.imaging.org/site/PDFS/Reporter/Articles/2010_25/Rep25_2_EI2010_HUANG.pdf) | -- | 220.6 | 610m chars, random 11m for test. MSR segmenter. |
 |  Neural Lattice Models [v5] [Buckman+Neubig, 2018](https://www.mitpressjournals.org/doi/pdf/10.1162/tacl_a_00036) | 32.19 | -- | *Guangming Daily subset, top 10k chars + UNK, length <150. 934k lines train, 30k line test. Data [here](https://github.com/jbuckman/neural-lattice-language-models). |
 
-### Other Resources
+## Other Resources
 
-## <span class="t">Common Crawl Data</span>
+### <span class="t">Common Crawl Data</span>
 
 [CommonCrawl](https://commoncrawl.org) has released enormous quantities of web-crawled data that can be mined for Chinese text. Several groups have built their own pipelines to do the extraction and filtering.
 
 The CLUE Organization extracted "Clue Corpus 2020" (also called "C5") from the Common Crawl data. It is 100G raw text with 35 billion Chinese characters.
 Intended to be a large-scale corpus for pre-training Chinese language models.
 Preprint paper by [Xu, Zhang, and Dong](https://arxiv.org/abs/2003.01355v2)
 
-## <span class="t">CLUECorpusSmall </span>
+### <span class="t">CLUECorpusSmall </span>
 
 Publicly-available data, collected at https://github.com/CLUEbenchmark/CLUECorpus2020 and https://github.com/brightmart/nlp_chinese_corpus
 Includes:
diff --git a/docs/machine_translation.md b/docs/machine_translation.md
@@ -35,9 +35,9 @@ The United States and China may soon reach a trade agreement.
 * BLEU-SBP ((Chiang et al 08)[http://aclweb.org/anthology/D08-1064]).  Addresses decomposability problems with Bleu, proposing a cross between Bleu and word error rate.
 * HTER.  Returns the number of edits performed by a human posteditor to get an automatic translation into good shape.
 
-## <span class="t">ZH-EN</span>.
+## ZH-EN
 
-### <span class="t">WMT</span>.
+## <span class="t">WMT</span>.
 
 The Second Conference on Machine Translation (WMT17) has a Chinese/English MT component, done in cooperation with CWMT 2017.
 * [Website](http://www.statmt.org/wmt17)
@@ -91,7 +91,7 @@ The Linguistic Data Consortium has additional resources, such as FBIS and NIST t
 
 
 
-### <span class="t">NIST</span>.
+## <span class="t">NIST</span>.
 
 NIST has a long history of supporting Chinese-English translation by creating annual test sets and running annual NIST OpenMT evaluations during the 2000s.  Many sites have reported results on NIST test sets.  
 
@@ -134,13 +134,13 @@ The Linguistic Data Consortium provides training materials typically used for NI
 
 
 
-### <span class="t">IWSLT 2015</span>.
+## <span class="t">IWSLT 2015</span>.
 
 * Translation of TED talks
 * Chinese-to-English track
 * [Shared task overview](https://cris.fbk.eu/retrieve/handle/11582/303031/9811/main.pdf)
 
-|  Dataset | Size (sentences) | # of talks | Genre |
+|  Test sets | Size (sentences) | # of talks | Genre |
 | --- | --- | --- | --- |
 |  tst2014 | 1068 | 12 | TED talks |
 |  tst2015 | 1,080 | 12 | TED talks |
@@ -203,9 +203,9 @@ English to Chinese
 [The Multitarget TED Talks Task (MTTT)](http://cs.jhu.edu/~kevinduh/a/multitarget-tedtalks/)
 
 
-## <span class="t">ZH-JA</span>.
+## ZH-JA
 
-### <span class="t">Workshop on Asian Translation</span>.
+## <span class="t">Workshop on Asian Translation</span>.
 
 [The Workshop on Asian Translation](http://lotus.kuee.kyoto-u.ac.jp/WAT/) has run since 2014.  Here, we include the 2018 Chinese/Japanese evaluations.
 
@@ -255,7 +255,7 @@ Participants must get data from [here](http://lotus.kuee.kyoto-u.ac.jp/WAT/paten
 |  Japanese-Chinese devtest | 2000 | Patents |
 
 
-### <span class="t">IWSLT2020 ZH-JA Open Domain Translation</span>.
+## <span class="t">IWSLT2020 ZH-JA Open Domain Translation</span>.
 
 [The shared task](http://iwslt.org/doku.php?id=open_domain_translation) is to promote research on translation between Asian languages, exploitation of noisy parallel web corpora for MT and smart processing of data and provenance.
 
@@ -298,9 +298,9 @@ Japanese to Chinese
 |  Existing parallel sources | 1,963,238 | mixed-genre |
 
 
-## <span class="t">Others</span>.
+## Others
 
-### <span class="t">CWMT</span>.
+## <span class="t">CWMT</span>.
 
 [CWMT 2017](http://ee.dlut.edu.cn/CWMT2017/index_en.html)
 and [2018](http://www.cipsc.org.cn/cwmt/2018/english/)
diff --git a/docs/relation_extraction.md b/docs/relation_extraction.md
@@ -17,7 +17,7 @@ Output:
 
 ```
 (entity1: 李晓华, entity2: 王大牛, relation: 夫妻) 
-````
+```
 
 ## Standard Metrics
 
diff --git a/docs/spell_correction.md b/docs/spell_correction.md
@@ -72,7 +72,7 @@ Results above are all on the SIGHAN 2015 test set.
 
   | Source | # sentence pairs | # chars | # spelling errors | character set | genre |
   | --- | --- | --- | --- | --- | --- |
-  | Synthetic training dataset ([Wang et. al. 2018](https://www.aclweb.org/anthology/P19-1578)) 271,329 | 12M | 382,702 | simplified | news |
+  | Synthetic training dataset ([Wang et. al. 2018](https://www.aclweb.org/anthology/P19-1578)) | 271,329 | 12M | 382,702 | simplified | news |
 
 ---
 
diff --git a/docs/topic_classification.md b/docs/topic_classification.md
@@ -1,4 +1,4 @@
-# Chinese Text Classification / Topic Classification
+# Chinese Text Classification
 
 
 ## Background
@@ -113,9 +113,9 @@ First paragraphs of Chinese news articles from 2006-2016 were evenly split into
 
 |   | Accuracy |
 | --- | --- |
-| [[Meng et al, 2019]](https://arxiv.org/pdf/1901.10125.pdf) | 85.8% |
+| [Meng et al, 2019](https://arxiv.org/pdf/1901.10125.pdf) | 85.8% |
 | [Sun, Baohua, et al](https://arxiv.org/abs/1810.07653) | 84.4% |
-| [[Zhang and Lecun 2017]](https://arxiv.org/abs/1708.02657) | 83.7% |
+| [Zhang and Lecun 2017](https://arxiv.org/abs/1708.02657) | 83.7% |
 
 ### Resources
 
@@ -140,8 +140,8 @@ Chinese news articles from 2008- 2016 were evenly split into 7 news channels, re
 |   | Accuracy |
 | --- | --- |
 | [Sun, Baohua, et al](https://arxiv.org/abs/1810.07653) | 92.0% |
-| [[Meng et al, 2019]](https://arxiv.org/pdf/1901.10125.pdf) | 91.9% |
-| [[Zhang and Lecun 2017]](https://arxiv.org/abs/1708.02657) | 90.9% |
+| [Meng et al, 2019](https://arxiv.org/pdf/1901.10125.pdf) | 91.9% |
+| [Zhang and Lecun 2017](https://arxiv.org/abs/1708.02657) | 90.9% |
 
 ### Resources
 
