We provide two groups of example inputs and outputs in the data appendix:

 * the running example shown in Figure 1:
   - the input sentence is 国际苦名丰持人
   - the golden output sentence is 国际著名主持人  (苦 --> 著, 丰 --> 主)
   - the example is spell checked by FASPell trained on OCR data

 * the examples of handling inconsistencies between SC and TC usages
   1. 周末(weekend):
      SC:
         - the input sentences are 这周末的考试很难。and 你考虑得相当周全啊。
         - the golden output sentences are 这周末的考试很难。 (nothing changed) and 你考虑得相当周全啊。 (nothing changed)
         - the example is spell checked by FASPell trained on OCR data
      TC:
         - the input sentences are 這周末的考試很難。and 你考慮得相當周全啊。
         - the golden output sentences are 這週末的考試很難。 (周 --> 週) and 你考慮得相當周全啊。 (nothing changed)
         - the example is spell checked by FASPell trained on SIGHAN15 data

   2. 旅游(trip):
      SC:
         - the input sentences are 我从台湾坐飞机到美国旅游。and 你会游泳吗。
         - the golden output sentences are 我从台湾坐飞机到美国旅游。 (nothing changed) and 你会游泳吗。 (nothing changed)
         - the example is spell checked by FASPell trained on OCR data
      TC:
         - the input sentences are 我從台灣坐飛機到美國旅游。and 你會游泳嗎。
         - the golden output sentences are 我從台灣坐飛機到美國旅遊。 (游 --> 遊) and 你會游泳嗎。 (nothing changed)
         - the example is spell checked by FASPell trained on SIGHAN15 data

   3. 制造(make):
      SC:
         - the input sentences are 这件衣服是中国制造。and 公司的这个制度我们无法接受。
         - the golden output sentences are 这件衣服是中国制造。 (nothing changed) and 公司的这个制度我们无法接受。 (nothing changed)
         - the example is spell checked by FASPell trained on OCR data
      TC:
         - the input sentences are 這件衣服是中國制造。and 公司的這個制度我們無法接受。
         - the golden output sentences are 這這件衣服是中國製造。 (制 --> 製) and 公司的這個制度我們無法接受。 (nothing changed)
         - the example is spell checked by FASPell trained on SIGHAN15 data



In the directory of each example, there are three types of files:
 * nohup.out -> this is the verbose log of the whole spell checking process; the details of filtering each candidate are fully logged;
 * history.json -> this records the the input sentences and output sentences in each round;
 * result_*.json -> each of these provides details of changes of sentences in each round;

