Hiroto Nakajima and Takeshi Yamada from NIKKEI Innovation Lab developed techniques teaching AI skill of article proofreading effectively, and presented it as an academic thesis at annual meeting of the Association for Natural Language Processing 2018, held on March 15, 2018.

It has characteristic deep learning, which was fundamental technology of AI, applied for proofreading. This is first thesis being presented by Nikkei Innovation Lab alone. Title of the thesis is “ Innovation of techniques identifying error (proofreading support) device”. Although automatic translation of Applied research advanced in deep learning of language, allied research for proofreading has started recently.

No one mentioned there is such a research found in previously along selection process, this study may effect on AI research worldwide.

The system works as follows.
In order to make incorrect sentence automatically, the part of “correct” sentence taken from newspaper article being changed in order following certain rule and replaced intendedly as false noise.

This makes “teacher data” in deep learning.

Secondly, inserting both correct sentences and incorrect sentences teaches AI to
After this process, putting into sentence for proofreading, evaluation value comes out automatically. Prototype was made for validation, which detects errors with approx. 95% success rate that is good level for practical use.

Teaching large volume of articles of various genres such as politics, economics, marketing can lead to developing system to spot errors reflects features of each category. It can be applied to multi languages as well as Japanese and English.

Nikkei Innovation Lab will continue working on research and development hoping for other purposes including finding bugs in programming or proofreading other articles as well as news article.