Yesterday, Dr. Zhou Ming from the natural language processing group of Microsoft Yayan Institute gave us a report.

Dr. Zhou's report is very interesting. It tells us a lot of "tricks" for natural language processing, one by one, and says, "This is enough, there is more land ", in addition, Dr. Zhou's northeast talk is particularly ridiculous, and the time passes without knowing it. I was impressed by my memory.

Zhou Ming proposed four levels of NLP

(1) Machine Translation (MT), an automatic Q & A system. The main idea of these early AI scientists is completely unnecessary.Personnel participationZhou Ming gave us some indicators, with xx reaching 89% and XX reaching 90%. It seems that some of them are part-of-speech tagging, some of them are syntactic analysis. These are almost bottlenecks, but they are not doing well.

(2) search engines, social searches, couplets, and so on. These need to be imported into the user's intelligence. log analysis has a huge impact on the establishment of the model, therefore, this level of work requires immediate use by users, comments from various users, and group intelligence to improve the effect.

If the above two types of research are involved, "it is hard work", which is basically hated by various cool people.

(3) Use social networks to do some work, mention social QA, and mention Facebook's I Like It function (as mentioned in the previous report by people on Renren ), as mentioned, Facebook will use social networks to Optimize search engines and incorporate search into one of its components. This is a great dream.

(4) blank spaces may change the way humans interact, new Internet forms, and new ways of communication.

The above two categories are the focus of scientific research and investment. We hope that we can make more efforts at these two levels.

 

Interestingly, Zhou Ming asked us if we wanted to start a business. As a result, no one in the teacher raised his hand. The teacher said that our Tsinghua team was very low-profile and said that when I went to Peking University to report, half of them raised their hands and had entrepreneurial ideas, it seems that there is indeed a huge ideological gap between different schools. Tsinghua's people are well-off.

 

Finally, let me talk about my ideas. The information is largely understandable, valuable, and influential, therefore, the processing of natural language is more of an automatic processing of information. This processing process can make the information more storage, presentation, and availability, zhou Ming gave an example of engkoo made by Microsoft. She is a adj. lady. this query is intended to know what is the adjective that describes lady. To do this function, you can simply think about how to process and align the syntax structure, however, without good information processing, it is impossible to support such queries. SoFirst, we should understand the application and push back what needs to be done from the application of information. Otherwise, there will be no human use in intelligence, and there will be no chance of improvement. It can only be a matter of high or low.