--- title: "AI Research in China: A Review" authors: - "Jiang Xinsong" - "Song Guoning" - "Chen Yu" doi: "" url: "https://ijcai.org/proceedings/1983/37" pdf: "https://ijcai.org/proceedings/1983/0037.pdf" venue: "Proceedings of the Eighth International Joint Conference on Artificial Intelligence" year: 1983 pages: "164-165" track: "Expert Systems" keywords: --- # AI Research in China: A Review Jiang Xinsong, Song Guoning, Chen Yu *Proceedings of the Eighth International Joint Conference on Artificial Intelligence*, pages 164-165. ## Abstract Since 1970's AI research has been getting more and more active in China and certain useful results have been achieved. This paper is intended to review briefly what waa and is g oin g on in AI field of China. ## Full text ## ABSTRACT Since 1970's AI research has been getting more and more active in China and certain useful results have been achieved. This paper is intended to review briefly what waa and is going on in AI field of China. ## INTRODUCTION In China AI research activities began ever from early 1960 1 s, but it had been cofined only to a small circle until the end of that decade. From then on many university departments majoring in com­ puter science were set up. They were try­ ing hard to expand applications of com­ puter and develop theories, expecting that computer would bring about improve­ ment in productivity process control. Considerable efforts have been made in this direction, but the results were not so satisfactory since a practical process is too complex to be identified. However the fact that a veteran worker or a tech~ nician often deals sith the control task some hat better than a computer. The fact suggested that AI might be a succesful approach. On the other hand, to tackle some open fascinating problems linguists and psychologists were also drawn into AI. And now AI research has been getting more and more active in China and certain use­ ful results have been achieved. This paper is intended to give a brief review on the following topics: 1. machine translation and natural language understanding, 2. theorem proving, 3. expert consulting system, 4. computer aided education, 5. others. MACHINE TRANSLATION AND NATURAL LANGUAGE UNDERSTANDING Machine translation is the earliest AI research work in China and experinced its ups and downs. We may list the follow­ ing three systems, the first one was developed by the Science Technology Infor­ mation Institute and the other two by the ## AI RESEARCH IN CHINA: A REVIEW Jiang Xinsong Song Guoning, Chen Yu Shenyang Institute of Automation Academia Sinica, Shenyang, China Computer Technology Institute Academia Sinica. · 1. ECTA (English-Chinese Automation Translation) system 0) , which processed correctly a science-technologe text con· sisting of 1600 words · (25 words/sentence in avera~ e, 60 words/sentence of the longest) with a direct chinese character output in J ·une, 1981. 'Phe.:-eecaracy was 90.8% when it dealt with a 3600-word article with most errors to be corrected by slightly modifying dictionary or program. The s ystem w9rks tx f ollowing procedures: Sentence segmetation -- to segment a sentence according to prpositions, con­ junctions, articles or punctuation marks. Template match -- to match each seg­ ment with its template and label with feature data. Segment tree formation -- to link segment with one other and result iri a tree structure. Tree scan-to turn out the goal lan­ guage sentence. - 2 . English-Chinese Title Translation System (~ , which could process no leas than 5000 metallurgical items with an. accuracy of 80% and a speed of 100 items / 60min. For the errors occurixg<6o% was of lexical nature and easy to eliminate. and the remaining 40% was grammatical. 3. JF-111, a universal machine trans­ lation system pJ , which could theoreti­ cally process anj language, if grammars of the two languages to be translated into each other were stored in the system. Research on natural language under­ stand~ng is mainly centred on chinese. The system RJD-80 (4), developed in the Institute of Linguistics, Chinese Academy .of Social Sciences, is a typical one. It is a QA system run on a small knowledge base of chinese literature. In the following are a few typical se ntences between man and the system· - Q (man): nt~fi'~1f.iA:!t~?" "Is the Author of the drama 'TEAHOUSE' LAOSHE?" - A (machine): 11 ~11~ " "Yes." Q: " e"i.?;;tJl~.::ot--~\_~ ?" "Has BAJIN ever written any drama ?II A: " . 7l1.J." "No." This system is a linguistic model of understanding modern chinese on basis :of linguistic theory. Another sucessful project concerning natural language is the interface for the Rice Resource Retrieval System, which was the system in English eliminating the need for users to memorize long boring string of commands. The system extract~ key words and supplemented with syntactic and semantic analysis. At first, sum up several typical questions in retrieval procedure, for example "Find out the value V, of attribure A1 corresponding to the value V1 of attribute A2 11 , i.e., Q(A1 A2 V2 ). ATN network method is employed in parsing (syntactic analysis) and each node represents a key word. Zhejiang University have also deve­ loped another similar system, a language interface for Warehouse Management System. One can inquire the system in any language if the system has been trained beforehand. To understand chinese language by machine, some institutions have set out to explore certain specific methods and fundamental corresponding theory. For instance, Qinghua University proposed a 4-level hierarchy structure: sentence-­ component--structure--word. ## THEOREM PROVING Prof. Wu Wenjun, Institute of Sys­ tem Science, Academia Sinica, has esta­ blished a novel approach on the decision problem and the mechanization of theorem­ proving in elementary geometry. Its basic idea, restricted to theorems with betwe­ enness but of consideration and based on an entirely different principle, aims at giving a mechanical procedure which per­ mits to prove quite non-trivial theorems in elementaty geometry even by hands. His algorithm has solved certain practical problems that have been hard nuts to crach. Prof. Wang Xianghu, Jilin University, has proposed a generalized Resolution Principle. its basic idea is as follows: In theore~ provin~ the axiom often takes implication form (P-q, R-s, ··· ). While turn it into Skolem normal form they will take disjunction form( P V q, R V S, ··· ) and lose the explicit meaning involved in primitive form. In order to make deduction more effective and natural he suggested to keep the natural form of Implication. The clause in natural form is named gerenalia­ ed clause, corresponding resolution is described as Generalized Resolotion. He proved that generalized resolution is com­ pl~te for first order predicate logic and many improved algorithm, developed in recent years, can be applied to general­ ized resolution, (e. g., lock resolution, linear resolution, semantic resolution etc.). ## EXPERT CONSULTING SYSTEM By the end of 1970's there were only certain small prototype Erpert Systems. After a few year groping and experiences accumulating R & D work on some large practical systems had begun. Most of these system. This is resulting from the fact that traditional chinese medicine diagno­ sis is mainly by experinece, and so to simulate doctors' or physician's diagnosis procedures has aroused a great interests among AI researchs. The first system of this kind is for hepatitis, developed jointly by Beijing Academy of Traditional Chinese Medicine and the Institute of Au­ tomatiion, Academia Sinica. The fuzzy set theory was applied in the system to make decisions in process of diagnosis. And the program is in BASIC. This system has been successfully employed in practice, diagno­ sing dozens patiants a day. Anther system of this sort was deve­ lopred by Shanghai Institute of Computing Technology, named General Diagnosis System of Traditional Chinese Medicine. The main idea of the system is the concept of "con­ sciousness package". The program are divi­ ded into several small pieces named con­ sciousness package. Once a certain condi­ tion has been met the corresponding con­ sciousness package would become active. And the package can also be activited by other package. This is a simulation of association thinking of human being. A consciousness package is a basic unit of knowledge. For example, a prescription of chinese herbal medicine may be described as a package in which its natures, what kind of disease to be cure, contraindica­ tion etc. are involoved. Based on similar idea the Institute has also developed a control system of blast furnace recertlJ. As chemical reac­ tion in blast furnace is so complicated that it could not get desired .result with t:radit:L.onal. identffication.The system adopted a v-e.ter.an s stra:tegy .. :to control some key parameters. And they have achieved same operating level as a group chief does with this system. There are still certain others of medical diagnosis system, such as for gynaecology diseases developed in Jilin University. It diagnosed 204 patie~ nts with a correctness rate of 91.7%. The system is programed in BASIC, has a 17K main storage. The system for heart dis­ eases, developed in Zhongshan University, it achieved 90% correct diagnosis rate. The system for nephitis, developed by Nanjing Technology College, it provided 278 different prescr i ptions from 98 s ymptoms with a s peed of 1 diagnosis/mi­ nute. There are also system for lung dis­ ease, paediatrics diseases and other dis­ eases. Some of these are still prototype system while some others have been put into practice. Most of them are simulating a doctor and-supplemented by some math processing . Some of them are MYCIN--like system. A group in Zhejiang University tried to ~ tack le certain technical problems in agriculture and sideline production with Expert System, crossbreeding of silkworms is an example. Conventional crossbreeding is to select parents to generate a hybrid and then with two new generations hybrid­ ized again to bear sencond generation of hybrid and so on. After several times of hybridization a nice variety would be yielded. Following such a way there are two major drawbacks: long period and low rate of success. Computer aided cross­ breeding system, in which certain experi ences of veteran technicians have been embodied, not only selects the parents based on knowledge rules but also de­ scribes properties of new generation beforehand. So that one could made a com­ par1s1on choose the best scheme. This system adopts MYCIN BACKWARD ~aasoning . Zhejiang University also developed ~ystems for wheat crossbreeding and dis­ igning patterns printed on cloth. In the last system a set of formal language were adopted to describe pattern structure rules. ## COMPUTER AIDED EDUCATION Jinlin University have developed a Factoring System, one can factorize pol­ ynome with positive integers of coeffi­ cients and exponentials with this system. It provided 16 standard formulas which sould be solved directly: (a~ -b· ), (a 5 .. 8 5 ) (a 4 +4b .. ), (a· .±.2ab+b·), ···· If a polynome to be solved is not a standard formula as metioned, the system factorizes them, till starua&r· for~laa. · Tae s7Btem 4.& eapa.hl.e of ahowing the whole procedure in detail, helps students understand the whole course, 90% questions in junior middle school exercise in math could be solved with this system and the way i s rather the same as human being does. ATEP System, capable of solving trigonometric equations, has solved 110 out of 138 of trigonometry questions appeared in textbook of senior middle school. SSH system simu­ lates a student to solve various applied arithmetic problems. ## OTHER FIELD IN AI 1. Program Synthisis. Bei jing Aeron­ autical Institute has worked out a sys tem for synthesizing recursive program on the base of structure induction approach. Some simple functions have been synthesized with this system, such as SORT (L) REVERSE ( L) · 2. Robotics. Shenyang Institute of Automation, Academia Sinica, has developed a playback-type robot. It has 5 freedoms: arm stretch out, draw back, pitching move­ ment and turn round; wrist turn round an d swing ; it has been implemented computer control in two ways: point to point and linear track and speed. In point-yo-poin~ control the playback accuracy is less than + 1.5 mm, in linear-track control retrack­ Tng accuracy not more than 10 mm, the maximum weight grasped by the hand is about 35kg. 3. Modal Logic about "knowing". Prof. Ma Xiwen, Beijing University, established a predicate calculus of modal logic about "knowing", including its formal system W; provided its semantic i:llt-erpretation JS; discuaaed certain principal features of system of W-KS; by some typical examples, especially with wellknown puzzle "Mr. S and Mr. P". 4. LISP-130 Interpreter, developed i n Shenyang Institute of Automation, Academia Sinica, was on a domestic minicomputer. It is based on LISP 1.5 having 94 functions and programed in Assembly. This is the first implementation in China (1980). ·. LISP-3000 was developed by Beijing Univer­ sity in 1982 written in WISP language. One of its peculiarities is the ability to transplant to any co~puter. They also develop a language called FT (From Trans­ lation), by which WISP could be translated into FASP, a specific assembler language.