Predication of Meaning of Bisyllabic Chinese Compound Words Using Back Propagation Neural Network

W Owa Predication of Meaning of Bisyllabic Chinese Compound Words Using Back Propagation Neural Network Lua Kim Teng Department Of Information Systems and Computer Science, National University of Singapore Kent Ridge, Singapore 119260 luakt@iscs.nus.sg http://www.iscs.nus.sg/–luakt Abstract A three layer back propagation neural net is set up to study the functional dependency between the semantic class of a bisyllabic Chinese word and that of its two constituent Chinese characters. Simulations were performed using a three-layer back-propagation neural net with various combination of inputs. The inputs are (1) semantic classes of the constituent characters, (2) Entropy of the characters and (3) semantic strength[1] of the characters. Our simulations show that we can obtain the meaning class of a bisyllabic word from the meaning classes of its two constituent characters to an accuracy of 81 % by taking the semantic classes and semantic strength of the characters as input. This research establishes the dependency between the meaning class of a Chinese compound word and that of its two constituent characters. 1. Introduction The processing of the meaning of a word is a difficult job. In all existing text books on semantics[2-4], only descriptive treatments are provided for explaning the meaning of a word. There have been some attempts to break meaning into more fundamental units (Chapter 6 of [2]) but without much success. The main obstacle is that we are still unable to quantify meaning. As there is no proper representation of meaning, it is also difficult to process meaning. In some of our earlier papers [5-7], we have proposed a scheme to quantify the change in the meaning when a bisyllabic Chinese compound word is built from two Chinese characters. We take the meaning of a word as one of the meaning class given in a dictionary Figure 1 Semantic Distance of meaning classes:� riq X it-33 ifa� [8]. This dictionary classifies 70,000 Chinese words into� Characters Between Words and Ch 49 12 major, 94 medium and 1428 minor classes. A triangular figure is used to represent the change in meaning during a word formation process(See Figure 1). In Figure 1, W is the word, Ca, Cb are the characters forming the word, Owa, 6wb , Oab are the so-called semantic distances between the word and the characters and between the two characters. The value of 6 is computed according to Table 1. Table : semantic distance 6 Type of distance 6 same major, medium and minor class 0 (Hex 000) same major class, medium class, different minor class 1 (Hex 001) same major class, different medium class 2(Hex 010) 4(Hex(100) different major class We found that many Chinese bisyllabic words are formed as so-called biased type of words (these are the so-called .1, )p* type of words). Examples are(See Table 2): Table 2 Semantic distances of some words W Ca Cb Owa Owb .5 ab Ewa- Owb OA% Ic01 Ea03 Ic01 4 0 4 4 gm Ec04 Bf01 Ec04 4 0 4 4 St—a Fb03 Fb03 Ea09 0 4 4 -4 …

Content truncated to 500 words