The first one is Machine-Tractable Dictionary (MTD) such as WordNet and LDOCE (Longman Dictionary of Contemporary English) especially LDOCE has been used extensively in NLP research and provide a broad set of senses for sense tagging. MTD contains word senses and their definitions are defined in term of descriptive and tagged primitives (words attached with sense number). Primitives are a set of words derived from dictionary (Guo, 1989b) and it is used to define the definition of a word sense. (For further information about primitives, please refer to Wilks.Y (1977)). For example, father#l has a definition defined by using four primitives that are 'titlel\'respect2\ 'priest3' and 'church^ (refer figure 1).

For SENSEVAL-2 competition, the pre­release WordNetl.7 was used for this purpose. After WordNetl.7 was downloaded, the entries including their definition, sense number and sense id in WordNet was extracted and written into a temporary file. Primitives (not tagged) were derived from the words used in the word senses' definition. Then, the first 7 words of the definition text of the WordNet dictionary were disambiguated using the information from an existing MTD (LDOCE) and the derived primitives (Guo, 1998a). The existing MTD (LDOCE) contained word senses and the words in their definition are already tagged.

Thus, a new MTD (the pre-release WordNet 1.7) was ready for the usage of tagging process.

Conceptual distance data is showing the relatedness between two tagged primitives.

Basically, the conceptual distance data is calculated by using content terms in the definition to determine the relatedness measure between two primitives layer by layer. The definitions of the primitives are getting from the existing MTD (LDOCE). It is important to note that a tagged primitive is also a word sense. For example, the first and second layer of referential definition for word sense förecast2' is:

forecast2 [def] predict 1 in2 advance3 predict 1 [def] makel a2 prediction3 about4; telll in2 advance3 1- referential layer: forecast2 predict 1 advance3 2- referential layer: predictl makel prediction3 telll (note:'advance3' is omitted because it has been counted in the first layer)

Formula used to compute the relatedness percentage:

target word sense n2 = no. of the element in the first layer of

second target word sense n3 = no. of the element in the second layer of

first target word sense n4 = no. of the element in the second layer of

second target word sense q, ql, q2, q3 = no. of common content terms

for each comparison pi, p2 = final value of the relatedness measure

The third one is sense tagger. Sense tagger will get the input from MTD and its conceptual distance data among primitives to do the word sense disambiguation. Currently, the tagger consists of three processes:

• Preprocess process.

• Dictionary look-up process

• Numerical calculation algorithm

In the preprocess process, test data, which is downloaded for the usage of SENSEVAL-2, is going through several processes before tagging process takes place. The first process is separating the given text into sentences using full stops as separator. After that, the words in the sentences that do not require tagging will be removed, leaving only the heads (words to be sense-tagged) behind. Then, each word in the sentences will be stemmed, leaving only morphological root. The list of heads is then cut into chunks of three successive heads to be tagged in seconds.

In dictionary look-up module, word senses with their definition for each of the words in a chunk is extracted from MTD.

After that, sense tagger will use numerical calculation algorithm to choose the suitable word sense for the words in the sentence. This algorithm is to compute the path value among the definition of the word senses in a sentence. This is done first, by summing up the semantic data from conceptual distance data when comparison among primitives in the definitions for the word sense pairs in the sentence. After that, the result of summation has to be multiplied with the distance value between the two words in the sentence. This distance value basically depends on total words in a sentence. For example, sentence "Father marry couple", the distance value for 'father' and 'marry' is 2 whereas the distance value for 'father and 'couple' is 1. This is because word father' is closer to the word 'marry' than 'couple'. Then this computation continues for the other of word sense pairs.

Finally, this algorithm will compare the path values for the combination of word senses in a sentence and find the highest path value. Then this algorithm assigns the best combination of senses to each word in the sentence.

For example, with reference to Figure 1, assume that words such as 'father', 'marry' and 'couple' have two senses only.

In the first step, definition of sense 1 from 'father' will compare with definition of sense 1 from 'marry'.

father 1 (Note: xl, x2, ....xl2 are the values accessed from conceptual data. )

In the second step, definition of sense 1 from 'father' will compare with definition of sense 1 from 'couple'. The total comparison is 4*4=16 and total value extracted from conceptual distance data is Y. Then in the third step, definition of sense 1 from 'couple' will compare with definition of sense 1 from 'marry'. The total comparison is 4*3=12 and total value extracted from conceptual distance data is Z. The calculations for second step and third are as same as the step.