Vol.11, No.8 2016
Copyright ⓒ 2016 SERSC
19
With those conditions in mind, the construction of hypernym-hyponym pairs that we propose consists of three processes:
1. Word Sense Disambiguation WSD every word acquired in a definition to find the
precise part-of-speech for each of them. In this case we use Lesk algorithm as the simple WSD process in this stage [22].
2. Definition simplification so that only the part of the definition that contains
information of hypernym of lemma is extracted. The main task in this process is to discard the synonym part of a lemma entry.
3. Acquisition of hypernym from the simplified definition. A very important heuristic
that we use in this process is that the first noun-phrase from the definition has high probability to be hypernym candidate of a synset even though definition
exploration of every word in the definition still needs to be done see Figure 4.
4.2. Is part-of Holonym-Meronym Construction
A mapping approach is used in the construction of holonym-meronym relationship or frequently referred as part-whole relationship in our research. For every Indonesian noun
synset s
i
, an English translation synset s
e
can be obtained. Holonym and meronym synset from s
e
, s
eh
and s
em
respectively, are acquired through the use of PWN. Furthermore, by translating s
eh
and s
em
into Indonesian, a set of synset s
ih
and s
im
is created. Finally, s
ih
and s
im
will be linked with s
i
, each as holonym and meronym synset in Indonesian. The main difficulty in this approach is the ambiguity in both translation to the target
language, whether to Indonesian or to English. When translating s
i
to English, a set of English synset, S
e
, is produced instead of a single synset. Thus, WSD is needed to choose the appropriate s
e
out of the member of S
e
. Similar case can be seen as well when translating s
eh
and s
em
into Indonesian. Figure 5 shows the approach that we offer for synset level.
We also have tried to do similar mapping approach for word level. This word mapping can be done by simplifying the algorithm shown in Figure 5; by removing all WSD tasks
required to get the appropriate synset, whether in English step 3 or in Indonesian step 7 and 12 since those are not yet required. Nevertheless, the post processing for assignment
to the Indonesian synsets Section 3 that have been acquired accurately is still needed.
1. For Each s
i
in AllIndonesianSynset 2.
S
e
TranslateIndonesianToEnglishs
i
3. s
e
GetWSDMappedSynsetS
e
, s
i
4. s
eh
GetAllPWNHolonym s
e
5. For Each h in s
eh
6. S
i
TranslateEnglishToIndonesianh 7.
s
ih
GetWSDMappedSynsetS
i
, h 8.
GenerateHolonymMeronymRelationInIndonesians
i
,s
ih
9. s
em
GetAllPWNMeronym s
e
10. For Each m in s
em
11. S
i
TranslateEnglishToIndonesianm 12.
s
im
GetWSDMappedSynsetS
i
, m 13.
GenerateHolonymMeronymRelationInIndonesians
i
,s
im
14. Return
Figure 5. Algorithm for Part-of Semantic Relations Construction with Synset Mapping
Online Version
Only.
Book made by this
file is
ILLEGAL.
20
Copyright ⓒ 2016 SERSC
5. Gloss Acquisition
