X hits on this document

PDF document

An LMF-based Web Service for Accessing WordNet-type Semantic Lexicons - page 4 / 7

14 views

0 shares

0 downloads

0 comments

4 / 7

Figure 4: Result of /wn-ja/word/form/

.

document is <LexicalResource> whose immediate daughters are <GlobalInformation> (decorated with the access URI) and <Lexicon>.

Another example that would be of interest is text search on the gloss definition of a synset.

full- For

example,

the

URI

/edr/synset/definition/the

capital of France/ gives us a whose gloss text contains the string

set of synsets the capital of

France

ample

as shown in Fig. 5.

Notice from this

that

not

only

the

<Synset>

elements

but

ex- the

associated

<LexicalEntry>

elements

are

included

in the XML quirement of

document to fulfill the structural the <Lexicon> element imposed

re- by

LMF. This the access

figure also serves as a nice example of to EDR, where the definition of a con-

cept is given both in English and Japanese: in Fig. the Japanese <LexicalEntry> (writtenForm=”

5

”), along with two other variants, and the English <LexicalEntry> (writtenForm=”Paris”) simulta- neously refer to a <Synset> (id=edr cph-104edc-x), whose definition is given in the two languages.

As described above, the core part in the URI pattern re- flects the lexicon structure specified by LMF, while the op- tional part that utilizes the query parameter construct is in- troduced to support other possible usage patterns. Here we show two synset search examples: (1) by specifying a {word-form, part-of-speech, sense-number} triple, and (2) by restricting a conceptual relation.

Figure 6 shows the former example, where a PWN synset is queried by specifying the triple {form=”bank”, pos=noun, sense-number=2}. Notice here that we introduced the query parameter (?synset by index) along with the /word/form-pos/ construct that specifies a combina-

tion of writtenForm and partOfSpeech attributes.

Figure 7 shows the latter example, where a set of PWN synsets is queried by specifying a base sysnset (id=eng- 09-08420278-n; it is the result synset shown in Fig. 6) and a conceptual relation (hyponym). Thanks to the rendering with Firefox browser, only two of eleven <Synset> ele- ments are unfolded and shown.

4.3. A Note on the Implementation

Each target lexical resource has been stored in a relational database in advance, and the user-specified URI is mapped

into

the

corresponding

SQL

statement.

We

employed

Python-based Django6 as the back-end mechanism. It is extremely effective in implementing this type of Web ser- vice, because it provides a built-in mechanism for URI-to- resource (database entry) mapping (URL dispatcher) along with a template mechanism for generating XML documents that observe an XML schema.

5. Modeling Multilingual Semantic Lexicons with LMF: A Revision Proposal

As already shown, EDR is an innately bilingual lexicon, in- sisting that the <Synset> configuration can be bilingual. In fact, a substantial number of concept nodes in EDR have both Japanese and English definitions. To encode such a bilingual/multilingual synset configuration, at least two so- lutions can be possible:

  • 1.

    allowing a <Synset> element to have multiple <Definition> elements, each of which is given in an explicitly specified language ;

  • 2.

    utilizing the Sense Axis device provided by LMF.

The former is the solution proposed in this paper, where lang attribute is being introduced to <Definition> el- ement in order to explicitly specify the language of a defi- nition text. Figure 5 exemplifies the case. On the other hand, with the latter solution, the ”the capital of France” entry in EDR, as already shown in Fig. 5, can be alternatively organized as illustrated in Fig. 8. As depicted in the figure, the original EDR concept node (id=edr cph-104edc-x) is separated into the Japanese Synset node (id=edr cph-104edc-J) and the English Synset node (id=edr cph-104edc-E). These two nodes are then interconnected by a Sense Axis node. This organization may be particularly effective in case where other synsets in other languages would be further incorpo- rated. With this organization, however, the Sense Axis node has to be linked to the EDR conceptual system via Interlingual External Ref. That is, the EDR conceptual system, now considered as an external ontol- ogy rather than a part of the dictionary system, has to be modeled and represented using some framework other than LMF. Comparing these two solutions, it would be still natural and simple to accept the revision of LMF as proposed to repre- sent innately bilingual/multilingual lexicons, as far as we

6http://www.djangoproject.com/

510

Document info
Document views14
Page views14
Page last viewedSat Dec 03 20:52:30 UTC 2016
Pages7
Paragraphs191
Words3665

Comments