Difference between revisions of "Full-Text: Japanese"

From BaseX Documentation
Jump to navigation Jump to search
(6 intermediate revisions by the same user not shown)
Line 1: Line 1:
This article is linked from the [[Full-Text]] page. It gives some insight into the implementation of the full-text features for Japanese text corpora. The Japanese version is [http://files.basex.org/etc/ja-ft.pdf also available as PDF].
+
This article is linked from the [[Full-Text]] page. It gives some insight into the implementation of the full-text features for Japanese text corpora. The Japanese version is [https://files.basex.org/etc/ja-ft.pdf also available as PDF].
Thank you to [http://blog.infinite.jp Toshio HIRAI] for integrating the lexer in BaseX!
+
 
 +
The lexer was contributed by Toshio HIRAI.
  
 
=Introduction=
 
=Introduction=
  
The lexical analysis of Japanese documents is performed by  
+
The lexical analysis of Japanese documents is performed by [https://igo.osdn.jp/ Igo]. Igo is a ''morphological analyser'', and some of the advantages and reasons for using Igo are:
[http://igo.sourceforge.jp/ Igo]. Igo is a ''morphological analyser'',
 
and some of the advantages and reasons for using Igo are:
 
* compatible with the results of a prominent morphological analyzer "MeCab"
 
* it can use the dictionary distributed by the Project MeCab
 
* the morphological analyzer is implemented in Java and is relatively fast
 
  
Japanese tokenization will be activated in BaseX if Igo is found in the
+
* Compatible with the results of a prominent morphological analyzer "MeCab".
classpath. [http://en.sourceforge.jp/projects/igo/releases/ igo-0.4.3.jar]
+
* It can use the dictionary distributed by the Project MeCab.
of Igo is currently included in all distributions of BaseX.
+
* The morphological analyzer is implemented in Java and is relatively fast.
 +
 
 +
Japanese tokenization will be activated in BaseX if Igo is found in the classpath. [https://osdn.net/projects/igo/releases/ igo-0.4.3.jar] of Igo is currently included in all distributions of BaseX.
  
 
In addition to the library, one of the following dictionary files must either be unzipped into the current directory, or into the <code>etc</code> sub-directory of the project’s [[Configuration#Home Directory|Home Directory]]:
 
In addition to the library, one of the following dictionary files must either be unzipped into the current directory, or into the <code>etc</code> sub-directory of the project’s [[Configuration#Home Directory|Home Directory]]:
* IPA Dictionary: http://files.basex.org/etc/ipadic.zip
+
 
* NAIST Dictionary: http://files.basex.org/etc/naistdic.zip
+
* IPA Dictionary: https://files.basex.org/etc/ipadic.zip
 +
* NAIST Dictionary: https://files.basex.org/etc/naistdic.zip
  
 
=Lexical Analysis=
 
=Lexical Analysis=
Line 63: Line 62:
 
=Token Processing=
 
=Token Processing=
  
"Fullwidth" and "Halfwidth" (which is defined by
+
"Fullwidth" and "Halfwidth" (which is defined by [https://unicode.org/Public/UNIDATA/EastAsianWidth.txt East Asian Width Properties]) are not distinguished (this is the so-called ZENKAKU/HANKAKU problem).
[http://unicode.org/Public/UNIDATA/EastAsianWidth.txt East Asian Width Properties])
+
 
are not distinguished (this is the so-called ZENKAKU/HANKAKU problem).
+
For example, <code>XML</code> and <code>XML</code> will be treated as the same word. If documents are ''hybrid'', i.e. written in multiple languages​​, this is also helpful for some other options of the XQuery Full Text Specification, such as the [https://www.w3.org/TR/xpath-full-text-10/#ftcaseoption Case] or the [https://www.w3.org/TR/xpath-full-text-10/#ftdiacriticsoption Diacritics] option.
For example, <code>XML</code> and <code>XML</code> will be treated
 
as the same word. If documents are ''hybrid'', i.e. written in multiple languages​​,
 
this is also helpful for some other options of the XQuery Full Text Specification,
 
such as the [http://www.w3.org/TR/xpath-full-text-10/#ftcaseoption Case] or the
 
[http://www.w3.org/TR/xpath-full-text-10/#ftdiacriticsoption Diacritics] Option.
 
  
 
=Stemming=
 
=Stemming=
Line 92: Line 86:
 
is returned for the following two types of queries:
 
is returned for the following two types of queries:
  
<pre class="brush:xquery">
+
<syntaxhighlight lang="xquery">
 
'私は本を書いた' contains text '書く' using stemming using language 'ja'
 
'私は本を書いた' contains text '書く' using stemming using language 'ja'
 
'私は本を書く' contains text '書いた' using stemming using language 'ja'
 
'私は本を書く' contains text '書いた' using stemming using language 'ja'
</pre>
+
</syntaxhighlight>
  
 
=Wildcards=
 
=Wildcards=
Line 104: Line 98:
 
queries both return <code>true</code>:
 
queries both return <code>true</code>:
  
<pre class="brush:xquery">
+
<syntaxhighlight lang="xquery">
 
'芥川龍之介' contains text '.之介' using wildcards using language 'ja'
 
'芥川龍之介' contains text '.之介' using wildcards using language 'ja'
 
'芥川竜之介' contains text '.之介' using wildcards using language 'ja'
 
'芥川竜之介' contains text '.之介' using wildcards using language 'ja'
</pre>
+
</syntaxhighlight>
  
 
However, there is a special case that requires attention. The following
 
However, there is a special case that requires attention. The following
 
query will yield <code>false</code>:
 
query will yield <code>false</code>:
  
<pre class="brush:xquery">
+
<syntaxhighlight lang="xquery">
 
'芥川龍之介' contains text '芥川.之介' using wildcards using language 'ja'
 
'芥川龍之介' contains text '芥川.之介' using wildcards using language 'ja'
</pre>
+
</syntaxhighlight>
  
 
This is because the next word boundary metacharacters
 
This is because the next word boundary metacharacters
Line 120: Line 114:
 
an additional whitespaces as word boundary:
 
an additional whitespaces as word boundary:
  
<pre class="brush:xquery">
+
<syntaxhighlight lang="xquery">
 
'芥川龍之介' contains text '芥川 .之介' using wildcards using language 'ja'
 
'芥川龍之介' contains text '芥川 .之介' using wildcards using language 'ja'
</pre>
+
</syntaxhighlight>
  
 
As an alternative, you may modify the query as follows:
 
As an alternative, you may modify the query as follows:
  
<pre class="brush:xquery">
+
<syntaxhighlight lang="xquery">
 
'芥川龍之介' contains text '芥川' ftand '.之介' using wildcards using language 'ja'
 
'芥川龍之介' contains text '芥川' ftand '.之介' using wildcards using language 'ja'
</pre>
+
</syntaxhighlight>

Revision as of 12:56, 2 July 2020

This article is linked from the Full-Text page. It gives some insight into the implementation of the full-text features for Japanese text corpora. The Japanese version is also available as PDF.

The lexer was contributed by Toshio HIRAI.

Introduction

The lexical analysis of Japanese documents is performed by Igo. Igo is a morphological analyser, and some of the advantages and reasons for using Igo are:

  • Compatible with the results of a prominent morphological analyzer "MeCab".
  • It can use the dictionary distributed by the Project MeCab.
  • The morphological analyzer is implemented in Java and is relatively fast.

Japanese tokenization will be activated in BaseX if Igo is found in the classpath. igo-0.4.3.jar of Igo is currently included in all distributions of BaseX.

In addition to the library, one of the following dictionary files must either be unzipped into the current directory, or into the etc sub-directory of the project’s Home Directory:

Lexical Analysis

The example sentence "私は本を書きました。(I wrote a book.)" is analyzed as follows.

<syntaxhighlight> 私は本を書きました。 私 名詞,代名詞,一般,*,*,*,私,ワタシ,ワタシ は 助詞,係助詞,*,*,*,*,は,ハ,ワ 本 名詞,一般,*,*,*,*,本,ホン,ホン を 助詞,格助詞,一般,*,*,*,を,ヲ,ヲ 書き 動詞,自立,*,*,五段・カ行イ音便,連用形,書く,カキ,カキ まし 助動詞,*,*,*,特殊・マス,連用形,ます,マシ,マシ た 助動詞,*,*,*,特殊・タ,基本形,た,タ,タ 。 記号,句点,*,*,*,*,。,。,。 </syntaxhighlight>

The element of the decomposed part is called "Surface", the content analysis is called "Morpheme". The Morpheme component is built as follows:

<syntaxhighlight> 品詞,品詞細分類1,品詞細分類2,品詞細分類3,活用形,活用型,原形,読み,発音 (POS, subtyping POS 1, subtyping POS 2, subtyping POS 3, inflections, use type, prototype, reading, pronunciation) </syntaxhighlight>

Of these, the surface is used as a token. Also, The contents of analysis of a morpheme are used in indexing and stemming.

Parsing

During indexing and parsing, the input strings are split into single tokens. In order to reduce the index size and speed up search, the following word classes have been intentionally excluded:

  • Mark
  • Filler
  • Postpositional particle
  • Auxiliary verb

Thus, in the example above, , , and 書き will be passed to the indexer for each token.

Token Processing

"Fullwidth" and "Halfwidth" (which is defined by East Asian Width Properties) are not distinguished (this is the so-called ZENKAKU/HANKAKU problem).

For example, XML and XML will be treated as the same word. If documents are hybrid, i.e. written in multiple languages​​, this is also helpful for some other options of the XQuery Full Text Specification, such as the Case or the Diacritics option.

Stemming

Stemming in Japanese means to analyze the results of morphological analysis ("verbs" and "adjectives") that are processed using the "prototype".

If the stemming option is enabled, for example, the two statements "私は本を書いた (I wrote the book)" and "私は本を書く (I write the book)" can be led back to the same prototype by analyzing their verb:

<syntaxhighlight> 書く 動詞,自立,*,*,五段・カ行イ音便,基本形,[書く],カク,カク

書い 動詞,自立,*,*,五段・カ行イ音便,連用タ接続,[書く],カイ,カイ た 助動詞,*,*,*,特殊・タ,基本形,た,タ,タ </syntaxhighlight>

Because the "auxiliary verb" is always excluded from the tokens, there is no need to consider its use. Therefore, the same result (true) is returned for the following two types of queries:

<syntaxhighlight lang="xquery"> '私は本を書いた' contains text '書く' using stemming using language 'ja' '私は本を書く' contains text '書いた' using stemming using language 'ja' </syntaxhighlight>

Wildcards

The Wildcard option in XQuery Full-Text is available for Japanese as well. The following example is based on '芥川 龍之介(AKUTAGAWA, Ryunosuke)', a prominent Japanese writer, the first name of whom is often spelled as "竜之介". The following two queries both return true:

<syntaxhighlight lang="xquery"> '芥川龍之介' contains text '.之介' using wildcards using language 'ja' '芥川竜之介' contains text '.之介' using wildcards using language 'ja' </syntaxhighlight>

However, there is a special case that requires attention. The following query will yield false:

<syntaxhighlight lang="xquery"> '芥川龍之介' contains text '芥川.之介' using wildcards using language 'ja' </syntaxhighlight>

This is because the next word boundary metacharacters cannot be determined in the query. In this case, you may insert an additional whitespaces as word boundary:

<syntaxhighlight lang="xquery"> '芥川龍之介' contains text '芥川 .之介' using wildcards using language 'ja' </syntaxhighlight>

As an alternative, you may modify the query as follows:

<syntaxhighlight lang="xquery"> '芥川龍之介' contains text '芥川' ftand '.之介' using wildcards using language 'ja' </syntaxhighlight>