JPH01211176A - Morpheme analyzing system - Google Patents

Abstract

PURPOSE:To enhance the performance of morpheme analysis by determining whether or not a character can link to an adjacent vocabulary item and form a compound word during an input. CONSTITUTION:The information of an independent word, an attached word, a prefix and a suffix is stored into a word dictionary 103. When an analysis control part 101 retrieves the word dictionary 103 at the character position of a present position 201 during the analysis, since the concerned word does not exist, it retrieves a letter dictionary 104, the letter of KANJI (Chinese character) 'HO' (to visit) is obtained. Since the concerned word does not exist at the next character position in the dictionary 103, the control part 101 retrieves the dictionary 104, and the letter of KATAKANA (square form of Japanese sillabary) 'SO' is obtained. Here, whether or not the character of KATAKANA 'SO' and the character of KANJI 'HO' can make a combined word is decided, and since dictionary information 301 of the KANJI 'HO' and dictionary information 302 of KATAKANA 'SO' do not contradict, it is decided that the both can be combined.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a B-format cumulative analysis method, and is particularly applicable to Japanese morphological analysis, which is an essential component of Japanese text-to-speech synthesis systems, Japanese-English machine translation systems, etc. It is related to the method.

[Conventional technology]

Conventionally, as described in "Information Processing" Vol. 27, No. 8, page 951, Japanese morphological methods such as the longest match method, the two-clause longest match method, the minimum number of segments method, and the minimum cost method on the extended clause model have been used. Analysis techniques are known.

[Problem to be solved by the invention]

In conventional morphological analysis methods, in addition to word-phrase categories such as independent words and adjunct words, only neologism elements such as prefixes and suffixes were used. By creating these word categories of prefixes and suffixes, it was possible to deal with coined words created by combining prefixes and suffixes with other words.

However, in Japanese texts, there are not only such types of coined words, but also combinations of "coined word elements" that cannot be considered words on their own, such as "visiting China,""visiting the Soviet Union," and "visiting Korea." In conventional morphological analysis methods, if such a coined word is not registered as a single word, it must be treated as an unknown word.

An object of the present invention is to provide a highly accurate morphological analysis method that improves this drawback.

[Means to solve the problem]

Although the morphological analysis method of the present invention cannot necessarily combine with words to create coined words, it has a word category called "jimo" that has the ability to combine with other affixes or "jimo" to create coined words. Then, the attribute information that a word entry item that can be combined with the ``jimo'' to form a compound word is held together with the ``jimo'', and the above [if the notation of ``jimo J'' exists during input] , using the attribute information to determine whether or not the "jimo" can be combined with adjacent word selection items to form a compound word during the input.

[Effect]

In Japanese texts, in addition to what is already considered to be a word, ``jimo'', that is, coined words formed from coined word components with strong word-coining ability, frequently appear. For example, "Shinmeikai Japanese Dictionary J (Sanseido, first edition 19
In 1972), it is called a ``coined word component''. In the present invention, we have developed an umbrella category called ``word'' such as independent words and adjunct words, and an umbrella category called ``ji'' such as prefixes and suffixes, which are conventionally treated in Japanese morphological analysis. In addition, although it is not necessarily possible to combine words to create coined words, it is possible to combine them with other affixes or
Morphological analysis is performed using the word category ``jimo'', which has the ability to combine with ``jimo'' to form coined words.

As a result, when there is a partial string of characters that are not registered as words but are registered as "jimo" in the input text, word candidates can be selected without treating that section as an unknown word. It becomes possible to form.

〔Example〕

Next, the present invention will be described in detail according to embodiments with reference to FIGS. 1 to 3. FIG. 1 is a block diagram showing an embodiment for realizing the principle of the present invention.

In FIG. 1, the connection table 102 is a table that gives a value of 1 or O to the ordered pair (r+, IJ) of the lower classification categories of parts of speech. If this is 1, r,
and 1 no are grammatically contiguous, otherwise they are not contiguous. The word dictionary 103 stores information on independent words, attached words, prefixes, and suffixes.

The alphabet dictionary 104 stores information on the notation of each alphabet, its category, and the categories of partners that can be combined to form words.

The analysis control unit 101 uses a word dictionary 103 for input sentences.
If there are multiple candidate words, the candidate word with the longest notation is selected first, and the word is stored in the connection table 102.
is used to test the possibility of adjacency with the word on the left, and if adjacency is possible, the candidate word is assumed to be a word in that section. If the candidate word does not exist in the word dictionary 103, the character dictionary 104 is searched, and the connection table 10 is similarly searched.
2 is used to test the possibility of adjacency with the word or character on the left, and if adjacency is possible, it is assumed that the candidate character can exist in that interval. If the candidate at the current position is a jimo and the left side is a jimo, the possibility of coining is determined using the category of the jimo obtained from the jamo dictionary 104 and the category of the other party that can be combined to form a coined word. Determine.

If a candidate word does not exist even if you look up a dictionary at a certain position, or if it is determined that adjacency is not possible by testing the possibility of adjacency using the connection table 102, or if there is a possibility of coining a word by combining a character and a character. If it is determined that a word cannot be coined, the process returns to the position of the previously assumed word, selects the next candidate at that position, and moves on.

If the dictionary lookup, connection test, or test of the possibility of coining a character is successful, the word dictionary 103 is searched again at the next character position of the candidate word or character notation, and if there are no candidates, the character 8104 is searched. Repeat the above process. When the character type reaches the position where the character type changes from kanji to other characters, the analysis control unit 101 determines the word assumed up to that point as the analysis result.

FIG. 2 is an explanatory diagram showing an example of input Japanese text. Assume that the dictionary has now been used to find ``Foreign Minister's''. The analysis control unit 101 first searches the word dictionary 103 for the character position 201, and since the corresponding word does not exist, it then searches the jimo dictionary 104 and obtains the jimo "vis". Since the corresponding word does not exist in the word dictionary 103 at the next character position, the analysis control unit 101 searches the character dictionary 104 and obtains the character "so". Here, it is determined whether it is possible to create a word by combining the character capital "SO" and the character capital "Vi".
2 is not contradictory, so it is determined that they can be combined.

〔Effect of the invention〕

As described above, according to the morphological analysis method of the present invention, it is possible to analyze Japanese sentences that contain compound words that are not registered in word dictionaries but are formed by the word-coining ability of "jimo". . In Japanese, words are often coined using "jimo", so the improvement in morphological analysis performance according to the present invention is extremely effective.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram showing an example of input Japanese text, and FIG. 3 is an explanatory diagram showing an example of the dictionary contents of "Jibo". 101...Analysis control unit, 102...Connection table,
103...Word dictionary, 104...Character dictionary.

Claims (1)

[Claims] Although it is not necessarily possible to create coined words by combining with words, there is a lexical category called "jimo" that has the ability to combine with other affixes or "jimo" to create coined words, and the above-mentioned "jimo"
The attribute information that a lexical item that can be combined with to form a compound word is held together with its "jimo", and when the "jimo" notation is present in the input, the attribute information is used to A morphological analysis method, characterized in that it is determined during the input whether the "jimo" can be combined with adjacent lexical items to form a compound word.