KR100918842B1 - Apparatus for substitute restoration and method thereof - Google Patents

Abstract

The alternative term reference solving method according to the present invention comprises: a substitute term recognition step of recognizing a substitute term included in the string by analyzing the input string; A semantic analysis step of analyzing the overall meaning of the substitute term; A target candidate selecting step of chunking a string and assigning a concept to each chunk, and comparing target concepts and overall meanings to select target candidates that can be represented by a substitute term; Ranking target candidates to select a reference target to be used for the restoration of the term, and reconstructing the term using the reference target. By such a configuration, it can be effectively used in various applications requiring the understanding of natural language sentences, and it is effective in restoring substitute words with high accuracy in long and complex sentences.

Description

Apparatus for substitute restoration and method

The present invention relates to a terminology reference solving apparatus and a method thereof, and more particularly, to a terminology reference solving apparatus and method for automatically resolving a reference term in natural language.

Substitute terms are frequently used in order to avoid repetition of natural language sentences and achieve unity of context, and have been variously developed in the field of computational linguistics, but they are not easy to solve. Substitution is a phenomenon in which a repeated element is replaced by using a pronoun, a directive, a recursive pronoun, an omission, and a general noun when the same element is repeated between a sentence or several sentences.

Referring to the related art of an apparatus and a method for automatically solving a terminology reference solution, a method of processing a terminology has been proposed in "Interactive Learning Aid Device and Its Conversation Analysis Method" of Korean Application No. 10-1998-0048384. .

1 is a view for explaining a method for solving the conventional substitute term reference.

The conventional interactive learning aid shown in FIG. 1 includes a speech input device 1, a speech / text converter 2, a morpheme interpreter 3, a syntax analyzer 4, a semantic interpreter 5, a speech analyzer ( 6), conversation manager 7, response generator 8, output device 9, pre-storage device 10, and knowledge base storage device 11.

The conventional term reference solution using the above-described configuration uses the result of detailed semantic analysis because it uses a method of simply mapping keyword content words and substitute terms with reference to the data of the pre-storage device 10. it's difficult. And since there is no process of ranking in the candidates indicated by a substitute term (ie, objects that can be pointed to in the previous context), it can be applied to simple conversational analysis, but it is accurate in long and complex sentences such as newspaper articles. The problem arises that it is difficult to guarantee.

Previously, the technique related to the reference resolution of the term used a simple mapping method according to the type of the term after defining the meaning of the limited number to solve the limited term in the sentence, but the present invention ranks the candidate candidates in the entire document. By using the method of finding a reference object of a substitute term, it has high accuracy in long and complicated sentences such as newspaper articles.

The present invention is to solve the above problems, by selecting the target candidate by means of semantic analysis and individual name recognition, ranking the selected target candidate to select the reference target, and restore the selected reference target in substitute terms It is an object of the present invention to provide a terminology reference solving apparatus and method having high accuracy in long and complex sentences.

An alternative term reference solving apparatus according to the present invention comprises: a term term recognition unit that receives a string and recognizes a term term; A semantic analysis unit that analyzes the overall meaning of the substitute term; An entity name recognition unit recognizing an entity name of a character string; A chunking unit that receives the string recognized by the entity name and assigns a concept to each chunk after chunking; It includes a terminology restoration unit that selects a reference object to be used for terminology restoration by comparing and analyzing the whole meaning of the terminology and the concept assigned to each chunk.

Preferably, the substitute word recognition unit may recognize 'pronoun' and 'directive + noun' as substitute words in the string.

In addition, the semantic analysis unit may grasp the overall meaning of the substitute term based on a plurality of semantic codes having a hierarchical semantic structure.

In addition, the chunk assigns a concept to each chunk based on the meaning of an entity name or noun in each chunk, and the concept assigned to each chunk is one of a plurality of semantic codes having a hierarchical semantic structure. It is characterized by that.

On the other hand, a substitute word reference solution according to the present invention is a substitute word recognition step of recognizing a substitute word included in the character string by analyzing the input string; A semantic analysis step of analyzing the overall meaning of the substitute term; A target candidate selecting step of chunking a string and assigning a concept to each chunk, and comparing target concepts and overall meanings to select target candidates that can be represented by a substitute term; Ranking target candidates to select a reference target to be used for the restoration of the term, and reconstructing the term using the reference target.

Preferably, the term recognition step is characterized by recognizing 'pronoun' and 'directive + noun' as a substitute by morphologically analyzing the character string.

In addition, the semantic analysis step is characterized by analyzing the entire meaning in the substitute using a plurality of semantic codes having a hierarchical semantic structure.

In addition, the target candidate selection step, the step of recognizing the individual name of the character string; And assigning a concept to each chunk based on the meaning or entity name of the noun in each chunk.

In the substitute word reconstruction step, the concept of the target candidate corresponds to a lower meaning in the overall meaning of the substitute word, so that the weight is given a higher weight.

In addition, when there are target candidates given the same weight, the recently selected target candidate is given priority, and under the same conditions, essential components of the sentence are given priority.

As described above, the present invention can perform natural language analysis more accurately and efficiently by recognizing and reconstructing a substitute word by using a substitute word overall meaning, individual name, hierarchical semantic structure, and chunking. Therefore, it can be effectively used in various applications requiring the understanding of natural language sentences, and it is effective in restoring substitute words with high accuracy in long and complex sentences.

1 is a view for explaining a method for solving the conventional substitute term reference.

2 is a block diagram for explaining a terminology reference solving apparatus according to the present invention.

3 is a view for explaining a substitute term reference solution according to the present invention.

<Explanation of symbols for main parts of the drawings>

110: term recognition unit 120: semantic analysis unit

130: object name recognition unit 140: chunking unit

150: Terminology Restoration Unit

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings. Here, the repeated description, well-known functions and configurations that may unnecessarily obscure the subject matter of the present invention, and detailed description of the configuration will be omitted. Embodiments of the present invention are provided to more completely describe the present invention to those skilled in the art. Accordingly, the shape and size of elements in the drawings may be exaggerated for clarity.

2 is a structural diagram for explaining a terminology reference solving apparatus according to the present invention.

Referring to FIG. 2, the apparatus for solving a terminology reference according to the present invention includes a terminology recognition unit 110, a semantic analysis unit 120, an object name recognition unit 130, a chunking unit 140, and a terminology restoration unit. And 150.

The substitute term recognition unit 110 receives a character string, performs morphological analysis, and tags parts of speech information corresponding to each morpheme. Subsequently, the substitute word recognition unit 110 recognizes the pronouns and the directive + correspondence as substitute words in the string. That is, in the following example, "the company" is recognized as a substitute.

"Microsoft has just started selling a new operating system, Windows Vista. The company aims to improve its previous operating system to provide a more convenient product for its users."

Directives are used instead to avoid unnecessary repetition when referring to the above-mentioned contents once again, such as 'yi', 'low', 'he', and 'other'. Directive types include noun phrases, pronouns, and reflexive pronouns. In the case of IT documents, there are many noun phrases and almost no pronouns or reflexive pronouns. Morphology refers to the smallest unit at the morphological level of language, which confers the function of meaning.

The semantic analysis unit 120 grasps the overall meaning of the substitute term recognized through the substitute term recognition unit 110 and tags all the meanings in the substitute term. In more detail, all nouns are tagged based on the semantic codes classified into a plurality of hierarchical semantic structures (for example, about 400), including all upper meanings of the substitute terms.

Standards> Normal> Institution Shape> Forms> Culture> Religion Standard> norm> institution> insurance Shape> Sign Standard> Regulations Shape> Format Standards> Regulations> Treaty Object> ring Standard> Cooking Object> tube Basis> Cooking> Ichi> Base Objects> Standards> Cooking> Ichi> Principle Object> Machine> Utensil Standard> instructions Objects> Machines> Instruments> Instruments Standard> standard Objects> Machines> Tools> Tools Symbol> Language Object> Machine> Utensil> Bowl Symbol> Name Stuff> Machine> Utensils> Tools Symbol> Table Stuff> Machine> Utensils> Tools> Stationery Unit> cell Object> Machine> Utensil> Tool> Phrase Target> Objective Stuff> Machine> Apparatus> etc Audience> Problem> Task Stuff> Machine> Utensils> Weapon Audience> Problem> Task> Problem Objects> Machine> Utensils> Future> Birds Object> Object Objects> Machine> Utensils> Furniture> House Object> Object> Crystal Stuff> Machine> Utensils> Musical Instruments Target> Object> Particle Objects> Machines> Utensils> Fitness Equipment Object> Object> Astro Stuff> Machine> Utensils> Medical Device Object> Object> Transparent Stuff> Machine> Utensils> Toy Object> Product> Art Object> Machine> Utensil> House Object> Product> Creative Stuff> Packs Grand Prize> Products> Creative Works> Artwork> Written Object> String Target> Relative Objects> Waste Shape> appearance> appearance> scenery Stuff> lump Shape> Appearance> Appearance> Sight> Video Stuff> Eating Shape> Text Stuff> Eat> Food> Grain Shape> Shape> Features> Details> Terrain Object> Food> Food Shape> Shape> Features> Shape> Image> Picture> Photo Object> Food> Food> Chapter Shape> Shape> Texture> System Object> Food> Food> jang> Red pepper paste Shape> Pattern Article> eat> food> jang> Miso Shape> Font Object> Food> Food> Fish Shape> actual Object> Food> Food> Fish> Mackerel Shape> Form Object> Eat> Food> Fish> Swordfish

As shown in Table 1, a semantic code classified into a hierarchical semantic structure applied to the present invention has a higher abstraction of meaning as the hierarchy goes up to the left and a higher specificity of meaning as the hierarchy goes down to the right. For example, in the term 'customer', the noun 'customer' is given the meaning of 'person', where all of 'biology> animal> human' including the upper meaning of 'person' is tagged.

The entity name recognition unit 130 receives a string from the semantic analysis unit 120, and the entity name for a set of words or a set of words that can be classified such as a name of a person, an institution, a song name, a broadcast name, or a place name. Tag (named entity: NE). For example, in the string "Buddhism is a religion that many believes in Asian cultures," the individual names 'OGG_RELIGION' for 'Buddhism' and 'LCG_CONTINENT' for 'Asia' are tagged. The entity name recognition unit 130 is well known, and named entity processing is known as an important step in language analysis.

The chunking unit 140 receives a string recognized by the entity name recognition unit 130 and receives a chunking (for example, recognizes a compound noun, a formula clause, etc. in the form of a chunk), and the meaning of the noun in the chunk or Assign a concept to each chunk based on the entity name. One concept assigned to the chunk is one of a plurality of semantic codes having a hierarchical semantic structure as shown in Table 1. When the meaning of a noun in a chunk is the basis for defining the concept of a chunk, one of a plurality of semantic codes having a hierarchical semantic structure can be directly assigned and used as the concept of a chunk. If the entity name in the chunk is the basis for defining the concept of the chunk, the concept of the chunk may be determined using a mapping table that maps the entity name and semantic code. For example, assuming that the entity name "OGG_Business" is in the chunk, the entity code "OGG_Business" is mapped to the meaning code "company", and the meaning code "company" is assigned to the concept of the chunk.

The substitute term restoring unit 150 selects a chunk (hereinafter, referred to as a reference object) to be used for restoring a term by comparing and analyzing the whole meaning tagged with the term and the concept assigned to the chunk. The reason for using a chunk as a reference object is that a reference object indicated by a substitute term may be a single entity name but may be a compound noun or a whole clause. First, if there is a chunk having a concept included in the overall meaning tagged in the substitute term, the substitute term restoration unit 150 selects the chunk as a candidate candidate. The term recovery unit 150 selects a reference object by comparing and ranking chunks selected as target candidates. Hereinafter, a method of selecting a reference target by ranking the selected target candidates will be described in detail. First, as the concept of the target candidate is lower in the overall meaning tagged in the substitute term, the greater weight is given and the higher the target candidate is in ranking. For example, if the overall meaning tagged in the terminology is 'shape> style> culture> religion', the 'religion' at the bottom of 'shape> style> culture> religion' is the most specific meaning. Higher weights are given to target candidates with concepts that match '. On the contrary, since 'shape', which is the highest meaning of 'shape> style> culture> religion', is the most abstract meaning, low weight is given to the target candidate having the concept of 'shape'. If the weights of the target candidates are the same after the above-described process, the recently inputted string is given priority. If the same is the same, the essential components (eg, subject and object) of the sentence are given priority. A reference object will be selected.

Hereinafter, a process of referencing a substitute term using a semantic code having a hierarchical semantic structure in the term reference solving apparatus according to the present invention will be described in detail with reference to the following.

3 is a view for explaining a substitute term reference solution according to the present invention.

First, assuming that a user receives a string of "Buddhism is a religion that is much believed in Asian culture. It has rules that prohibit killing."

The substitute term recognition unit 100 receives the character string, performs morpheme analysis, and performs tagging of parts of speech information corresponding to each morpheme (S10). the results are as follow.

"Buddhism / nc is much in Asia / nc culture / nc / mag believe / pv is religion / nc. He / mm religion / nc prohibits killing / nc / pv has discipline / nc / pv (table 2).

Tier 1 Tier 2 Tier 3 1.s (symbol): 21 2. f (foreign language) 3.n (noun) 3.1 nc (independent noun): 0 3.2 nb (dependent noun): 2 4. np (pronoun): 1 5.nn (investigation): 3 6.pv (verb): 4 7.pa (adjective): 6 8.px (22): 22 Co (designated company): 13 10. ma (adverb) 10.1 mag (general adverb): 8 10.2 maj (connection adverb): 27 11.mm (tubular thread): 23 12.ii (interjection): 11 13.x (macro) 13.1 xp (prefix): 16 13.2 xs (suffix) 13.2.1 xsn (noun-derived suffix): 20, 17 13.2.2 xsv (verb-derived suffix): 18 13.2.3 xsm (adjective derived suffix): 19 14.j (investigation) 14.1 jc: 12 14.2 jx: 28 14.3 jj (connection survey): 25 14.4 jm: 26 Ep (fresh ending): 15 16.e (the ending) 16.1 ef (end ending): 14 16.2 ec (connection mother): 24 16.3 et 16.3.1 etn: 30 16.3.2etm (tubular mother): 29 17.uk (unregistered word): 31 18.nk (user pre-registered noun): 32 19.nr (unique noun): 33

Subsequently, the terminology recognition unit 110 is a string tagged with the part-of-speech information for each morpheme "Buddhism / nc is a lot / mag believe / pv is a religion / nc in Asia / nc culture / nc. / Mm religion / nc Is banning killing / nc / pv has discipline / nc ”, and recognizes that he / mm religion / nc consists of 'directive + noun' as a substitute (S20).

The semantic analysis unit 120 grasps the entire meaning 'shape> style> culture> religion' of 'mm / mm religion / nc' based on a plurality of semantic codes having a hierarchical meaning structure as shown in Table 1. Then, the entire meaning 'shape> style> culture> religion' is tagged with the term 'he / mm religion / nc' (S30).

"Buddhism / nc is a lot in Asia / nc culture / nc / mag believe / pv is religion / nc. He / mm religion / nc [shape>style>culture> religion] is killing / nc ban / pv is discipline / I have nc. "

Subsequently, the entity name NE is tagged in the character string through the entity name recognition unit 130 (S40). In the string "Buddhism is a religion that is much believed in Asian culture. It has rules that prohibit killing." In Buddhism and Asia, individual names are tagged as follows.

"<Buddhism: OGG _ RELIGION > is a religion that is much believed in <Asia: LCG _ CONTINENT > culture. It has rules that prohibit killing."

Subsequently, the chunking unit 140 chunks the string as "Buddhism / In Asian Culture / Much / Believe / Religion" and assigns a concept to each chunk as follows (S50). .

"[<Buddhism: OGG_RELIGION> is: Religion ] [<Asia: LCG_CONTINENT> In culture: Form ] [Much] [Believe] [Religion: Religion]"

Subsequently, the term restoring unit 150 compares the overall meaning (shape> style> culture> religion) tagged with the terminology through the semantic analysis unit 120 with the concept assigned to each chunk, and among the concepts of the chunk. If there is something corresponding to the whole meaning tagged in the substitute term, it is selected as the target candidate (S60). In operation S70, the substitute word restoration unit 150 selects a reference object by comparing and ranking chunks selected as target candidates.

As the concept of the target candidate is lower in the overall meaning tagged in the substitute term, the higher weight is assigned, and the higher candidate target is ranked during ranking. Then, the substitute term is restored using the chunk having the highest ranking (S80). In other words, 'religion', the lowest meaning among the hierarchical meanings assigned to substitute terms, is the most specific meaning. Thus, chunks with concepts that match 'religion' are given high weights. On the contrary, 'shape', the highest meaning among the hierarchical meanings assigned to substitute terms, is the most abstract meaning. Thus, a low weight is given to chunks that have a concept that matches 'shape'. If there is a chunk given the same weight, the target candidate included in the recently inputted string is given priority, and under the same conditions, the essential components of the sentence like the subject or object are given priority over the expression clause.

Applying this rule to the above example, "[<Buddhism: OGG_RELIGION>: Religion ]" is given a greater weight than "[<Asia: LCG_CONTINENT> Culture: Style ]" and is selected for reference.

Accordingly, restoring a substitute word included in the input string is as follows.

"Buddhism is a religion that is much believed in Asian culture. Buddhism has rules that prohibit killing."

While the preferred embodiments of the present invention have been shown and described, the present invention is not limited to the specific embodiments described above, and the present invention is not limited to the specific embodiments of the present invention, without departing from the spirit of the invention as claimed in the claims. Various modifications can be made by those skilled in the art, and these modifications should not be individually understood from the technical spirit or the scope of the present invention.

Claims (10)

A substitute word recognition unit that receives a string and recognizes a substitute word; A semantic analysis unit for analyzing the overall meaning of the substitute term; An entity name recognition unit recognizing an entity name of the character string; A chunking unit that receives the string recognized by the entity name and assigns a concept to each chunk after chunking; And a substitute term restoring unit for comparing and comparing the whole meaning of the terminology with a concept assigned to each chunk to select a reference object to be used for restoring the term. The method according to claim 1, The terminology recognizing unit, the terminology reference solving apparatus, characterized in that for recognizing the 'directive + correspondence' in the input string. The method according to claim 1, And the semantic analysis unit grasps the overall meaning of the substitute term based on a plurality of semantic codes having a hierarchical semantic structure. The method according to claim 1, The chunk assigns a concept to each chunk based on the meaning of an entity name or a noun in each chunk, and the concept assigned to each chunk is one of a plurality of semantic codes having a hierarchical semantic structure. Terminology reference solving device, characterized in that one. A substitute word recognition step of recognizing a substitute word included in the string by analyzing the input string; A semantic analysis step of analyzing the overall meaning of the substitute term; A target candidate selection step of chunking the string and assigning a concept to each chunk, and comparing and analyzing the concept and the overall meaning to select target candidates to which the substitute term can indicate; And a term restoring step of ranking the candidate candidates to select a reference object to be used for the term restoration and restoring the term using the reference object. The method according to claim 5, The term recognition step, A grammatical analysis of the string to recognize the term "directive + correspondence" as a substitute, reference word resolution method characterized in that. The method according to claim 5, The semantic analysis step, Alternative word reference solution characterized in that for analyzing the overall meaning of the substitute terms using a plurality of semantic codes having a hierarchical meaning structure. The method according to claim 5, The target candidate selection step, Recognizing the entity name of the character string; And Assigning a concept to each chunk based on the meaning of the noun in each chunk or the individual name. The method according to claim 5, Terminology restoration step, The higher term weighting of the target candidate corresponds to a lower meaning in the overall meaning of the substitute term, characterized in that the ranking (ranking). The method according to claim 9, If there are a plurality of target candidates that are given the same weight, the term reference solution method of claim 1, wherein the target candidate included in the character string having a later input time is given priority.