CN107402933A - Entity polyphone disambiguation method and entity polyphone disambiguation equipment - Google Patents

Abstract

A kind of entity polyphone disambiguation method and entity polyphone disambiguation equipment is disclosed, the wherein entity polyphone disambiguation method includes：Entity recognition step, for identifying at least one entity including polyphone from the text of input；And determine pronunciation step, for each entity at least one entity, by the corresponding entity in the data set of the entity link to the open data of association, and its property value based on corresponding entity includes at least one attribute of pronunciation and/or the pronunciation associated with corresponding entity, determines the pronunciation of the entity.In accordance with an embodiment of the present disclosure, the pronunciation of entity can be found out from the open data of association, disambiguation is carried out so as to the pronunciation to entity polyphone.

Description

Entity polyphonic character disambiguation method and entity polyphonic character disambiguation device

technical field

The present disclosure relates to the field of information processing, and more specifically, relates to a method for disambiguating entity polyphonic characters and a device for disambiguating entity polyphonic characters. Disambiguate.

Background technique

TTS (Text To Speech) technology, also known as text-to-speech technology, is a representative content of current speech synthesis, which refers to the technology of converting arbitrary text into speech by using a computer. Because the input text needs to be converted into the corresponding pronunciation, disambiguation of polyphonic characters is the core issue of text-to-speech conversion. Whether the conversion of polyphonic characters is correct or not greatly affects the user's comprehension of synthesized speech. If the accuracy of polyphone disambiguation is high, the synthesized language is easier for users to understand and sounds more natural and fluent.

In Chinese or Japanese, there are a large number of polyphonic characters, so how to determine the pronunciation of polyphonic characters has become a problem that needs to be solved in the field of speech synthesis of Chinese or Japanese texts. At present, the disambiguation of polyphonic characters mainly includes two methods: one is to manually summarize and formulate rules; the other is to use machine learning to disambiguate polyphonic characters. Among them, the method based on artificial rules is labor-intensive, and in some cases, the pronunciation of polyphonic characters has no rules to follow, and people cannot judge their pronunciation. For example, in Japanese, the same Chinese character may be pronounced differently in different people's names . In machine learning methods, a large amount of manually labeled corpus is often required, which is also time-consuming and labor-intensive.

Linked Data (Linked Data) is a series of best practice methods that use the Web to create semantic associations between different data sources. Linked Data uses Uniform Resource Identifiers (URI) to identify resources (understandable as entities), so each entity Unique, while also providing per-resource metadata in the form of triplets, that is, associated attributes and attribute values. The development and release of linked data on the Internet is called linked open data (Linked Open Data, LOD). Commonly used large-scale LOD datasets include DBpedia, Freebase, etc. For example, DBpedia is a structured dataset of Wikipedia. When people edit the Wikipedia page of an entity, they often give its pronunciation at the same time, but it is not given in a fixed way. In LOD, some resources have attributes like pronunciation, and each resource has a unique identifier. Therefore, we consider that LOD can be used to disambiguate polyphone resources.

Contents of the invention

A brief summary of the present disclosure is given below in order to provide a basic understanding of some aspects of the present disclosure. It should be understood, however, that this summary is not an exhaustive summary of the disclosure. It is not intended to identify key or critical parts of the disclosure, nor is it intended to limit the scope of the disclosure. Its sole purpose is to present some concepts of the disclosure in a simplified form as a prelude to the more detailed description that is presented later.

In view of the above problems, the purpose of this disclosure is to provide a method for disambiguating entity polyphonic characters and a device for disambiguating entity polyphonic characters, which can find out the pronunciation of entities from associated open data as the disambiguation result of entity pronunciation, so as to be able to disambiguate entities The pronunciation of polyphonic characters is disambiguated.

According to an aspect of the present disclosure, a method for disambiguating entity polyphonic characters is provided, including: an entity recognition step, which can be used to identify at least one entity including polyphonic characters from the input text; and a determining pronunciation step, which can be used for all each of the at least one entity, the entity is linked to a corresponding entity in the dataset of associated open data, and may contain at least one attribute of an utterance and/or an utterance associated with the corresponding entity based on its attribute value of the corresponding entity , which determines the pronunciation of the entity.

According to another aspect of the present disclosure, there is also provided an entity polyphone disambiguation device, including: an entity recognition unit configured to identify at least one entity including a polyphone from an input text; and a pronunciation determination unit, Can be configured such that, for each entity in the at least one entity, the entity is linked to a corresponding entity in the data set of associated open data, and at least one attribute of the pronunciation can be included based on its attribute value of the corresponding entity and/or related to The pronunciation associated with the corresponding entity, determines the pronunciation of the entity.

According to other aspects of the present disclosure, there are also provided computer program codes and computer program products for realizing the above-mentioned methods according to the present disclosure, and computer-readable computer program codes on which the computer program codes for realizing the above-mentioned methods according to the present disclosure are recorded. storage medium.

Further aspects of embodiments of the present disclosure are given in the following descriptive section, wherein the detailed description serves to fully disclose preferred embodiments of the embodiments of the present disclosure without imposing limitations thereon.

Description of drawings

The present disclosure can be better understood by referring to the following detailed description given in conjunction with the accompanying drawings, wherein the same or similar reference numerals are used throughout to designate the same or similar parts. The accompanying drawings, together with the following detailed description, are incorporated in and form a part of this specification, and serve to further illustrate preferred embodiments of the present disclosure and explain principles and advantages of the present disclosure. in:

Fig. 1 is a flow chart showing an example of the process of a method for disambiguating entity polyphonic characters according to an embodiment of the present disclosure;

FIG. 2 is a diagram illustrating an example of an entity in a dataset of associated open data;

FIG. 3 is a diagram illustrating an example of another entity in a dataset of associated open data;

Figure 4 is a diagram illustrating an example of yet another entity in a data set of associated open data;

FIG. 5 is a block diagram showing an example of a functional configuration of an entity polyphone disambiguation device according to an embodiment of the present disclosure; and

FIG. 6 is a block diagram showing an example structure of a personal computer as an information processing device employable in an embodiment of the present disclosure.

detailed description

Exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. In the interest of clarity and conciseness, not all features of an actual implementation are described in this specification. It should be understood, however, that in developing any such practical embodiment, many implementation-specific decisions must be made in order to achieve the developer's specific goals, such as meeting those constraints related to the system and business, and those Restrictions may vary from implementation to implementation. Moreover, it should also be understood that development work, while potentially complex and time-consuming, would at least be a routine undertaking for those skilled in the art having the benefit of this disclosure.

Here, it should be noted that in order to avoid obscuring the present disclosure due to unnecessary details, only the device structure and/or processing steps closely related to the solution according to the present disclosure are shown in the drawings, and the Other details that are not materially relevant to the present disclosure are omitted.

Embodiments according to the present disclosure will be described in detail below with reference to the accompanying drawings.

First, a flow example of a method for disambiguating entity polyphonic characters according to an embodiment of the present disclosure will be described with reference to FIG. 1 . Fig. 1 is a flow chart showing an example of the procedure of a method for disambiguating entity polyphonic characters according to an embodiment of the present disclosure. As shown in FIG. 1 , the method for disambiguating entity polyphonic characters according to an embodiment of the present disclosure may include an entity recognition step S102 and a pronunciation determination step S104.

First, in the entity recognition step S102, at least one entity including polyphonic characters may be recognized from the input text.

Specifically, in the entity recognition step S102, entities in the input text may be recognized using named entity recognition technology. But this is only an illustration rather than a limitation, and those skilled in the art can also use other techniques to identify entities in the input text. For example, in the Japanese sentence "The World's Strongest Player が集うATPツアー・ファイナルに麦维基(Nissin Foods) が初见", the name "Nishinori Kei" and the organization name "Nissin Foods" can be recognized, among which "Nishin Gui" is an entity that includes the polyphonic word "Nishikiri".

In determining the pronunciation step S104, for each entity in the at least one entity, the entity may be linked to a corresponding entity in the data set of associated open data, and at least one attribute of the pronunciation may be included based on its attribute value of the corresponding entity and/or the pronunciation associated with the corresponding entity, determining the pronunciation of the entity.

In this disclosure, by way of example and not limitation, the dataset for LOD is DBpedia. In addition, the data set of LOD can also be Freebase and so on.

For the "Kei Nishikori" and "Nissin Foods" identified in the entity recognition step S102, these entities can be linked to corresponding entities in the LOD dataset using entity linking technology. For example, "Kei Nishikori" may be linked to the corresponding entity "http://ja.dbpedia.org/resource/Kei Nishikori" in DBpedia, and may contain at least one attribute of pronunciation based on its attribute value of the corresponding entity and/or Or the pronunciation associated with the corresponding entity, determine the pronunciation of the entity "Kei Nishikori". In addition, "Nissin food" may be linked to the corresponding entity "http://ja.dbpedia.org/resource/Nissin food" in DBpedia, and at least one attribute of pronunciation may be contained based on its attribute value of the corresponding entity and/or the pronunciation associated with the corresponding entity, determine the pronunciation of the entity "Nissin Foods". Because each entity in the LOD is unique, the resulting pronunciations are also unambiguous.

Preferably, the at least one attribute may include at least one first predetermined attribute whose attribute value is directly the pronunciation. For the entities in the LOD dataset, the pronunciation exists in the attribute value. In some cases, the attribute value is directly the pronunciation of the entity. FIG. 2 is a diagram showing an example of one entity in a dataset of associated open data. Specifically, FIG. 2 shows the entity "http://ja.dbpedia.org/resource/Atsushi Nishikori" in the dataset of associated open data. For example, the property value "にしこおりあつし" of the property "http://ja.dbpedia.org/property/national language notation" shown in FIG. 2 is directly the pronunciation of "Nishikori Jun". FIG. 3 is a diagram illustrating an example of another entity in a data set associating open data. Specifically, FIG. 3 shows the entity "http://ja.dbpedia.org/resource/Nishikori Ichiki" in the dataset of associated open data. For example, the attribute value "にしきおかずきよ" of the attribute "http://xmlns.com/foaf/0.1/name" shown in Figure 3 is directly the pronunciation of "Nishikori Yiqing"; in addition, the attribute shown in Figure 3 The property value "にしきおりずきよ" of "http://ja.dbpedia.org/property/ふりがな" is also the pronunciation of "Nishikiri Ichiki". However, these attributes representing pronunciation in different entities are often different, and some general attributes do not always represent pronunciation, so we need to select these attributes.

Preferably, the at least one first predetermined attribute may be obtained by obtaining the name of each entity in the LOD data set, and listing all the pronunciations of each character in the entity's name in the dictionary All the pronunciations of the entity are used as candidate pronunciations. If there is an attribute whose attribute value completely matches any of the candidate pronunciations of the entity in the attributes of the entity, this attribute is selected as a candidate attribute, and in the LOD Among all candidate attributes selected by all entities in the data set, at least one candidate attribute whose probability of representing pronunciation is greater than a predetermined threshold is selected as at least one first predetermined attribute.

Specifically, in order to select the attribute representing the pronunciation, we first obtain the name of each entity in the LOD dataset, and list all possible pronunciations of each name as candidate pronunciations according to all the pronunciations of each word in the dictionary in the name . Compare the candidate pronunciation of the entity with the attribute value of the entity one by one. If one of the candidate pronunciations corresponds to one of the attribute values, select the corresponding attribute as the candidate attribute, and the attribute value is the pronunciation of the entity. Then, for all candidate attributes selected from all entities in the LOD dataset, the probability that they represent pronunciation is calculated. If the probability of the candidate attribute representing the pronunciation is greater than the predetermined threshold, then retain the candidate attribute as a first predetermined attribute, that is, select at least one candidate attribute whose probability of representing the pronunciation is greater than the predetermined threshold as at least one first predetermined attribute.

Preferably, the probability of the candidate attribute representing pronunciation may be the ratio of the number of times the attribute value of the candidate attribute is pronounced to the total number of occurrences of the candidate attribute in the LOD data set.

If the candidate attribute is represented by a, the probability P(a) of the candidate attribute a representing the pronunciation is the ratio of the number of times the attribute value of the candidate attribute a is pronounced to the total number of occurrences of the candidate attribute a in the LOD data set, such as the formula (1 ) as shown:

P(a) = the number of times the attribute value of a is pronounced / the total number of occurrences of attribute a (1)

Preferably, the aforementioned predetermined threshold can be determined by those skilled in the art based on experience or experiments.

According to an embodiment of the present disclosure, for the example shown in FIG. 2 , it can be determined that the attribute value "にしこおりあつし" of the attribute "http://ja.dbpedia.org/property/national language representation" is directly the value of "Nishikori Jun". pronounce.

According to an embodiment of the present disclosure, for the example shown in FIG. 3 , it can be determined that the attribute value "にしきおりずきよ" of the attribute "http://xmlns.com/foaf/0.1/name" is directly the value of "Nishikori Ichiki". pronounce. In addition, it can also be determined that the attribute value "にしきおりずきよ" of the attribute "http://ja.dbpedia.org/property/ふりがな" is directly the pronunciation of "Nishikiri Ichiki".

Preferably, the at least one attribute may further include at least one second predetermined attribute whose attribute value contains a pronunciation that can be extracted by using at least one pronunciation extraction template. In addition to the fact that the attribute value is directly the pronunciation of the entity, in some cases the pronunciation is included in the attribute value, and generally the position of occurrence is regular. At this time, we can use some pronunciation extraction templates to determine the pronunciation of the entity. FIG. 4 is a diagram illustrating an example of still another entity in a dataset of associated open data. Specifically, FIG. 4 shows the entity "http://ja.dbpedia.org/resource/Kei Nishikori" in the dataset of associated open data. Unlike the examples shown in Figures 2 and 3, the attributes in Figure 4 do not include attributes whose attribute values are directly the pronunciation of the entity, that is, the pronunciation of "Kei Nishikori" cannot be directly determined from the attribute values of the attributes in Figure 4 . However, the pronunciation of "Kei Nishikori" is included in the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment". In this case we can obtain the pronunciation of the entity in Figure 4 by utilizing the pronunciation extraction template.

Preferably, the at least one pronunciation extraction template can be generated in the following manner: for each entity in the data set of LOD that includes any first predetermined attribute in the at least one first predetermined attribute, according to the entity's The attribute value of any first predetermined attribute determines the pronunciation of the entity, and the regularity of the occurrence position of the pronunciation in the attribute values of other attributes containing the pronunciation of the entity can be determined, so that according to the LOD data set including all All entities of any first predetermined attribute in the at least one first predetermined attribute are used to automatically generate the at least one pronunciation extraction template.

In the LOD data set, in some cases, the pronunciation is not only directly an attribute value of a certain attribute, but also exists in other attribute values, that is to say, the other attribute values also include the pronunciation of the entity. And in some cases, the pronunciation of the entity is not directly the attribute value, but the pronunciation is only included in the attribute value and its occurrence position in the attribute value is regular. In this case we can automatically generate some pronunciation extraction templates to match these pronunciations.

To generate a pronunciation extraction template, the training template corpus must first be collected. Specifically, for each attribute in the attribute list composed of the obtained at least one first predetermined attribute, each entity containing the attribute is searched in the LOD data set, and according to the attribute value of the attribute of the entity Determine the pronunciation of the entity, and look for other attribute values containing its pronunciation for the entity as a template training corpus. For example, for the entity "http://ja.dbpedia.org/resource/Nishikori Jun" shown in FIG. 2, the property "http://ja.dbpedia.org/property/national language token" is directly pronounced attribute, so we can see that this entity is pronounced as "にしこおりあつし", and then check other attribute values, and find that the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" contains pronunciation "にしこおりあつし", the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" is used as the template training corpus. In addition, for the entity "http://ja.dbpedia.org/resource/Nishikori Ichiki" shown in FIG. 3, the attribute "http://xmlns.com/foaf/0.1/name" is an attribute directly pronounced , it can be seen that this entity is pronounced as "にしきおりずきよ", and then check other attribute values, and find that the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" contains pronunciation "にしきおりずきよ", the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" is also used as the template training corpus.

Secondly, it is necessary to determine the regularity of the occurrence position of the pronunciation in the attribute values of other attributes of the entity, so as to automatically generate the pronunciation extraction template. For each training corpus, characters whose window length is N before and after the pronunciation string in the training corpus can be intercepted, and the numbers and Chinese characters in the intercepted N characters can be generalized, so as to determine the pronunciation in other entities containing pronunciation Occurrences in the attribute value of an attribute, from which a candidate template is generated. For example, as mentioned above, the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" shown in Figure 2 is a template training corpus, for which the training corpus can be intercepted including The character string of the pronunciation string "Nishikiri Jun (にしこおりあつし, July 30, 1945 -) は", generalize the numbers and Chinese characters in the intercepted string, so as to determine the pronunciation "にしこおりあつし" in the attribute occurrence in the attribute value of "http://www.w3.org/2000/01/rdf-schema#comment", thereby generating a candidate template. In addition, as mentioned above, the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" shown in Figure 3 is also a template training corpus, for which the training corpus can be intercepted The character string "Nishikori Ichiyo (にしきおりずきよ, May 22, 1965-)" including the pronunciation character string, the numbers and Chinese characters in the intercepted character string are generalized to determine the pronunciation "にしきおりずきよ" in the attribute value of attribute "http://www.w3.org/2000/01/rdf-schema#comment", thereby generating a candidate template. In order to simplify the description, only the entities shown in Figure 2 and Figure 3 are described above as examples, but in fact, all entities including any first predetermined attribute in the at least one first predetermined attribute in the LOD data set can be entity to perform the above processing.

In addition, for each entity found that includes the first predetermined attribute, generalize the characters in the attribute values of the entity's other attributes that include pronunciation, and extract the common subset of sentences after the generalization, Therefore, the occurrence position of the pronunciation in the attribute values of other attributes including the pronunciation of the entity is determined, thereby generating a candidate template. Still taking Fig. 2 and Fig. 3 as an example, the attribute value "Nishikori Jun(にしこおりあつし, July 30, 1945 -) は, Japan's の弗 nurse and politician. Member of the Yuan House of Representatives (1st period). Born in Izumo City, Shimane Prefecture (Old Hirata City)." Generalization, thus obtaining a generalized structure . In addition, the attribute value "Nishikori Ichiki (にしきおりずきよ, May 22, 1965) of the attribute "http://www.w3.org/2000/01/rdf-schema#comment" shown in FIG. -)はくるるグーズににののるるるープトリーダートスカグループ「Youth Team」のリーダー. Nickname は, Fankara は "ニッキ", メンバーティク "ニシキ".にオーディションを得け for the 5th year of elementary school, and entered the office in July 1977 for the 6th year of Edogawa Ward Hirai Minami Elementary School. Born in Tokyo. Youth Team のイメージカラーは红. "To generalize, and thus obtain a generalized structure. Then, extract the common subset of the above two generalized structures, so as to determine the regularity of the occurrence position of the pronunciation, thereby generating a candidate template In order to simplify the description, the above only describes the entity shown in Figure 2 and Figure 3 as an example, but in fact, according to the LOD data set that includes any first predetermined attribute in the at least one first predetermined attribute All entities to do the above processing.

Finally, all the generated candidate templates are sorted, and the candidate templates whose occurrence times are greater than a certain threshold are selected as the final pronunciation extraction templates. Thus, the at least one pronunciation extraction template may be automatically generated according to all entities in the LOD data set that include any first predetermined attribute in the at least one first predetermined attribute.

Preferably, the attribute value of the at least one first predetermined attribute and the attribute value of the at least one second predetermined attribute may be attribute values of a character string type.

Preferably, for each entity in the at least one entity, if the attribute of the corresponding entity to which the entity is linked contains one of the at least one first predetermined attribute, the one first predetermined attribute can be The attribute value of a predetermined attribute is used as the pronunciation of the entity, and if the attribute of the corresponding entity to which the entity is linked does not contain any one of the at least one first predetermined attribute, the at least one predetermined attribute may be used. A pronunciation extraction template to determine the pronunciation of the entity.

Specifically, check whether the attribute of the corresponding entity to which the entity is linked contains any first predetermined attribute, and if so, that is, the corresponding entity contains an attribute whose attribute value is directly pronounced, then any one of the first predetermined attributes can be taken The attribute value of is used as the pronunciation of the entity. Taking the entity "Jinzhichun" including polyphonic characters as an example, its corresponding entity is "http://ja.dbpedia.org/resource/Jinzhichun" as shown in Figure 2, because the corresponding entity contains its attribute value, which is directly the pronunciation The attribute "http://ja.dbpedia.org/property/International language notation", so you can directly get the attribute value "にしこおりあつし" of the attribute "http://ja.dbpedia.org/property/International language notation" As the pronunciation of "Nishikori Jun". In addition, taking the entity "Nishikori Ichiyo" including polyphonic characters as an example, its corresponding entity is the entity "http://ja.dbpedia.org/resource/Nishikori Ichiyo" shown in Figure 3, because the corresponding entity contains Its attribute value is directly the pronounced attribute "http://xmlns.com/foaf/0.1/name" and "http://ja.dbpedia.org/property/ふりがな", so the attribute "http:// xmlns.com/foaf/0.1/name" and "http://ja.dbpedia.org/property/ふりがな" attribute value "にしきおりずきよ" is used as the pronunciation of "Nishikori Ichiki".

And if the attribute of the corresponding entity to which the entity is linked does not contain any first predetermined attribute, that is, the corresponding entity does not contain an attribute whose attribute value is the pronunciation directly, then the pronunciation extraction template can be used to determine the pronunciation of the entity. For the entity "http://ja.dbpedia.org/resource/Nishikori Kei" shown in Figure 4, since the attributes of the corresponding entity do not contain attributes whose attribute values are directly pronunciation, it is necessary to use the pronunciation extraction template to determine the pronunciation .

Preferably, using the at least one pronunciation extraction template to determine the pronunciation of one entity in the at least one entity may include: using the at least one pronunciation extraction template to match at least one of the corresponding entities to which the one entity is linked. The attribute value of the character string type of the attribute, and the matched character string is used as the pronunciation of the entity.

Specifically, if the attribute of the corresponding entity to which the entity is linked does not contain any first predetermined attribute, the pronunciation extraction template may be used to match the attribute value of the string type of the corresponding entity, and the matched character string may be used as The pronunciation of this entity.

Taking the entity "Nishikori Kyu" including polyphonic characters as an example, its corresponding entity is "http://ja.dbpedia.org/resource/Nishikori Kyu" shown in Figure 4, since the attribute of the corresponding entity does not contain any first For a predetermined attribute (that is, the attribute of the corresponding entity does not include the attribute whose attribute value is the pronunciation directly), it is necessary to use the pronunciation extraction template to determine the pronunciation. Specifically, using the above pronunciation extraction template generated by taking the attribute value in the entity shown in Figure 2 and Figure 3 as an example of the template training corpus, the entity "http://ja.dbpedia.org/resource/Nishikori Kei" Match the attribute value of the string type in , assuming that when using the pronunciation extraction template to match the attribute value of the attribute "http://www.w3.org/2000/01/rdf-schema#comment", the string "にしこり" is matchedけい", the matched character string "にしこりけい" can be used as the pronunciation of the entity "Kyu Nishikori".

It has been described in detail above that the pronunciation of an entity is determined based on at least one attribute of the corresponding entity whose attribute value contains the pronunciation.

In addition, as described in the entity recognition step S102, the utterances of the entities may be determined based on the utterances associated with the corresponding entities. Specifically, a separate pronunciation attribute of the corresponding entity may be established, and after the pronunciation is found in the LOD data set as described above, the found pronunciation may be stored in the attribute value of the separate pronunciation attribute. For example, for the entity "http://ja.dbpedia.org/resource/Nishikori Jun" shown in Figure 2, a separate pronunciation attribute is established, and the pronunciation "にしこおりあつし" is used as the attribute value of its "pronunciation attribute". For the entity "http://ja.dbpedia.org/resource/Nishikori Ichiyo" shown in Figure 3, a separate pronunciation attribute is established, and the pronunciation "にしきおりずきよ" is used as the attribute value of its "pronunciation attribute". For the entity "http://ja.dbpedia.org/resource/Nishikori Kei" shown in Figure 4, a separate pronunciation attribute is established, and the pronunciation "にしこりけい" is used as the attribute value of its "pronunciation attribute". These pronunciation attributes and their attribute values can not only be stored locally, but also published on the network. In this way, after the entity is linked to the corresponding entity in the LOD dataset, the pronunciation of the entity can be obtained by querying the attribute value of the "pronunciation attribute" of the corresponding entity. For example, if the entity "Atsushi Nishikori" is identified from the input text, after linking this entity to the corresponding entity "http://ja.dbpedia.org/resource/Atsushi Nishikori" in the LOD's dataset, by querying the According to the attribute value of the "pronunciation attribute" of the corresponding entity, the pronunciation "にしこおりあつし" of "Nishikori Jun" can be obtained. Also, if the entity "Kei Nishikori" is recognized from the input text, after linking this entity to the corresponding entity "http://ja.dbpedia.org/resource/Kei Nishikori" in the LOD's dataset, by querying the According to the attribute value of the "pronunciation attribute" of the corresponding entity, the pronunciation "にしこりけい" of "Nishikori Kei" can be obtained.

According to the above description, the entity polyphonic character disambiguation method according to the embodiment of the present disclosure links the entity polyphonic character whose pronunciation is to be determined to the corresponding entity of the LOD, and finds the pronunciation from the related attribute value of the corresponding entity. Because each entity in the LOD is unique, the obtained pronunciation is also unambiguous, so as to disambiguate the pronunciation of entity polyphonic characters.

Corresponding to the above method embodiments, the present disclosure also provides the following device embodiments.

FIG. 5 is a block diagram showing an example of a functional configuration of an entity polyphone disambiguation device 500 according to an embodiment of the present disclosure.

As shown in FIG. 5 , an entity polyphonic word disambiguation device 500 according to an embodiment of the present disclosure may include an entity recognition unit 502 and a pronunciation determination unit 504 . A functional configuration example of each unit will be described next.

In the entity recognition unit 502, at least one entity including polyphonic characters may be recognized from the input text.

For the specific method of identifying entities including polyphonic characters from the input text, refer to the descriptions at corresponding positions in the above method embodiments, and will not be repeated here.

In determining the pronunciation unit 504, for each entity in the at least one entity, the entity may be linked to the corresponding entity in the data set of the associated open data, and at least one attribute of the pronunciation may be included based on its attribute value of the corresponding entity and/or the pronunciation associated with the corresponding entity, determining the pronunciation of the entity.

For the entities identified in the entity identification unit 502, these entities can be linked to the corresponding entities in the LOD dataset by using the entity linking technology. And the utterance may be determined based on at least one attribute of the corresponding entity whose attribute value contains an utterance and/or an utterance associated with the corresponding entity. Because each entity in the LOD is unique, the resulting pronunciations are also unambiguous.

Preferably, the at least one attribute may include at least one first predetermined attribute whose attribute value is directly the pronunciation. For the entities in the LOD dataset, the pronunciation exists in the attribute value. In some cases, the attribute value is directly the pronunciation of the entity. However, these attributes representing pronunciation in different entities are often different, and some general attributes do not always represent pronunciation, so we need to select these attributes.

Preferably, the at least one first predetermined attribute may be obtained by obtaining the name of each entity in the LOD data set, and listing all the pronunciations of each character in the entity's name in the dictionary All the pronunciations of the entity are used as candidate pronunciations. If there is an attribute whose attribute value completely matches any of the candidate pronunciations of the entity in the attributes of the entity, this attribute is selected as a candidate attribute, and in the LOD Among all candidate attributes selected by all entities in the data set, at least one candidate attribute whose probability of representing pronunciation is greater than a predetermined threshold is selected as at least one first predetermined attribute.

Specifically, in order to select the attribute representing the pronunciation, we first obtain the name of each entity in the LOD dataset, and list all possible pronunciations of each name as candidate pronunciations according to all the pronunciations of each word in the dictionary in the name . Compare the candidate pronunciation of the entity with the attribute value of the entity one by one. If one of the candidate pronunciations corresponds to one of the attribute values, select the corresponding attribute as the candidate attribute, and the attribute value is the pronunciation of the entity. Then, for all candidate attributes selected from all entities in the LOD dataset, the probability that they represent pronunciation is calculated. If the probability of the candidate attribute representing the pronunciation is greater than the predetermined threshold, then retain the candidate attribute as a first predetermined attribute, that is, select at least one candidate attribute whose probability of representing the pronunciation is greater than the predetermined threshold as at least one first predetermined attribute.

Preferably, the probability of the candidate attribute representing pronunciation may be the ratio of the number of times the attribute value of the candidate attribute is pronounced to the total number of occurrences of the candidate attribute in the LOD data set.

Preferably, the aforementioned predetermined threshold can be determined by those skilled in the art based on experience or experiments.

For a specific example of obtaining at least one first predetermined attribute, reference may be made to the descriptions at corresponding positions in the above method embodiments, which will not be repeated here.

Preferably, the at least one attribute may further include at least one second predetermined attribute whose attribute value contains a pronunciation that can be extracted by using at least one pronunciation extraction template. In addition to the fact that the attribute value is directly the pronunciation of the entity, in some cases the pronunciation is included in the attribute value, and generally the position of occurrence is regular. At this time, we can use some pronunciation extraction templates to determine the pronunciation of the entity.

Preferably, the at least one pronunciation extraction template can be generated in the following manner: for each entity in the data set of LOD that includes any first predetermined attribute in the at least one first predetermined attribute, according to the entity's The attribute value of any first predetermined attribute determines the pronunciation of the entity, and the regularity of the occurrence position of the pronunciation in the attribute values of other attributes containing the pronunciation of the entity can be determined, so that according to the LOD data set including all All entities of any first predetermined attribute in the at least one first predetermined attribute are used to automatically generate the at least one pronunciation extraction template.

In the LOD data set, in some cases, the pronunciation is not only directly an attribute value of a certain attribute, but also exists in other attribute values, that is to say, the other attribute values also include the pronunciation of the entity. And in some cases, the pronunciation of the entity is not directly the attribute value, but the pronunciation is only included in the attribute value and its occurrence position in the attribute value is regular. In this case we can automatically generate some pronunciation extraction templates to match these pronunciations.

For the specific method of generating the pronunciation extraction template, refer to the description at the corresponding position in the above method embodiment, which will not be repeated here.

Preferably, the attribute value of the at least one first predetermined attribute and the attribute value of the at least one second predetermined attribute may be attribute values of a character string type.

Preferably, for each entity in the at least one entity, if the attribute of the corresponding entity to which the entity is linked contains one of the at least one first predetermined attribute, the one first predetermined attribute can be The attribute value of a predetermined attribute is used as the pronunciation of the entity, and if the attribute of the corresponding entity to which the entity is linked does not contain any one of the at least one first predetermined attribute, the at least one predetermined attribute may be used. A pronunciation extraction template to determine the pronunciation of the entity.

Specifically, it is checked whether the attribute of the corresponding entity to which the entity is linked contains any first predetermined attribute, and if so, the corresponding entity contains an attribute whose attribute value is directly pronounced, so the value of any one of the first predetermined attributes can be taken The attribute value serves as the pronunciation of the entity. And if the attribute of the corresponding entity to which the entity is linked does not contain any first predetermined attribute, then the corresponding entity does not contain the attribute whose attribute value is the pronunciation directly, then the pronunciation extraction template can be used to determine the pronunciation of the entity.

Preferably, using the at least one pronunciation extraction template to determine the pronunciation of one entity in the at least one entity may include: using the at least one pronunciation extraction template to match at least one of the corresponding entities to which the one entity is linked. The attribute value of the character string type of the attribute, and the matched character string is used as the pronunciation of the entity.

Specifically, if the attribute of the corresponding entity to which the entity is linked does not contain any first predetermined attribute, the pronunciation extraction template may be used to match the attribute value of the string type of the corresponding entity, and the matched character string may be used as The pronunciation of this entity.

For a specific example of obtaining the pronunciation of the entity based on the attribute of the corresponding entity to which the entity is linked, refer to the description of the corresponding position in the above method embodiment, which will not be repeated here.

In addition, as described in the entity recognition unit 502, the utterances of the entities may also be determined based on the utterances associated with the corresponding entities. Specifically, a separate pronunciation attribute of the corresponding entity may be established, and after the pronunciation is found in the LOD data set, the found pronunciation may be stored in an attribute value of the separate pronunciation attribute. If a polyphonic character entity is identified from the input text, after linking the entity to the corresponding entity in the LOD dataset, the pronunciation of the polyphonic character entity can be obtained by querying the attribute value of the corresponding entity's "pronunciation attribute" .

For a specific example of determining the pronunciation of an entity based on the pronunciation associated with the corresponding entity, reference may be made to the description of the corresponding position in the above method embodiment, which will not be repeated here.

According to the above description, the entity polyphonic character disambiguation device according to the embodiment of the present disclosure links the entity polyphonic character whose pronunciation is to be determined to the corresponding entity of the LOD, and finds the pronunciation from the related attribute value of the corresponding entity. Because each entity in the LOD is unique, the obtained pronunciation is also unambiguous, so as to disambiguate the pronunciation of entity polyphonic characters.

It should be pointed out that although the functional configuration of the entity polyphonic word disambiguation device according to the embodiment of the present disclosure has been described above, this is only an example rather than a limitation, and those skilled in the art can modify the above embodiment according to the principle of the present disclosure , for example, the functional modules in each embodiment may be added, deleted or combined, and such modifications all fall within the scope of the present disclosure.

In addition, it should also be pointed out that the device embodiments here correspond to the above-mentioned method embodiments, therefore, for the content not described in detail in the device embodiments, refer to the descriptions in corresponding positions in the method embodiments, and the description will not be repeated here.

It should be understood that the machine-executable instructions in the storage medium and the program product according to the embodiments of the present disclosure may also be configured to execute the above-mentioned entity polyphonic word disambiguation method, so for the content not described in detail here, please refer to the previous corresponding position description, which will not be repeated here.

Correspondingly, a storage medium for carrying the above-mentioned program product including machine-executable instructions is also included in the disclosure of the present invention. Such storage media include, but are not limited to, floppy disks, optical disks, magneto-optical disks, memory cards, memory sticks, and the like.

In addition, it should also be noted that the series of processes and devices described above may also be implemented by software and/or firmware. In the case of realization by software and/or firmware, the program constituting the software is installed from a storage medium or a network to a computer having a dedicated hardware configuration, such as a general-purpose personal computer 600 shown in FIG. , can perform various functions and so on.

In FIG. 6 , a central processing unit (CPU) 601 executes various processes according to programs stored in a read only memory (ROM) 602 or loaded from a storage section 608 to a random access memory (RAM) 603 . In the RAM 603, data required when the CPU 601 executes various processes and the like is also stored as necessary.

The CPU 601 , ROM 602 , and RAM 603 are connected to each other via a bus 604 . The input/output interface 605 is also connected to the bus 604 .

The following components are connected to the input/output interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage section 608 , including a hard disk, etc.; and the communication part 609, including a network interface card such as a LAN card, a modem, and the like. The communication section 609 performs communication processing via a network such as the Internet.

A driver 610 is also connected to the input/output interface 605 as needed. A removable medium 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 610 as necessary, so that a computer program read therefrom is installed into the storage section 608 as necessary.

In the case of realizing the above-described series of processes by software, the programs constituting the software are installed from a network such as the Internet or a storage medium such as the removable medium 611 .

Those skilled in the art should understand that such a storage medium is not limited to the removable medium 611 shown in FIG. 6 in which the program is stored and distributed separately from the device to provide the program to the user. Examples of the removable media 611 include magnetic disks (including floppy disks (registered trademark)), optical disks (including compact disk read only memory (CD-ROM) and digital versatile disks (DVD)), magneto-optical disks (including )) and semiconductor memory. Alternatively, the storage medium may be the ROM 602, a hard disk contained in the storage section 608, or the like, in which programs are stored and distributed to users together with devices containing them.

The preferred embodiments of the present disclosure have been described above with reference to the accompanying drawings, but the present disclosure is of course not limited to the above examples. A person skilled in the art may find various alterations and modifications within the scope of the appended claims, and it should be understood that they will naturally come under the technical scope of the present disclosure.

For example, a plurality of functions included in one unit in the above embodiments may be realized by separate devices. Alternatively, a plurality of functions implemented by a plurality of units in the above embodiments may be respectively implemented by separate devices. In addition, one of the above functions may be realized by a plurality of units. Needless to say, such a configuration is included in the technical scope of the present disclosure.

In this specification, the steps described in the flowcharts include not only processing performed in time series in the stated order but also processing performed in parallel or individually and not necessarily in time series. Furthermore, even in the steps of time-series processing, needless to say, the order can be appropriately changed.

In addition, the technology according to the present disclosure may also be configured as follows.

Additional Note 1. A method for disambiguation of entity polyphonic characters, comprising:

An entity recognition step for recognizing at least one entity comprising polyphonic characters from the input text; and

determining the pronunciation step, for each entity of said at least one entity, linking the entity to a corresponding entity in the data set of the associated open data LOD, and containing at least one attribute of the pronunciation based on the value of its attribute of said corresponding entity and/or Or the pronunciation associated with said corresponding entity, determining the pronunciation of the entity.

Supplement 2. The method for disambiguating entity polyphonic characters according to Supplement 1, wherein the at least one attribute includes at least one first predetermined attribute whose attribute value is directly pronounced.

Supplementary Note 3. The method for disambiguating entity polyphonic characters according to Supplementary Note 2, wherein the at least one attribute further includes at least one second predetermined attribute whose attribute value includes at least one pronunciation that can be extracted using at least one pronunciation extraction template .

Supplement 4. The method for disambiguating entity polyphonic characters according to Supplement 2, wherein the at least one first predetermined attribute is obtained in the following manner:

Obtain the name of each entity in the dataset for the LOD;

List all pronunciations of the entity as candidate pronunciations according to all pronunciations of each word in the entity's name in the dictionary;

If there is an attribute among the attributes of the entity whose attribute value exactly matches any one of the candidate utterances of the entity, that attribute is selected as a candidate attribute; and

Among all candidate attributes selected for all entities in the data set of the LOD, at least one candidate attribute whose probability of representing an utterance is greater than a predetermined threshold is selected as the at least one first predetermined attribute.

Supplementary Note 5. The method for disambiguating entity polyphonic characters according to Supplementary Note 4, wherein the probability that the said candidate attribute represents pronunciation is the number of times the attribute value of said candidate attribute is pronounced and said candidate attribute is in the The ratio of the total number of occurrences in the data set for the LOD described above.

Supplementary Note 6. The method for disambiguating entity polyphonic characters according to Supplementary Note 3, wherein the at least one pronunciation extraction template is generated in the following manner:

For each entity in the data set of the LOD that includes any of the at least one first predetermined attribute, determining the pronunciation of the entity based on an attribute value of the entity's any first predetermined attribute;

determining the regularity of occurrences of the utterance in attribute values of other attributes of the entity that contain utterances; and

The at least one pronunciation extraction template is automatically generated according to all entities in the data set of the LOD that include any first predetermined attribute of the at least one first predetermined attribute.

Supplementary Note 7. The method for disambiguation of entity polyphonic characters according to Supplementary Note 3, wherein the attribute value of the at least one first predetermined attribute and the attribute value of the at least one second predetermined attribute are attribute values of the character string type .

Supplementary Note 8. The method for disambiguating entity polyphonic characters according to Supplementary Note 3, wherein, for each entity in the at least one entity:

If the attribute of the corresponding entity to which the entity is linked includes a first predetermined attribute of the at least one first predetermined attribute, using the attribute value of the one first predetermined attribute as the pronunciation of the entity; and

If the attribute of the corresponding entity to which the entity is linked does not contain any one of the at least one first predetermined attribute, the at least one pronunciation extraction template is used to determine the pronunciation of the entity.

Additional note 9. The method for disambiguating entity polyphonic characters according to additional note 8, wherein, using the at least one pronunciation extraction template to determine the pronunciation of an entity in the at least one entity includes:

The at least one pronunciation extraction template is used to match the attribute value of the string type of at least one attribute of the corresponding entity to which the one entity is linked, and the matched string is used as the pronunciation of the one entity.

Additional note 10. An entity polyphone disambiguation device, comprising:

an entity recognition unit configured to recognize at least one entity comprising polyphonic characters from the input text; and

Determining an utterance unit configured to, for each entity in the at least one entity, link the entity to a corresponding entity in the data set of the associated open data LOD, and contain at least one of the utterances based on the value of its attribute of the corresponding entity attributes and/or pronunciations associated with the corresponding entity, determine the pronunciation of the entity.

Supplement 11. The entity polyphonic character disambiguation device according to Supplement 10, wherein the at least one attribute includes at least one first predetermined attribute whose attribute value is directly the pronunciation.

Supplementary Note 12. The entity polyphone disambiguation device according to Supplementary Note 11, wherein the at least one attribute further includes at least one second predetermined attribute whose attribute value contains a pronunciation that can be extracted by using at least one pronunciation extraction template .

Supplement 13. The entity polyphone disambiguation device according to Supplement 11, wherein the at least one first predetermined attribute is obtained in the following manner:

Obtain the name of each entity in the dataset for the LOD;

List all pronunciations of the entity as candidate pronunciations according to all pronunciations of each word in the entity's name in the dictionary;

If there is an attribute among the attributes of the entity whose attribute value exactly matches any one of the candidate utterances of the entity, that attribute is selected as a candidate attribute; and

Among all candidate attributes selected for all entities in the data set of the LOD, at least one candidate attribute whose probability of representing an utterance is greater than a predetermined threshold is selected as the at least one first predetermined attribute.

Supplementary Note 14. The entity polyphone disambiguation device according to Supplementary Note 13, wherein the probability of the said candidate attribute representing the pronunciation is the number of times the attribute value of the candidate attribute is pronounced and the number of times the candidate attribute is pronounced. The ratio of the total number of occurrences in the data set for the LOD described above.

Supplement 15. The entity polyphone disambiguation device according to Supplement 12, wherein the at least one pronunciation extraction template is generated in the following manner:

For each entity in the data set of the LOD that includes any of the at least one first predetermined attribute, determining the pronunciation of the entity based on an attribute value of the entity's any first predetermined attribute;

determining the regularity of occurrences of the utterance in attribute values of other attributes of the entity that contain utterances; and

The at least one pronunciation extraction template is automatically generated according to all entities in the data set of the LOD that include any first predetermined attribute of the at least one first predetermined attribute.

Supplement 16. The entity polyphone disambiguation device according to Supplement 12, wherein the attribute value of the at least one first predetermined attribute and the attribute value of the at least one second predetermined attribute are attribute values of the character string type .

Supplement 17. The entity polyphone disambiguation device according to Supplement 12, wherein, for each entity in the at least one entity:

If the attribute of the corresponding entity to which the entity is linked includes a first predetermined attribute of the at least one first predetermined attribute, using the attribute value of the one first predetermined attribute as the pronunciation of the entity; and

If the attribute of the corresponding entity to which the entity is linked does not contain any one of the at least one first predetermined attribute, the at least one pronunciation extraction template is used to determine the pronunciation of the entity.

Supplement 18. The entity polyphone disambiguation device according to Supplement 17, wherein, using the at least one pronunciation extraction template to determine the pronunciation of one entity in the at least one entity includes:

The at least one pronunciation extraction template is used to match the attribute value of the string type of at least one attribute of the corresponding entity to which the one entity is linked, and the matched string is used as the pronunciation of the one entity.

Claims (10)

1. a kind of entity polyphone disambiguation method, including：

Entity recognition step, for identifying at least one reality including polyphone from the text of input Body；And

It is determined that pronunciation step, for each entity at least one entity, by the entity link To the corresponding entity in the open data LOD of association data set, and based on the corresponding entity Its property value includes at least one attribute of pronunciation and/or the pronunciation associated with the corresponding entity, Determine the pronunciation of the entity.

2. entity polyphone disambiguation method according to claim 1, wherein, described at least one It is directly at least one first predetermined attribute of pronunciation that individual attribute, which includes its property value,.

3. entity polyphone disambiguation method according to claim 2, wherein, described at least one Individual attribute also includes its property value and includes and can extract what template was extracted using at least one pronunciation At least one second predetermined attribute of pronunciation.

4. entity polyphone disambiguation method according to claim 2, wherein, described at least one Individual first predetermined attribute obtains in the following manner：

Obtain the name of each entity in the data set of the LOD；

The entity is listed in all pronunciations of each word in dictionary in the name of the entity All pronunciations are pronounced as candidate；

If its property value and any in candidate's pronunciation of the entity in the entity attributes be present The attribute that individual pronunciation matches completely, then select the attribute as a candidate attribute；And

All candidate attributes gone out selected by all entities in the data set for the LOD are worked as In, select its pronouncing probability to be more than described at least one candidate attribute conduct of predetermined threshold extremely Few first predetermined attribute.

5. entity polyphone disambiguation method according to claim 4, wherein, candidate's category Property the pronouncing probability be the candidate attribute property value for pronunciation number with it is described The ratio of appearance total degree of the candidate attribute in the data set of the LOD.

6. entity polyphone disambiguation method according to claim 3, wherein, described at least one Individual pronunciation extraction template generates in the following manner：

For including appointing at least one first predetermined attribute in the data set of the LOD Anticipate each entity of the first predetermined attribute, according to the attribute of any first predetermined attribute of the entity Value determines the pronunciation of the entity；

Determine appearance position of the pronunciation in the property value of other attributes comprising pronunciation of the entity Regularity；And

Include appointing at least one first predetermined attribute in the data set of the LOD Anticipate the first predetermined attribute all entities come automatically generate it is described it is at least one pronunciation extraction template.

7. entity polyphone disambiguation method according to claim 3, wherein, described at least one The property value of the property value of individual first predetermined attribute and at least one second predetermined attribute is character The property value of string type.

8. entity polyphone disambiguation method according to claim 3, wherein, for it is described extremely Each entity in a few entity：

If it is predetermined that the attribute for the correspondent entity that the entity is linked to includes described at least one first First predetermined attribute in attribute, then it regard the property value of one first predetermined attribute as this The pronunciation of entity；And

If it is pre- that the attribute for the correspondent entity that the entity is linked to does not include described at least one first Determine any one first predetermined attribute in attribute, then using at least one pronunciation extraction template come really The pronunciation of the fixed entity.

9. entity polyphone disambiguation method according to claim 8, wherein, using it is described extremely A few pronunciation extracts template to determine that the pronunciation of an entity at least one entity includes：

Pair that one entity is linked to is matched using at least one pronunciation extraction template The property value of the character string type of at least one attribute of entity is answered, and the character string matched is made For the pronunciation of one entity.

10. a kind of entity polyphone disambiguation equipment, including：

Entity recognition unit, it is configured to identify at least one including polyphone from the text of input Individual entity；And

Pronunciation unit is determined, is configured to for each entity at least one entity, by this Entity link is based on the corresponding entity to the corresponding entity in the data set of the open data of association Its property value include pronunciation at least one attribute and/or the hair associated with the corresponding entity Sound, determine the pronunciation of the entity.