API
Various APIs for natural language processing and speech recognition are available.
Parsing
Analyze the structure and semantics of Japanese text.
- RESTful API for parsing statement structure and semantics.
- The input sentence is decomposed into sentence clauses and morphemes, and semantic information such as interrelation between sentences, interrelation between morphemes, and part of speech information is added.
- Since information such as "Who (what)" and "What's wrong?" can be extracted by word unit called morpheme, it can be applied to data mining to extract and analyze information on a large number of text data.
Use Case
NTT COMWARE CORPORATION
Sales Support BOT
Named Entity Extraction
Extract named entities such as person names and geographical names.
- It is a RESTful API that outputs 8 Named-entity Class of person names, place names, date representations (Time, Date), organization names, quantitative representations (Amount, Percentage), artifacts, and Extended Named-entity Class with over 200 classes based on "Extended Named-Entity Class by Mr.Sekine".
- Phrases such as names and place names can be extracted from input sentences, so it can be applied, for example, to applications that analyze topics in text data.
Company Name Normalization
Extracts and normalizes proper names (Company Name) from text.
- RESTful API for extracting normalized company names.
- Provide normalized company name information for company names that contain misrepresentation or fluctuation.
- Since unique company names can be extracted from input statements, it is possible to apply this method to the aggregation and analysis of a large number of text data, including company names, by company.
Reference Resolution
Detect demonstrative word such as "that" "his/her" "same" "its" as well as abbreviations, and then output sentences that specify the contents.
- This is a RESTful API that receive multiple sentences written in Japanese as an input, extract pronouns such as "そこ", "それ","彼", "彼女", extract an antecedent that correspond to those pronouns, and outputs them as a pair of pronoun and object.
- For example, in the analysis of the dialogue log between the dialogue engine and the user, words indicated by pronouns are extracted from sentences containing pronouns and their surrounding contexts, and words such as "he" and "she" that are not very meaningful in the log analysis are replaced with antecedents, enabling more precise log analysis.
Keyword Extraction
Extract major keywords from sentences such as person name, places, organizations, etc.
- This is a RESTful API that receives multiple sentences written in Japanese as an input, and extracts characteristic phrases and words contained in input sentence as a keyword.
- Outputs the specified number of phrase words in descending order based on the characteristic score that is computed.
- The extracted phrases can be used as tags for articles, for example, in applications that facilitate searching.
Similarity Calculation
Calculate similarity between sentences, and output the result as score 0 to 1 .
- A RESTful API for receiving two sentences written in Japanese as input, calculating and outputting semantic similarity between sentences.
- The degree of similarity is defined from 0 to 1, indicating that the closer to 1, the greater similarity between the texts.
- Since the similarity is calculated using semantic information of words contained in sentences, we can also estimate the similarity between the sentences containing different words.
- It can be applied to applications that extract and respond to the most similar texts and answers in FAQs to various user questions in language processing such as search system and FAQ automatic answer system.
Sentence Type Classification
Receive a sentence written in Japanese as an input, identifies and outputs the Modality types (descriptions/questions/commands) and speech act types.
- A RESTful API for receiving a sentence written in Japanese as an input, identifies and outputs the Modality types (descriptions/questions/commands) and dialog act types.
- Since it is possible to determine whether a sentence is an interrogative sentence or an instructional sentence, it can be applied to applications that appropriately select a response module for a user's utterance in language processing in robots and dialogue engines.
User Attribute Estimation (β)
Estimate the user's age, occupation, and other attributes from the text.
- A RESTful API for receiving sentence written in Japanese as an input, estimates and outputs the attributes of people such as age, gender, hobby, occupation, etc.
- We have tuned it so that the input of user's tweet and profile on twitter is assumed.
Filler Removal (β)
Removes filler words from the user’s voice input.
- It is a RESTful API that receives the text after speech recognition process, specify the fillers such as 「えーと」, 「あのー」, outputs the information and text without fillers. Also, it normalizes the wrong expressions which caused by long vowel and sound disruption, and outputs the normalized text strings.
- It can be used to improve the accuracy of data utilization after recognition by applying speech removal to text such as minutes written by speech recognition.
Detect Misrecognition (β)
Detect and extract words in the text that may cause recognition errors after speech recognition process.
- It is a RESTful API that receives the text after speech recognition process, outputs the suspected recognition error and the score at the same time.
- The errors are extracted with a score of 0 -1, with the closer to 1 the more likely the error is.
- For example, by applying speech recognition error detection to text such as minutes written by speech recognition, it is possible to extract only those parts that require manual correction, so that finishing work can be performed efficiently.
Sentiment Analysis
Sentiment Analysis is text classification tool that analyses an incoming message and determine whether the underlying sentiment is positive or negative. Sentiment Analysis API can also recognize types of feelings such as anger, joy, fear, disgust, sadness, surprise, etc.
- A RESTful API that is text classification tool that analyses an incoming message and determine whether the underlying sentiment is positive or negative. Sentiment Analysis API can also recognize types of more than 50 different feelings such as joy, surprise, fear, pleasure etc. within the sentences.
- Judging whether the input is a positive sentence or a negative sentence, you can receive word-of-mouth and reviews from users of your product. It can be applied to an application that analyzes the evaluated points and dissatisfaction points of the product.
Speech-to-Text
API for converting audio data of natural human speech into text.
- This Text to Speech API accepts and transcribes audio data in file or streaming format.
- You can also use the API for registering and deleting the text to speech dictionary for each user.
Use Case
NTT COMWARE CORPORATION
Sales Support BOT
Text-to-Speech
Generates speech artificially from text.
- RESTful API for receiving text and artificially generating speech.
- You can use a word dictionary for text to speech and a RESTful API to register, list, update and delete sentences.
- In addition to being able to read in context, multiple speakers can be selected, and speech speed and intonation can be adjusted, making it possible to generate synthesized speech according to use cases.
Summarization
This API summarizes text written in Japanese or English.
- Japanese and English Models
- A RESTful API that provides two summarization methods: abstractive summarization and extracted summarization for Japanese and English.
- Call Center Model
- A RESTful API that provides two summarization methods: abstractive summarization and extracted summarization, that are specific to summarizing contact history in a call center.
- We only accept Japanese.
- Abstractive Summarization
- This is a summarization method that analyzes the original sentence and creates a new natural sentence with the number of characters specified by the Japanese and English models created by learning many summary cases.
- Extractive Summarization
- It is a summary method that analyzes the original sentence and selects important sentences of a specified number of sentences.
- Notes
- Summary results using this API may not be as expected.
- The number of characters generated by the abstractive summary will be close to the specified number of characters, but may exceed it.
- If you specify the number of characters or sentences that exceeds the original document, the response is returned with the original sentence.
- If you make a request not using the specified character code, the summary text in the response could be garbled.
- In the API, use a period, question mark, or exclamation mark to identify sentence breaks.
- When summarizing a bulleted text, add a period "." as a delimiter at the end.
Text Classification
Classify a text into pre-learned classes.
- A RESTful API for classifying a text into pre-learned classes.
- You can also use the RESTful API to learn and delete classification models and to check the status of models.
- Because classification classes can be learned arbitrarily, they can be applied to a variety of applications, such as "Business Contact Classification" and "Customer Review Categorization".
