Japanese grammar proofreading/notation fluctuation & rough word checking tool | ZeroTools

This feature is a system that uses Kuromoji.js, a morphological analysis engine for natural language processing, at its core to extremely precisely decompose and evaluate input Japanese character strings at the part-of-speech level.

In particular, it accurately captures the continuous use of particles, which is an issue in elaborating Japanese. For example, complex nested possessive structures, such as "no" occurring three or more times in the same clause, are separated from the syntax tree and a warning is issued as a factor that significantly reduces readability.

In addition, the asynchronous text parsing process instantly identifies `words left out'' that are easily mixed unconsciously in everyday writing, redundant expressions such as `I have a headache'' that cause semantic duplication, and double negative constructions that impose an excessive cognitive load on the reader.

These analysis processes score the context and dependency relationships of input sentences on a clause-by-clause basis, and comprehensively identify parts that are likely to violate Japanese syntactic rules using a hybrid probabilistic approach and rule-based approach.

A unique lexical analysis algorithm thoroughly equalizes the duplication of sentence endings and spelling fluctuations caused by the author's cognitive bias when writing long texts.

The frequency of occurrence of different character strings for the same concept such as "server" and "server" and "do" and "do" is calculated across the entire document, and expressions with a low occurrence rate are extracted as exceptional values.

At the same time, when detecting misspellings, we refer to an N-gram language model built from a general Japanese corpus and flag character sequences with extremely low transition probabilities as abnormal values.

In addition, when evaluating sentence final expressions, we use the syntactic analysis tree to identify places where monotonous final forms such as "It is ~ desu.

~ desu. ~ desu." are repeated three or more times in a row, and point them out as factors that impede the reader's reading rhythm.

In addition to these, the system not only points out errors, but also dynamically generates suggested correction text that is estimated to be the most appropriate based on the context, logically supporting the author's revision process.

The system is equipped with a unique dictionary layer that is flexible and highly scalable in order to comply with company regulations and strict terminology definitions in specific academic fields.

Users can freely define custom rule dictionaries that support regular expressions via the interface and have them intervene in the analysis engine's evaluation criteria in real time.

This allows automatic replacement of specific NG words and control of project-specific terminology with extremely low latency. The unique rules set are evaluated with higher priority than the standard parsing dictionary, ensuring that expressions that are correct in standard Japanese but inappropriate in a specific business context are captured.

Dictionary data can be imported and exported in JSON format, and the architecture allows for easy sharing of terminology control rules within a team and version control of regulations during large-scale editing operations.

In order to safely edit confidential data that cannot be leaked externally, such as a company's unpublished press releases, highly confidential internal documents, or the text of unpublished papers, this system has an architecture that completes all text analysis processing within the browser's local environment.

All scripts for the morphological analysis engine including Kuromoji.js, language determination module, and orthographic fluctuation extraction algorithm are loaded on the client side, and there is no physical or network structure in which input text data is sent to an external server.

Background processing using Web Workers enables high-speed analysis by maximizing the device's CPU resources without blocking the main thread's UI drawing, even for large text data containing tens of thousands of characters.

This guarantee of complete offline operation allows users to immerse themselves in editing work while completely eliminating the risk of information leakage.

The extracted redaction candidates and locations with spelling fluctuations are dynamically mapped as DOM elements on the text editor and highlighted to make them easier to recognize visually.

This highlighting element is more than just a decoration; it is internally bound one-to-one to a node in the abstract syntax tree, and deploys a pop-up with contextual correction suggestions the moment the user clicks on it.

Furthermore, it is equipped with a batch correction mechanism that reflects the differences all at once for the same spelling fluctuations and consecutive misused expressions scattered throughout the document.

By using the one-click batch correction function, replacement processing is performed transactionally on all target nodes, even for long data spanning hundreds of pages.

The replaced text is immediately sent to the reparsing pipeline, forming a circular elaboration loop that verifies in real time whether the corrections introduce new syntax errors or particle sequences.

This system implements application-specific elaboration profiles that dynamically change the rigor of analysis and evaluation criteria depending on the nature of the target document.

For example, in a web article elaboration profile, we prioritize crawlability for search engines and readability for readers, and strongly recommend reducing redundant expressions and converting them to plain language.

The business email proofreading profile strictly censors excessive duplication of honorific language and frank expressions that are inappropriate for documents intended for external use, and encourages you to revise the text to be dignified and in accordance with business etiquette.

In addition, if you select the paper draft check guide profile, advanced context analysis will be enabled to ensure academic objectivity, such as eliminating subjective adjectives, strictly unifying the key of da and da, and using logical conjunctions appropriately.

This makes it a single tool that serves as the perfect elaboration assistant for a wide variety of writing scenarios.