Documents have an important role as a way of expressing, communicating, and storing information.
Artificial intelligence (AI) has progressed to a level where people expect it to understand the content of a document as part of a solution that helps streamline and automate tasks.
Natural language processing (NLP) is a key technology in such advanced document solutions.
Deep Alignment is an NLP technology developed by Ricoh. It automatically aligns two documents, associating sentences and paragraphs with similar content with each other.
The technology visualizes the differences between two documents instantly. For instance, you can compare a draft contract with another or compare similar articles and clarify information that is absent/present in one or other of the documents.
Deep Alignment consists of the two new technologies described below.
A complete sentence can often have several meanings. Thus, a sentence is too large a unit to be used for association based on meaning alone. In contrast, a word, which is the smallest unit of meaning, is too weak to be used for association because it tends to appear in multiple sentences.
Deep Alignment uses phrases, which consist of multiple words, as keys for association. It synthesizes the meaning of words obtained through deep learning into the meaning of phrases, thus enabling precise association of meaning.
In the area of machine translation, technologies have been developed to associate original and translated sentences in two texts. Conventional technologies have only limited applications, as they assume a correlation between both texts in terms of their sentence order.
Deep Alignment, however, works independently of the sentence order, so it can be applied to tasks of association more versatilely. It can be applied to one-to-many associations, where one sentence with multiple meanings is associated with multiple different sentences, or even to tasks where association counterparts are missing.
Besides contracts, Deep Alignment has many potential applications e.g. proposals, specifications, provisions, and more. Deep Alignment associates items at the meaning level, and will greatly accelerate and enhance the checking process in many tasks.
Ricoh will continue to promote the technology concurrently with its many partner companies and further develop new NLP technologies.