Be the Change
Yes, it is possible. 5,000 words are enough to carefully study each author’s writing identity. To do this profoundly, the technology analyzes texts at morphological, lexical, syntactic and semantic levels. This magic works thanks to support vector machine algorithm. To make it operate smoothly requires regular customization of feature and training sets. It was the biggest challenge for us. But we met it successfully. To achieve the desired effect, we also applied stylometric techniques and were able to carry out a deeper analysis of each writing.
See latest publications of our R&D team: https://link.springer.com/chapter/10.1007/978-3-319-59569-6_27
Checking writings for grammar errors is usually implemented as a primitive pattern-based tool. The truth is, it can be easily misled. The outcome is pitiful. Many errors are left unnoticed, which is of little help for a user, especially if he/she is a non-native speaker. We took another path. The technology we developed is based on augmented transition network capable of reading texts like humans do. It self-learns to be able to assess the correctness of the whole sentence and achieve 100 percent accuracy in grammar error detection.
This technology can take plagiarism detection to new heights. Paraphrase recognition is another achievement we are proud of. This technology is a multi-level classification system that has a lot in common with machine translation. It recognizes words replaced with synonyms, active verb constructions changed into passives and soon will be able to detect idea-based plagiarism. This technology also operates fast and is a good fit for different types of online solutions.
See latest publications of our R&D team: https://link.springer.com/chapter/10.1007/978-3-319-59692-1_14
Content is often lifted and translated to other languages to hide malicious intentions of cheaters. Crosslingual check is what can solve the problem efficiently. The technology comprises a huge set of dictionaries in different languages and analyzes semantic relations between words. Here is when paraphrase recognition comes in handy too. In a word, the latter allows applying metrics for measuring language translation quality.
This technology is a good fit for both lawmen and nonlawyers. Writing legal memos or wading through legalese to evaluate possible risks on one’s own is too time-consuming. Here is when technology can help handle the problem efficiently. Legal document extraction is what we are working on too. The task is laborious. But we are optimistic about the outcome. The trickiest thing is to pick up the right machine learning methods, since one and the same legal text fragment may refer to several categories. It requires expert advice. Hence our team of lawyers help us create a classifier with manually selected features.