SignIt! is an NLP language translator that takes in a text/file input and returns a language-translated document in more readable terms. SignIt! will give users the resources to understand complex jargon-filled legal documents received from doctors, lawyers, and others regardless of their education level or language fluency. Currently, there exist quick and high-quality text translators, such as DeepL, and paraphrasing software, such as simplish.org, which attempts to paraphrase text by constraining the simplified version to a dictionary of 1,000 common words. However, the latter often results in unclear or incomprehensible text, and the integration of these technologies does not exist.
To further explore the usability of SignIt!, we asked a user who thoroughly looks through all his legal documentation and gave him an experience prototype of our product to test. We wanted to measure how much faster his reading time improved and how thoroughly he was able to understand the document. The user was given a manually translated document with key phrases pertaining to numbers and binding agreements highlighted. The user was then evaluated based on how quickly he read the document and how much of the document he was able to comprehend. This study concluded that the users saved 50% of the time during reading and expressed significant improvement in ease of comprehension.
With strong market needs and usability, we moved on to identifying key vulnerabilities that needed to be addressed. Below are a few ranked in order of importance along with proposed solutions:
- Translation accuracy: Enhancing the model’s translation and predictive power. We plan initially manually identify key legal jargon, phrases, and other important numerical data in an effort to speed up translation accuracy.
- Time-Saving ability: Create a sidebar with highlighted phrases. This should allow users who do not want to read word-for-word translations to save a large amount of time as they only need to glance over the keynotes.
- Document privacy: Initialized steps for the model to recognize words like “SSN”, “ID”, etc, and not store the numerical numbers that directly follow that blank. This way our model can’t store your personal information but rather just copy pasting it over during a period of time.
We will know if we achieved our product is working if the tool can translate text seamlessly between languages, highlight important key terms in complex documentation, and no important information is “taken” during translation. All main features of SignIt! will be built using NLP libraries, React, and MongoDB. Apart from the key translation feature, we have considered incorporating:
- B2B user interface: Providing extra webpages for administrative users to oversee the different documentation users stored throughout the company
- Summary report: Provide a report on the content of the legal document, and educate users on why certain decisions were made in the simplification process
To accomplish the build of SignIt below is a rough timeline of the project:
- 1-3weeks: Do user research to identify customer needs and key features required
- 1-3months: Develop hi-fi prototype, and text, and start the development of the web application for basic language translation. Begin training model. Use a human in the workflow to catch errors
- 1-3quarters: Continue training model and have customers reject suggested simplifications
- 1-3years: Expand the app out for hierarchical business models
- 1-3decades: Lease SignIt! out for larger companies that perform translation on a regular basis
