Project Jay-Walker
Image Credit: Business Insider - https://www.businessinsider.com/r-bin-laden-called-for-americans-to-rise-up-over-climate-change-2016-3
Problem
This project addresses processing information from handwritten text and from text written in obscure languages, content to which traditional processing methods have failed. The use cases for these problems likely exists in:
Intel Agencies - the processing of a tremendous amount of backlogged content.
Tactical Operators on Target - the quick translation and analysis of content gathered during sensitive site exploitation operations when proper interpreters are not accessible.
Processing handwritten content at scale is difficult, as Optical Character Recognition (OCR) lacks the accuracy to make it reliable. Moreover, natural language processors (NLP’s) work well on popular languages, but are less accurate when applied to more obscure languages. Some content (even ‘typed-text’) is in niche languages, such as Dari, Uzbek, Turkmen, Balochi, Pashayi, Nuristani, Pashto, and Urdu. Because these languages are not popular, NLP does not yield strong results, and produces ‘less than desirable’ confidence in the translated output.
hey princess are you gonna come with the boyz to
babe’s house to see judge smack
one into the lot next tuesday against your dirty sawx?
The above may be difficult to understand even though it is typed in English, unless you live in The Bronx. Cultural nuance, local dialect, regional geography, and events all impact language.
The level at which we are processing this extremely difficult content to a) quickly provide intel and analysis to the operator in the field, and b) to save the results to be stored and retrieved for traditional work and future AI/ML processing, allows room for improvement. This creates a hole in the intelligence gathering and analysis processes as we can foresee information is being omitted in both tactical operations and analysis.
Solution
For very difficult content, we propose a single capability to process small amounts within seconds for operators, and large back-logged amounts at scale for intelligence agencies. Project Jay-Walker is a human (service) and computer (product) system that transforms and translates high-valued classified content at scale using dispersed independent humans with cultural nuance and native language skills, by breaking content up into ‘snippets’, (sentences fragments, clauses, section of a page, and other naturally grouped context) and then quantitatively comparing the results processed by more than one human for high confidence. This solution as currently defined is focused on the processing of content, not analysis.
Strike Labs aims to build a system that will process up to thousands of pages per day, and return to the operator intel within seconds of that evidence being sent for processing.
A master solution comes from combining these two solutions together so that handwritten information is converted into machine readable format and then translated from a different language.
Contact us to learn more.