The Intelligence Advanced Research Projects Activity (IARPA) is holding a proposers’ day for an upcoming research program aimed at both identifying an author by their writing style, and also making an author anonymous by removing their “linguistic fingerprints.”
IARPA will hold a virtual proposers’ day on January 19 for its Human Interpretable Attribution of Text Using Underlying Structure (HIATUS) research program.
According to the HIATUS program description:
HIATUS seeks to develop novel human-useable AI systems for attributing authorship and protecting author privacy through identification and leveraging of explainable linguistic fingerprints.
The program will develop novel techniques to generate representations that capture author-level linguistic variation and will use these representations to build human-interpretable algorithms to perform authorship attribution and ensure author privacy (i.e., via removal of author-identifying characteristics from text).
In other words, IARPA’s HIATUS program has two mirroring parts — attribution and anonymization.
Part one is to identify an author with the help of artificial intelligence that analyzes patterns in their personal writing style — their “linguistic fingerprints” — which can include their unique forms of spelling, syntax, vocabulary, phrases, punctuation, rhythm, formatting, etc. for attribution purposes.
They call this “attributing authorship.”
The other part is to reverse the process by removing an author’s linguistic fingerprint — their “identifying characteristics from text” — thus making them anonymous.
They call this “protecting author privacy.”
The HIATUS program contact point is Dr. Timothy McKinnon, who was also a program manager for IARPA’s Better Extraction from Text Towards Enhanced Retrieval (BETTER) program, which in 2019 awarded a contract to Raytheon BBN — a company that harvests the text of social media postings and other data.
“HIATUS seeks to develop novel human-useable AI systems for attributing authorship and protecting author privacy through identification and leveraging of explainable linguistic fingerprints” — IARPA HIATUS program
To get an idea of the type of experts IARPA’s HIATUS program is looking for, the teaming form asks for expertise in:
- Explainable NLP
- Software integration
- Forensic linguistics
- Text generation
- Authorship attribution
- Human computation
- Authorship privacy
According to the HIATUS program description, “Successful technical approaches will be scalable across diverse topic domains, genres and languages.”
While the research funding arm of the US spying apparatus doesn’t give any specifics on practical applications or real-world use cases for the technology, the fruits of the HIATUS program could be used across multiple scenarios.
In a research and analysis setting, the attribution aspect could be applied to something as simple as making sure the correct authors are identified and attributed from a cache of nameless documents for a variety of intel purposes and/or in-house record keeping.
In the real world, the anonymization aspect could be another layer of protection for journalists, whistleblowers, refugees, or spies in the field by knowing which linguistic features could be used by adversaries to identify them, while at the same time, the attribution component could be used by the government to identify an adversary’s communication by their distinct, linguistic fingerprints, (i.e. forensic linguistics, code breaking).
HIATUS program “will develop novel techniques to generate representations that capture author-level linguistic variation and will use these representations to build human-interpretable algorithms to perform authorship attribution and ensure author privacy (i.e., via removal of author-identifying characteristics from text)” — IARPA HIATUS program
Through authentic authorship attribution, the flow of information can become increasingly more transparent, or at the very least, more organized.
On the flip side, the flow of information can become even more distorted by anonymizing the source, concealing its origins, and adding more noise to the channel — a tactic used by spy agencies in which, “You create so much noise in the channel that people start to have overall doubts on all information that’s available in the media, social media, and other places,” as one former NSA foreign surveillance agent told The Sociable.
Whatever real-world applications come out of IARPA’s HIATUS program, they will involve one or more of the following characteristics: analyzing text for linguistic fingerprints, attributing authorship, removing author-identifying characteristics, enlisting artificial intelligence.
This opens the door for the US intelligence community to know more accurately who wrote what, but also, how to better conceal or fake the source.
Another tool for the kit.
Spies will be spies.
‘IoT devices are a growing source of data that can be collected to learn intent’: IARPA director
Intel agency awards contract to company that harvests social media text, data
Govt Geopolitical Forecasting Challenge offers $250K to predict the future