Lab
Video NLP for verified news
Extending natural language processing to news video — so claims made on screen can be transcribed, extracted and checked against sources, with deeper analysis than text-only tools allow.
Much public information now arrives as video: speeches, interviews, broadcast clips and short-form posts. Most verification tools are built for text. This experiment works on the video end of that gap.
What it explores
- Transcription of news video, with timestamps so every claim can be traced to the moment it was said.
- Claim extraction — separating checkable statements of fact from opinion and commentary.
- Source checking — matching each claim against primary records and published fact-checks.
- Analysis — how claims are framed, repeated or changed across clips and outlets.
Ground rules
Retrieving a matching source is not the same as verifying a claim. The design keeps the evidence visible and leaves the final judgement to a human reviewer. Results will be shared here only once they have been tested.