RootWords
Meaning. What does this message actually say? Not what it gestures at. Not what its author hopes it implies. What it commits to. RootWords decomposes any text into its load-bearing elements: the claims it makes, the terms doing the heavy lifting, the entities it invokes, and the intent it is organized around.
Use it to
- Strip the spin off a competitor’s press release and see the five claims it is really making.
- Verify that your own draft actually asserts what your strategy memo promised it would.
- Turn a folder of unstructured transcripts into a queryable claim database.
- Feed clean, structured meaning into RootCause, Windfall, or your own models.
| Verb | Signature | Description |
|---|---|---|
extract | extract(text) | Decompose a text into its claims, terms, entities, and intent |
extraction = client.radiance.rootwords.extract(
text=open("competitor_press_release.txt").read()
)
for claim in extraction.claims:
print(claim.text, claim.commitment)Notice the commitment field. “Asserted” and “implied” are different legal, strategic, and rhetorical animals, and RootWords never confuses them. Notice also valence: contested on the term “streamlining”: RootWords knows which words mean different things to different rooms. That single field has saved careers.
ROI in one line: every downstream mistake starts as an upstream ambiguity. RootWords is the cheapest place in your entire pipeline to catch it.
Each layer of the decomposition is also addressable as its own class in the rail. Start at the top of the chain: Docs.