How search works¶

Hansard search uses two layers, in order:

Semantic search — your query is converted to a vector (a numeric fingerprint of its meaning) and compared against the same kind of fingerprint computed for every speech chunk in the database.
Optional keyword filters — you can narrow results by speaker, party, jurisdiction, date range, etc. without changing the semantic ranking.

Results are sorted by semantic similarity by default. You can re-sort by date or speaker.

Why semantic, not just keywords¶

A keyword search only finds speeches that contain the exact words you typed. Semantic search finds speeches that mean something similar to what you typed, even if they use entirely different words.

Concrete example

Search: government should not subsidize fossil fuel companies

A keyword search would only find speeches with that exact phrase.

Semantic search returns speeches arguing against oil and gas tax breaks, against pipeline funding, for ending subsidies, against LNG support, and so on — including ones that never use the word "subsidize" or "fossil fuel."

What gets searched¶

Every speech in our Hansard corpus is split into smaller chunks (typically a few sentences each). Each chunk is embedded into a 1024- dimensional vector and indexed. When you search:

Your query is embedded with the same model.
The system retrieves the most-similar chunks across the whole corpus.
Chunks are grouped back into the speeches they came from for display.
Each result shows the most relevant chunk, with a link to the full speech in context.

What does not affect ranking¶

The popularity of the speaker.
The political party of the speaker.
How recently the speech was made (unless you sort by date).
How long the speech is.

The ranking is purely about semantic similarity to your query. This is deliberate — it would be easy to bury small-party voices behind a "boost ruling-party speeches" rule, and we don't.

Tips for better results¶

Ask the way you'd ask a person. "What does X think about Y?" works better than just "X Y."
Be specific where it matters. child care fee reductions outside Quebec returns better results than child care.
If the first results are wrong, rephrase rather than scrolling. A better query usually beats deeper paging.
Use filters to slice, not to define. Filter to "Liberal" if you want Liberal MPs' speeches on a topic — but the semantic query still defines what you're looking for.

What if I want exact phrases?¶

Wrap a phrase in straight double quotes ("...") to require it appear verbatim. This re-ranks results to favour matches with that phrase, but still uses semantic similarity for ordering. For pure boolean / regex search, this isn't the right tool — Hansard.ca and OpenParliament both expose keyword search if that's what you need.

Languages¶

Most provincial legislatures publish in English, with the federal Parliament and the Quebec National Assembly publishing in both English and French (and Quebec primarily in French). The embedding model is multilingual, so a French query can return French-language speeches; an English query about a Quebec topic will return relevant French speeches alongside English ones.

Translation between the two is not yet a built-in feature — we surface the original. For mixed-language searches, ask the question in the language you want most of the answers in.

Privacy of your searches¶

Search queries are not associated with your account unless you explicitly save the search. Anonymous searches are logged in aggregate (counts, latencies) for performance tuning, never with your IP or the query string itself.