ABOUT
The translation process
Every substantive statement in Hansard is said again in plain words, beside the official wording. Here is how that version is written.
Our translation
The reader model reads each statement and writes a short, neutral brief: the speaker’s position, their argument, and what they are asking for. The brief makes sure the writer has the speaker’s side and point right, especially in a long statement.
Then the writer model uses that brief and the original wording to say the statement again in plain, grade-8 words, in the first person, as the speaker. It keeps what the speaker wants and the reason they give, and drops greetings, procedure and repetition.
Summaries and analysis
The reader model writes the short summary at the top of each debate and the profile of each MP. A debate summary says what the debate was about and what members said, with each side's points in its own terms. It treats every party the same, including the government, and does not say whether a reply answered the question. An MP profile describes what the member talks about and how, including the habits their grades show.
The writer model writes a plain summary of what each bill would do, taken from the bill's titles and official summaries published by Parliament and the Library of Parliament. It never uses the debates. It is written once per bill and updated only when that material changes, for example when the bill becomes law. A debate on a bill shows that bill's summary at the top in place of the debate's own summary.
The reader model also reviews statements for their Analysis page. It reads the official wording, never our translation, in several focused passes. It marks where a statement falls short on five measures: Signal, Relevance, Reasoning, Respect and Clarity, and gives a reason for every point lost. The site then turns those marks into five scores, an overall letter grade and score out of 100, and tags like Answered, Reasons given, Didn't answer, Off topic, Personal attack, Little substance, or Vague. The grade measures how a statement is argued, not whether its policy is right.
The newest statements are graded overnight. You can ask for a grade on any other statement, and it is usually ready within half an hour.
Mistakes
Language models make mistakes. If a translation or summary misstates what someone said, please get in touch.