238k speeches: Guardian and UCL trained a non-generative ML model on 100 years of House of Commons debates. Both Labour and Conservative MPs are currently at or near their most hostile on immigration, driven by competition with Reform UK.
- Before 2000, immigration accounted for 1% of all Commons speeches; by 2023 it had reached a record 5.4% — the debate isn't just getting harsher, it's getting louder.
- The methodology is unusually transparent: human annotators first, LLMs used only to scale training data, stress-tested against different keyword combinations, with sarcasm and indirect quotation accounted for.
