Sarcasm, banter and British English: why sentiment analysis is hard here
"Well that's just brilliant, that is." Every British reader knows this is a complaint. For a decade, sentiment software scored it glowing, which is roughly why marketers stopped trusting sentiment dashboards.
The national style
British English routinely inverts surface meaning: understatement ("not ideal" for catastrophes), ironic praise ("lovely stuff" under a photo of a flooded kitchen), and stoic humour that wraps genuine grievance in a joke. Regional flavours add slang layers; football culture adds an entire dialect where abuse can be affection and "we were robbed" is Tuesday.
Why it matters commercially
Mis-scored sarcasm does not just noise the data; it inverts it. A campaign that generated a wave of ironic mockery can read as a sentiment triumph to a naive tool. Decisions made on inverted data are worse than decisions made on none.
What actually helps
- Models that read whole-context, current-generation language models, rather than keyword scorers
- Prompts that name the client and ask for sentiment toward them specifically, so anger at the football result is not billed to the sponsor
- UK-specific instruction and examples: tuned prompts measurably improve accuracy on ironic text
- Human review of negatives and samples, catching the residual misreads and teaching the prompt
Perfection is not available. What is available: accuracy high enough that the monthly trend is real, with humans auditing the edge cases. That standard is now cheaply achievable, and it is the one that matters.
MyView watches every mention of your brand and tells you what needs a reply.
Book a demo