Why Fact-Checking in Portuguese Needs a Boost

Fact-checking the internet is like trying to mop up an ocean. It’s time-consuming and struggles to keep pace with the torrent of misinformation. The problem? Misinformation spreads faster than the truth can catch up. Automating fact-checking could help, but progress is uneven. English has a head start, thanks to abundant annotated data. But what about languages like Portuguese?

ClaimPT: A Portuguese Breakthrough

Enter ClaimPT, a dataset of European Portuguese news articles annotated for factual claims. It’s a collaboration with LUSA, the Portuguese News Agency, featuring 1,308 articles and 6,875 annotations. Unlike existing datasets that lean heavily on social media or parliamentary transcripts, ClaimPT focuses on journalistic content. That’s a big deal for understanding misinformation in the media.

But here's the kicker: Portuguese, like many other languages, still suffers from a lack of accessible, licensed datasets. This shortage stifles research and development in natural language processing (NLP). If we’re serious about combating misinformation globally, we can't just focus on English. Whose data? Whose labor? Whose benefit?

The Need for Equitable Investments

The creators of ClaimPT didn’t just gather data. They ensured quality by having two trained annotators label each article, with a curator validating all annotations. They proposed a new scheme for this purpose. But the real question is, will this effort get the backing it needs from the tech community? Ask who funded the study.

If fact-checking doesn’t expand beyond English, we risk leaving non-English speakers vulnerable to misinformation. When it comes down to it, this is a story about power, not just performance. The tech world’s focus needs a broader lens, or it risks perpetuating existing inequities. Look closer at who benefits from these advancements.

ClaimPT provides baseline models for claim detection, establishing benchmarks for future NLP applications. But the benchmark doesn't capture what matters most if it's only for one language. Without continued investment and attention, the potential for this dataset to drive meaningful change is limited. Who will step up to bridge the gap?