Come sit with us at the FilBench!
•
I have high hopes for Filipino NLP.
Three years ago, I didn’t have any involvement in the Filipino NLP space. It just so happened that I was the only Filipino in the spaCy team1, and despite that, there’s no Tagalog language support in the library. Perhaps it was nice to represent: it felt like I was alone in a grocery aisle and I spotted a cereal box on the floor. I’m already there, I’m quite capable, so why not pick it up?
Fortunately, I realized that I wasn’t alone. I still remember cold-emailing Blaise and sharing initial results of an NER pipeline (which soon became a core component of calamanCy). Through him, I got connected with Joseph, whose works I’ve been reading for quite a while. I still remember us meeting in person back in 2023, and I guess the rest is history!
Fast-forward to 2025, across many life events and other collaborations, I also met Ely and Conner through cold emails. We started with an annotation project that didn’t really materialize, but we pivoted into a more impactful project on language model evaluation. That project soon turned into FilBench, and together with Joseph and Blaise, it was published in EMNLP Main!
I know it’s confusing to call both the evaluation benchmark and this collective as FilBench,2 though I find it quite apt (and fortuitous if I may add!). Back in my undergrad, I remember the idea of “org benches” where people hang out to do work. There was something special about having a designated bench, where people could gather, plan, and collaborate. That’s what I think FilBench represents: a bench for Filipino NLP researchers to call home.
In my FilBench-Eval blog post, I mentioned that hopefully next time there will be more of us. I’m happy to see it come into fruition: our first meeting to plan out our future projects had a nice turnout! Who would’ve thought that in just three years, there’s more than ten (!) people willing to collaborate with us. We also have a very chill Discord group with other Filipino NLP researchers.
This collective is not a non-profit, for-profit, or a research institute. We are PhD students, software engineers, enthusiasts who have shared interests in improving the state of Filipino NLP through open research, tools, and datasets. I’ve seen grassroots efforts worked well in other language communities such as Masakhane, SEACrowd, and IndoNLP— I believe it’s possible to do the same thing here. I hope the momentum continues.
Kaya, tara! Come sit with us at the FilBench!