Meet the first AI that understands Filipino languages—made in Iloilo

“Maayo nga adlaw! Ako ang imo abyan nga AI assistant, ari para buligan ka sa bisan ano, bisan san-o, bisan diin. Ano ang pwede ko mabulig sa imo subong?”

Imagine if an AI chatbot suddenly opened with Hiligaynon instead of English. Imagine if AI is not just intelligent, but also local, fluent in a dialect.

The future may be closer than we think. Recently, a Filipino developer has claimed to have created the first-ever Hiligaynon language model, bringing AI closer to truly becoming locally fluent.

Such a feat represents a significant advancement in the field of artificial intelligence. It opens the possibility of localized AI for underrepresented languages and cultures. 

The leap and the leaper

Cedrick Alegroso, a software developer and entrepreneur, said in a Facebook post that he has developed the first Hiligaynon language model, and possibly the first language model trained locally in Iloilo City. 

According to the 2020 Census of Population and Housing, Hiligaynon (or Ilonggo) is spoken in almost two million households, making up more than 7% of households in the Philippines. 

Alegroso remarked that while ChatGPT can understand and translate Hiligaynon, it does so only to a limited extent. His goal is more ambitious: a language model that can think in Hiligaynon itself.

He believes that every language deserves AI representation.

Curiosity into concept 

To build the model, Alegroso gathered a large dataset of Hiligaynon text from publicly available Bombo Radyo Iloilo articles. The dataset comprises 80,000 tokens — chunks of text that AI reads — which he estimates are equivalent to more than 100 full articles in Hiligaynon.

He then developed the AI entirely from scratch, starting it from a blank slate, and trained it step by step until it could process and respond in Hiligaynon.

The project was completed over a weekend using a consumer-grade RTX 3060 Ti graphics card. Alegroso noted that large budgets are not always necessary for this type of work. 

He also emphasized that the quality of data used in training is more important than the amount. “Quality data beats quantity every time,” he said. 

Naming the bot “Curiosity-Hiligaynon”, Alegroso admits that it currently serves a very specific niche. 

The base model of it can outperform other AIs when it comes to Hiligaynon news writing, but will struggle when it comes to general tasks.

He also clarified that it is still a raw base model, and would “hallucinate”. It would generate confident but false or made-up information that isn’t backed by real data.

However, Alegroso stressed that this was not the point of Curiosity-Hiligaynon. It was never about beating larger language models, but about proving the possibility of building AI locally and laying the groundwork for future development.

When AI speaks your language 

What Alegroso’s weekend project revealed is more than just a technical feat; it would be a glimpse into the future where AI not only understands our words but also individual cultures as well. 

Yet, for many local languages in the country, it’s a prospect still far away. 

According to the United Nations Development Programme (UNDP), almost half of the world’s population speaks languages that are rarely found on the internet. 

UNDP says AI tools often don’t work well for these less common languages, causing slower and less accurate results. Including local languages in digital tools is important to make technology fair and useful for everyone. 

Most AI systems today are dominated by English, as the majority of their training data comes from English sources. This can result in bias against speakers of other languages.

According to the Center for Democracy & Technology (CDT), AI detectors often mistake writing by non-native English speakers as AI-generated. This shows a clear bias in how these tools work.

In a CDT study using Test of English as a Foreign Language (TOEFL) essays, 61% of submissions from English learners were wrongly flagged. In contrast, the tools were almost perfectly accurate for native English writers.

This shows how English-centered AI can misjudge and disadvantage those who use other languages. It points out the need for localized AI models that can think in and respect different linguistic and cultural contexts.

In a 2024 study in the Journal of Business Research, adding local language options to AI chatbots can change how people use them. The researchers tested this on an Indian shopping app, which introduced a Hindi-and-English chatbot instead of only English. 

They found that having a bilingual chatbot made users interact more during each visit, which led to more purchases. This shows that people are more engaged when they can speak in a language they are comfortable with. 

The study found higher uninstallation rates for bilingual chatbot users, especially when making complex or costly purchases. Researchers suggested language access must be paired with smooth app performance to keep users.

Both studies show language strongly shapes our use of technology. Projects like Curiosity-Hiligaynon prove local efforts can make AI more inclusive.

According to the Philippine Information Agency, the University of the Immaculate Conception in 

Davao and the DOST-PCIEERD have launched the Mindanao Natural Language Processing Laboratory to preserve endangered indigenous languages using AI. The lab received ₱5 million to develop translation and chatbot systems for Mindanaoan languages.

It currently works with Manobo, Mansaka, and Kalagan, with plans for six more endangered languages. These tools aim to document, digitize, and share resources through AI translation and conversation.

Similar to Alegroso’s Curiosity-Hiligaynon, they aim for a seamless integration of local languages into the world of AI.

These efforts are proving that AI can go beyond global languages to understand and honor local voices and cultures. Projects like Curiosity-Hiligaynon and the Mindanao NLP Lab are small steps toward a future where AI truly belongs to everyone. 

When technology speaks a local language, it carries culture, history, and words, affirming the place of everyone in the digital world. It transforms cold code into a living bridge between human identity and the humane future.

So, imagine an AI that greets you in “Maayo nga adlaw!”’ speaking not just English but the many dialects that make up our identity. The future may be closer than we indeed thought.