Filipino social media users have tons of ways to say things, and AI can’t translate that yet.
Filipinos are just so creative at making memes. For example, they opened 2023 with “amaccana accla”—a meme for “Tama ka na, bakla” (“You’re already too much”) to warn others to avoid doing cringe-worthy activities. It became viral after actor Nadine Lustre tweeted a rather unsightly picture of her with that caption in December 2022.
Recent catchphrases, memes, and other forms of jargon like these are called Filipino netspeak.
Photo by @jpcleds on Twitter
Beyond Filipinos, this netspeak is basically foreign. Not even Google Translate or ChatGPT can translate it. But why is that so?
—
A dynamic language
Before we go into that, what exactly is Filipino netspeak?
“Filipino netspeak is an ‘informal language’ that has a lot of variations,” Justine Daphnie Salvacion, a Filipino language teacher and linguist, says. She and her professor Dr. Marilou Limpot published a study in January 2022 about Filipino netspeak after observing how a small class of Filipino students conversed in a Facebook group.
“Filipino netspeak has themes including the unconventional use of abbreviations, punctuation marks, and the like.”
Filipino netspeak is often formed through combinations of jargon, dialects and regional languages, and code-switching. It is also peppered with cultural references or contextual knowledge.
This process is called remediation, or a repurposing of a medium for another medium. Current Filipino netspeak is a remediation of textism—or messages abbreviated to fit in the 160-character SMS format used in mobile phones.
Textism was popular in the Philippines in the late 1990s. It also gave rise to argots such “Swardspeak” or gay lingo. Filipino textism has been replaced by netspeak used by more Generation Z and younger millennial Internet users. However, some Filipinos from older generations still use it when chatting on social media.
Marianne, who asked to change her name, often chatted with other migrant Filipinos on WeChat using Swardspeak.
But one thing stands out: her chats are untranslatable! WeChat’s AI-powered function translates content and group messages originally written in Chinese or other languages into the user’s phone language.
We ran a sample experiment where we sent ten messages written in various Filipino netspeak combinations to a Filipino user. WeChat’s translate function failed to make sense of all messages. We also tried feeding those messages to Google Translate and ChatGPT, but the same happened.
So why can’t AI-powered apps translate Filipino netspeak?
The simple answer: AI language models have not yet fully learned languages in the Philippines.
Language models like GPT-3 and GPT-4 learn from available texts on the Internet using a bot that indexes web pages. One crawler has indexed billions of pages since 2011—and 40 percent of it was in English.
How about Filipino—in this case, “Tagalog”? A dataset of one crawler indexed 701 million web pages in Filipino between January and December 2018. 701 million sounds enormous, but it’s not that much compared to English.
In short, AI doesn’t know much about other Philippine languages, let alone Filipino netspeak.
—
Secret codes
This unavailability might be useful for some Filipinos, especially those using WeChat.
Jerry, who asked to change his name for security reasons, sometimes uses code-switching and Filipino netspeak to communicate potentially risky terms such as “VPN” (virtual private network). He refers to it creatively, sometimes by syllables (“bee-pee-enn”), mixed spellings (“N-P-V”), or using metaphors (like “jump the [fire] wall”).
Others like May, who also asked to be renamed, comment on politics intentionally combining regional languages and netspeak to avoid detection.
This behavior happens on other apps, but it isn’t new. Marilyn Allanigue, a former professor at Olivarez College, says it has happened in textism, a topic she studied more than 20 years ago.
Here, she also found that Filipino textism negatively impacted the writing and reading behaviors and English language skills of youth. However, it also allowed groups of young people to form communities and find connections.
As textism developed and transformed into netspeak, Filipinos have found more ways to reconstruct communication.
“Textism has allowed us to create secret codes, but understanding it depends on the sender and receiver of the message. Foreigners can use apps to understand the Filipino language. But they cannot get anything when we use jejemon.”
Jejemon (or jejenese) is a prime example of textism, which combines digits and lower and uppercase letters.
“Filipinos are mataba ang utak (“very brainy” or highly creative). We can easily create our own codes,” Allanigue added. “But it is only a matter of time before AI can understand our netspeak.”
—
AI-powered Filipino netspeak
Salvacion says that since Filipino netspeak is a “non-formal, dynamic language”, AI naturally cannot run direct translations or detect cultural contexts yet. But once it does so, she foresees AI suggesting new buzzwords or trending phrases, potentially expanding the Filipino netspeak vocabulary.
However, she encourages young Filipinos to use netspeak sparingly.
“If Ais can translate Filipino netspeak, there may come a time that the youth will be heavily dependent on AI and become fond of using netspeak. It can lead to the degradation or gradual loss of the Filipino identity and the identity of our language,” she says.
Salvacion still recognizes the positive effects of Filipino netspeak. In her study, she suggests teachers use it as a “classroom strategy” because it can be an “effective way for online feedback”.
“Nonetheless, netspeak has proven that language is dynamic, timely, and adapting to the needs of the present,” she adds. “When used correctly, it can be a useful and productive language.”