Argh Matey, Let AI Help You Speak a New Language

Learning a New Language (even Pirate) is Easier than Ever

“If you were waiting for the opportune moment, that was it” Captain Jack Sparrow. When we think pirates, oftentimes Johnny Depp as Captain Jack Sparrow comes to mind. What’s not to love about pirates? They sail the seas looking for buried treasure, drink a lot of rum, and talk in their own pirate language.

The best part about being (or pretending to be) a pirate, aside from the swashbuckling, seemingly endless stores of rum, and the opportunity to wear a hook like it’s next season’s Birkin, is the unique and very nuanced language. Sure, we all know the words “ahoy” and “parley,” but there’s actually a lot more to the language than a handful of key phrases, and that’s where AI and quality unbiased training data comes into play. Thanks to AI, we can create new languages (conlang), bring old ones back to life, or expand on existing ones that we know and love like Klingon or Elvish. The more common, straightforward use is that it helps us learn another language.

Language Learning Powered by AI

Language learning apps and websites are the go-to for people who want to pick up a few words before an international trip or to learn a new language on a budget and not have to take classes or hire a tutor.

Duolingo is one of the most used programs for language learning, and they utilize AI to create the best language lessons for all who use it. As a user completes a lesson, their successes and mistakes are saved, and that data is sent back and analyzed. Specifically, the company wants to see what the most common mistakes made with each lesson are. This is where AI steps in, their program Birdbrain personalizes each user’s experience based off of their strengths and weaknesses to create lesson plans that ensure maximum learning potential.

Duolingo first started using AI back in 2013 to help figure out how long it takes users to forget vocabulary they learn. Their new AI, Birdbrain was implemented in 2019. The algorithm wasn’t all knowing at first. Initially, the program just knew that a user was right or wrong, over time it had to learn why exactly a mistake was wrong. The program to figure out why a mistake was wrong is called Blame.

While Duolingo doesn’t offer a pirate course they do have both Klingon and High Valyrian but if it’s a pirate life’s for ye, the Daily Press has published a pirate vocabulary guide.

AI Preserves History: One Language at a Time

Dr. Mark Harvey teamed up with the Larrakia Nation Aboriginal Corporation of People and Appen with the goal of improving the database of usable text and audio data language samples of the Larrakia language. One of the reasons for this project was that the last fluent speaker of the language passed away over 20 years ago and creating a database would ensure that future generations could learn the language, so it wouldn’t join the ranks of extinct ones. He chose to work with Appen as he knew we would be up to the challenge of providing unbiased diverse training data to meet the challenges of this project.

One of the biggest challenges of this project was that there were two databases, one for audio, and one for text, that weren’t linked together. This is where we stepped in and helped align the two databases together. To find out exactly how we accomplished this, you can read our case study here.

AI is Making Languages You Didn’t Know You Needed

If you ever wondered if AI programs spoke to each other and in what language it was – you might get the answer sooner than you thought. There have been two instances where AI programs created their own language to communicate with each other, while it may be gibberish to us, it’s a real language to them.

One instance occurred when Facebook chatbots were trained to negotiate. They had to tweak one of their models because otherwise the bot-to-bot conversation “led to divergence from human language as the agents developed their own language for negotiating.”

The other instance occurred with DALL-E2 an AI system created by OpenAI. DALL-E2 was given simple tasks of image generation based on a provided description. Recently the AI model was providing what was first believed as gibberish when generating the images. Yet, once someone typed said gibberish into the system it generated back the same requested image from earlier. This sparked the creation of a paper on the Hidden Vocabulary of DALL-E2.

This begs the question, what would happen if we asked AI to create their own language, would we get something nautical like another pirate language, or perhaps something more magical like Elvish?

Conlang Has More Fun

Conlang, or constructed language, is a term used to describe a language that was created by an individual (like High Valyrian from Game of Thrones) and not a language that occurred naturally in society (like the language this article is in). Conlang isn’t just for speaking, it also includes other constructed languages like programming language whereas natural language processing (NLP) is just giving computers the ability to understand natural existing languages like humans do.

In celebration of Talk Like a Pirate Day, why not make your own new language?

Talk Like a Pirate Day

Created in 1995 as a joke, John Bauer and Mark Summers created this new holiday when they were playing a game of racquetball. The holiday become known to the world when John and Mark mailed a letter about it to Dave Barry, a humor columnist, who gave it the exposure it needed to become a world-wide holiday.

This holiday occurs during translation month, which is fitting as you need to translate from pirate to the language you speak to learn about what those unique nautical words actually mean.

As a company that works with NLP, translation, and localization we enjoy preserving languages and sharing them for everyone to appreciate, after all communication is essential in our daily lives.

To get into the pirate spirit use a pirate translator to learn to say your favorite phrases like a swashbuckling sailor.

Website for deploying AI with world class training data