New Maltese spell checker aims to catch errors beyond spelling

A spell checker in development aims to spot mistakes even when every word is spelt correctly, using context to help people write Maltese more confidently

The Department of Maltese at the University of Malta is a partner in the project. Its academics provide language expertise, helping the team understand and classify the types of errors people make
The Department of Maltese at the University of Malta is a partner in the project. Its academics provide language expertise, helping the team understand and classify the types of errors people make

A spell checker in development aims to spot mistakes even when every word is spelt correctly, using context to help people write Maltese more confidently.

Project manager and Professor Claudia Borg gave the example of ‘fjura sabiħ.’ Both words are correctly spelt, but the adjective should be ‘sabiħa’ [beautiful] to agree with ‘fjura.’ [flower]

A spell checker works in two stages, Borg explained: it detects an error, then corrects it.

The correction can take the form of a red line beneath a word, allowing the user to right-click and choose from suggested alternatives. Autocorrect, by contrast, changes the word automatically.

The Department of Maltese at the University of Malta is a partner in the project. Its academics provide language expertise, helping the team understand and classify the types of errors people make.

Team members correct existing texts and consult documentation to collect examples of mistakes. Borg said the many books published on Maltese orthography give them plenty of material to draw on.

Maltese is a Semitic language rooted in Arabic, with vocabulary shaped by centuries of contact with Sicilian, Italian, and later English. Written in the Latin alphabet, it follows a different tradition from Arabic.

Maltese orthography reflects not only pronunciation but also word origins and structure. Its basic rules emerged in the 1920s and have since been updated to clarify spelling variations and the treatment of English loanwords.

The project, funded by Xjenza Malta, began on 1 November 2025, and runs for two years.

About half of the team’s effort goes into building a dataset. The other half focuses on the computing side, where researchers are testing different AI techniques.

The system is fed examples of mistakes alongside correct sentences. That balance is essential, Borg said. Without enough correct examples, it can overcorrect, changing an already correct sentence and introducing an error.

One way the team collected errors was through a master’s Dissertation by Alana Busutill, who developed a website which collected spelling errors through an interface where participants listened to audio and typed what they heard. The backspace key was blocked so the system recorded their original mistakes.

Busuttil told MaltaToday the errors collected were more varied than those she could have generated artificially. She also noticed more mistakes at the beginning of sentences than towards the end, suggesting participants may have become more accustomed to typing or more careful as they went along.

She said it was important for Maltese to be represented in the technology people use every day, warning that leaving the language behind technologically could contribute to it becoming obsolete.

In one approach being tested, the system examines words character by character, deciding whether a character should be deleted or replaced. Given a sentence, it either leaves it unchanged or returns corrections.

Working with very limited data makes the project an interesting research problem, Borg said. Challenges also arise on the language side, where some errors prompt extensive discussion about how to correct or classify them.

The checker works well for some types of errors but still needs improvement for others. The team plans to release an online version for testing and feedback and will continue improving it during the project’s second year.

Through the website, users will be able to paste their text, run a check and receive the output.

At the end of the project, the spell-checking engine will be released as open source, free of charge. Borg said the aim is to give something back to the community.

She hopes companies will incorporate it into their products, including IT systems used by the public and government. “Developers could integrate the checker into their systems so it can automatically correct text entered in Maltese,” Borg said.

Once the engine is available and the researchers are more confident in its performance, she hopes enthusiastic developers — possibly including the university’s AI students — will create tools that work on laptops and mobile phones.

For Borg, the project also aims to ease the pressure people feel when writing in Maltese.

She said she sometimes avoids writing in the language because she knows she will make mistakes. A tool that helps people write well and gives them confidence “is definitely going to encourage more language use”.

Even colleagues in the Department of Maltese, who use the language all the time, tell her that “even we have to stop and think” about how something should be written.

Accepting that Maltese spelling is difficult, and that tools can help, could allow people to relax a little, she said.

Borg called for more room to make mistakes, urging people not to “bring out the firing squad whenever someone makes an orthographic mistake”.

“We need to change our attitudes as well,” she said. The team also includes Adrian Grima and Michael Spagnol from the Department of Maltese, alongside Ray Fabri from the Institute of Linguistics and Language Technologies.

Three people are employed on the project: Daniel Attard and John Paul Grima work on the language side, while Lucien Noel works on the computing side.