“There were great early successes with code-breaking but human language is much more complex and very hard to model.”

But by deploying a data-driven, statistical machine translation approach – “It’s far more effective,” she says – Siahbani has made invaluable progress, publishing several well-received papers at leading neuro-linguistic programming conferences. She’s also made some fundamental practical breakthroughs.

Encouraged by SFU supervisor Anoop Sarkar, her work addresses the problem of translating natural languages as a machine learning challenge. This ultimately led her to create her own left-to-right hierarchical phrase-based translation system – LR-Hiero, for short. The project, she says, was far more demanding than she imagined it would be.

“I thought it was a great idea but after working on it for six months, I began to realize just how hard it was. It was incredibly challenging to make it work,” Siahbani recalls, adding that she never gave up. “In some ways, it was a nice journey! I didn’t know anything about machine translation systems in the beginning so I had to learn everything. But my supervisor believed in me and kept saying it should work. I just had to keep going back through everything.”

Eventually, she resolved the issues and LR-Hiero was born. “It was a very interesting problem to work on,” Siahbani says modestly of a solution that could prove invaluable to corporations from Amazon to Microsoft that continually strive to streamline user experiences. There are wider potential uses, too.

“One of its direct applications is online translation, such as chat translation or speech-to-speech translation. I’ve also tried it for speech-to-text translation with promising results. We balanced the trade-off between accuracy of translation and the time a user should wait to get the translation of what the speaker said and obtained reasonable translations twenty times faster than regular systems.”

The real-world possibilities, she continues, are limitless. “Sometimes computing science problems are very abstract and no-one really cares about them. But this is something people are interested in because they can really use it.”

With these initial successes under her belt, what next? “I’m looking at several different opportunities and I’ve been thinking about different research labs, as well as maybe applying for tenure track. But I’d like to continue my research – there are so many more areas for me to explore in machine translation.”

Watch below: SFU computing science professor Anoop Sarkar, who supervises Maryam Siahbani's research, envisions a world where language is no longer a barrier, thanks to a system that could understand and translate every single language in the world.