Marsal Gavaldà, PhD, heads the machine learning team for the commerce platform at Square, a financial services company that helps facilitate transactions for millions of business owners. Working out of Square’s Atlanta office, Gavaldà pursues automated solutions to a variety of internal operations that revolve around the company’s ethos of economic empowerment. His academic backbone is firmly entrenched in computational linguistics, but he has a knack for developing machine intelligence of all sorts, particularly related to customer interaction.
A look at Gavaldà’s resume from the past few years includes highlights like overseeing research and engineering at respective startups MindMeld and YikYak. Still, his passions cover broader ground than computer science. Each year, Gavaldà organizes a summer science and culture summit that covers “topics as diverse as machine translation, music, or the neuroscience of free will.” Gavaldà certainly knows how to stay busy, but I had the opportunity to chat with him about his experiences and outlooks regarding NLP, his varied interests, and technology at large.
Data as human behavior
Entering the finance sphere wasn’t exactly in Gavaldà’s plans. Growing up in Barcelona, Spain, the capital of Catalonia, he recalls being struck by the dichotomy in which some of his classmates spoke Spanish and others spoke Catalan. From a young age, this drove him to contemplate the whirlwind of linguistic diversity. What is common across languages? What is unique?
“You start seeing how each language provides a slightly different perspective on the same events,” Gavaldà explained. He experienced a newfound shift in perspective when his dad brought home an HP-85 one day, one of the early home computers. “A beautiful machine” that Gavaldà and his father swiftly started learning to program. Always drawn to the swirling mechanics of language and computers, Gavaldà had yet to find himself equally entranced by financial transactions before he joined Square. But he dove into the data after coming aboard the Square team and eventually found himself immersed in a field he had never anticipated.
“It dawned on me that commerce is rooted in human behavior, it’s a conversation that the buyer and seller are having,” said Gavaldà. With the richness of the transactions and the accompanying data, there is no shortage of NLP work to be done. Customer interactions are one of the clearer instances for wielding NLP — optimizing marketing campaigns, driving customer-facing product features, automating the process of identifying merchants who might be ready for a loan. Ontology, however, is another important realm, wherein NLP can assist with judiciously structuring inventory catalogs based on product descriptions. At the end of the day, Square’s goal is simply to make it simpler to run a business. NLP, where Gavaldà comes in, is a crucial piece of that human-centered puzzle.
Square (Image source: WooCommerce)
A wide open field
When it comes to the future of NLP and AI, Gavaldà is resolutely optimistic. “Some people criticize deep learning as just glorified logistic regression or glorified statistics, but it’s a step change. In its matrix, it’s quite different… The main successes have been in very structured environments, but I do think the evolution is unstoppable,” Gavaldà mused. Although certain areas may be more prone to roadblocks, countless unsolved problems are far from intractable, and the data grows by the day.
Gavaldà accrued his MS in Computational Linguistics and PhD in Language and Information Technologies — both from Carnegie Mellon University — but he stresses that for those who are “super motivated and have a very specific interest,” going the PhD route isn’t necessarily necessary. “The field is very open. It’s quite easy to follow what the latest research is. Most of the best papers are on the arXiv website, and there’s a lot of open source [research], so it’s easy to replicate results,” Gavaldà said.
While a PhD can help ground your research in a historical and theoretical context, it’s not a prerequisite for getting your hands dirty. Gavaldà insisted that “if you have a clear idea of something you want to build, just by looking at existing projects and attending hackathons, you can accomplish it faster than spending five years in the ivory tower.” To Gavaldà, AI is here to stay, and it’s also here for the taking.
Consorci Universitat Internacional Menéndez Pelayo Barcelona (Image source: CUIMPB)
Tech as economic empowerment for all
Even with openness as a tagline of the tech community, Gavaldà recognizes the existing barriers to widespread tech literacy and accessibility. For one, many machines lack the facilities to be universally useful. Speech technology is a particular domain that suffers from a debilitating lack of data. “There [are] certain classes of people that aren’t as well-recognized as others,” noted Gavaldà as an example. “Children, the elderly, people with an accent.” He sees clear gaps in the veneer of equalizing progress, chasms that extend to a broader public dearth of knowledge when it comes to tech. Amidst a wide landscape of AI achievements, many people throw their hands up in exasperation at the always shifting frontier of innovation. A concerning question emerges: if the majority of the public doesn’t understand the ins and outs of technology, how can we make informed decisions in our 21st century environment?
In an effort to promote understanding and stimulate conversation, Gavaldà started a yearly seminar on language technology through the Consorci Universitat Internacional Menéndez Pelayo Barcelona. The summit, held in Barcelona, began in 2000 with a focus on language technology, each year tackling a relevant subdiscipline: machine translation, speech recognition, information retrieval, and more. Eventually, however, Gavaldà decided to expand the summit’s reach beyond language technology to consider topics at the intersection of technology and culture. In 2006, the curriculum started branching out, later including more amorphous concepts like rational choice, identity, the digital revolution, and the nature of time. Gavaldà wanted to start building bridges — between tech and culture, between techies and society as a whole — in hopes that all parties would be better equipped to make decisions about the future we want to build.
“Part of why I enjoy doing this is because I think there are ways of making any topic accessible to the public at large without simplifying it. Certainly with things like ML and AI, that we do know will have huge consequences for society, I think it’s important that we all engage in a dialogue,” said Gavaldà.
Whether working on machine intelligence at Square or encouraging cultural intelligence abroad, Gavaldà is caught up in conversations — conversations about finance, about language, about the future. Sometimes he’s analyzing them; sometimes he’s lending his own voice. In any case, the dialogue goes on.
Stay up to date on all things data science here.