Spanish and Portuguese are Romance languages derived from vulgar Latin and share many similarities, but they also have key differences in grammar, phonetics, and vocabulary. It is a case they are comparable to the Germanic languages, German and Dutch.
I was the 60th employee hired by Nuance Communications, a startup spinoff of the Stanford Research Institute (SRI), and its mission was to commercialize products and services related to Automatic Speech Recognition (ASR) at a time – in the late nineties – when “talking to machines” still sounded like science fiction. The company had survived following its initial investment phase and was now desperately trying to prove that this technology could be profitable in the contact center industry.
Despite the many myths surrounding this new technology, there was nothing magical about it. Part of my job involved collecting hundreds, sometimes thousands, of recordings of telephone users interacting with contact center operators. From this natural language database, it was necessary to classify and transcribe all the ways people could ask the same question. For example, “The time of the next flight to New York.” As I said, this query could be phrased in many ways:
● “Can you tell me what time the next flight to New York departs?”
● “I want to know what time the next flight to New York is, please?”
● “When is the next flight to New York?”
● “What time does the next flight to New York leave?”
● etc.
Following specific coding rules, each of these phrases had to be written in a text file, also registering the percentage of times speakers used each variation. The resulting file was called “grammar”, and these files were the input that a sophisticated Markovian network needed to decode in real-time, determining whether the incoming sounds resembled the phonetics of any of the transcribed and encoded phrases. I know this sounds like Mandarin Chinese, but for this story, there’s no need to delve deeper. One thing was the recognition software, and another entirely different thing was the professional services required to develop the grammars that would feed the recognition software. In this context, the value of software licenses was proportional to the number of phone lines the contact center allocated to the automated service, while the value of professional services was proportional to the time spent developing the grammars.
“There’s something funny about my last name, you know?”
For over three months, I had attended James’ training sessions. He was the expert grammar builder and was responsible for inducting new employees into the company. We had taken a break. From the cafeteria window, you could see the Sun Microsystems building. It was a bright July day – I remember it well. Not far from there were the offices of Cisco and Microsoft.
“Giangola?” I asked, “What’s funny about Giangola?”
“Do you speak Portuguese, Carlos?”
“Very little… I mean, I understand some words and phrases, but I can’t speak it.”
“I’ll tell you something: years ago, when I was studying for my Master’s in Linguistics, I based my thesis on the Portuguese language, you know? As part of my thesis work, I needed to collect data in Brazil, so I wrote to the Language Faculty at the Pontifical Catholic University of São Paulo, asking for permission to work there for a few months. Word spread quickly: James Giangola was coming to conduct linguistic studies. The funny thing is that phonetically, my last name sounds very similar to the phrase ‘de Angola’ in Portuguese – that is, ‘James from Angola’. So, no one expected an American; everyone thought an African professor would be coming to work at the university, you see? ‘A professor from Angola’…hah”, James laughed, enjoying the joke.
The next morning, I had an early breakfast and got into the Pontiac Firebird I had rented. If I was lucky, it would take no more than 15 minutes to drive from the Best Western Inn Los Gatos to Nuance’s offices.
Around midday, I ran into the Vice President of Sales for South America. He looked upset.
“Everything okay?”
“No. Everything’s a mess,” Pablo responded, visibly irritated. “They refused to sign the contract.”
“Who?”
“The executives from Algar… They changed their minds. I don’t know why. They’re upset about something.”
Algar Telecom was a telecommunications company operating in Uberlândia, in the state of Minas Gerais, Brazil. Pablo was taking them to lunch and then to their hotel. Their return flight was scheduled for 8 p.m. “Want to come?” he invited me. I gladly accepted.
The fish was delicious. Salmon has always been my favorite, and I enjoyed it thoroughly. I was seated between two Algar executives, trying to communicate in Spanish – my native language – and English. Slowly, I began to understand why they were upset. It was a problem of perception. As I mentioned, back then, there were many myths about ASR. The Brazilians’ perception was that the recognition software could understand anything spoken to it, all on its own. Since it was a preconceived idea, every time Pablo tried to explain the project’s importance – the importance of grammar development – they reinforced their perception. They thought Pablo was trying to sell them something they didn’t need, and for no less than two million dollars. No wonder why they were upset. I realized we needed to break that perception, but I also understood that if grasping the technology was already difficult, it was even harder to do so in a non-native language. I stood up, looked at everyone at the table. “We must go back,” I said with great determination. “We need to return to Nuance.” Thank God I managed to convince them.
In the meeting room, the rhythm, tone, and nasalized sounds of Brazilian Portuguese filled the air. With infinite patience, James was able to build a bridge between the executives’ perception and the true essence of ASR technology. Finally, they signed the contract. I felt proud of myself.
It was one of the first deals I was involved in.
After a few years, I accepted a job offer and returned to Chile, my home country. I know Nuance is not a household name, but let me tell you something: in 2010, Apple signed a contract with Nuance. Together, they developed Siri, the first virtual assistant in history. It debuted a year later on the iPhone 4S.