Job′s with a future: More and more linguists will be needed all over the world, but why ?
With the advance of voice control, the demand for computational linguists is also growing. The systems now also learn dialect and “sociolectâ€Machines now speak different languages - but there is often a lack of diversity in programming. In Silicon Valley many have proclaimed the "voice revolution". Lower the shutters or play music with Artificial Intelligence (AI) ? The consumer world can be so simple. But a lot of manpower is necessary for the voice control to work smoothly. At Amazon alone, 10,000 employees work on optimizing Alexa.Speech recognition means a lot of manual work: audio recordings have to be transcribed, texts dismantled, words have to be labeled. This rather stupid work is done by contract workers all over the world, while the programmers and computer linguists, who are mostly employed in-house, are responsible for more complex tasks such as developing language models. Their job is to fine-tune the machine.Since the computer pioneer Joseph Weizenbaum presented the first chatbot with Eliza in 1966, voice computers have continued to develop. Alexa can even whisper or tell jokes today. As Amazon Vice President Steve Rabuchin said, there is even a team working on the "personality" of Alexa.Lessons for machinesThousands of employees also work on the language software at Apple and Google. There is even a real personal sketch for the Google Assistant, as developer James Giangola once revealed to the Atlantic: "She is the youngest daughter of a librarian and a physics professor who has a Bachelor of Art in history from Northwestern (University)." A very sophisticated voice computer that accompanies you in everyday life. Cars, hotels, schools, hospitals - voice control is penetrating more and more areas of society. This also increases the need for computational linguists.
At Amazon at the end of February, dozens of positions for "data linguists" were advertised for various languages. The task profile includes phonemic transcription and data processing. Apple is also looking for software engineers for its Siri language assistant who have knowledge of Natural Language Processing (NLP) and who can give the computer language lessons. The new methods of language processing are no longer based on abstract grammar rules, but on the acquisition of the mother tongue. Like a toddler, artificial intelligence (AI) learns to recognize syllables and words, which it then combines according to a probability model. Siri now "speaks" over a dozen languages, from Cantonese to Hebrew to Thai. Alexa even speaks several English, French and Spanish dialects. From Hindi to Icelandic But AI is not quite that polyglot yet. Language assistants have problems with Swiss German, and there are also problems with Arabic because there are neither usable dictionaries, audio recordings nor a pronunciation database for the Arabic dialects that could be accessed. But the technology is getting better and better. Since 2019 Alexa has also been answering questions in Hindi, which has more than half a billion speakers in India. Tech companies focus not only on the large "language markets", but also on niches. Years ago, for example, Amazon expanded its text-to-speech function Polly, which converts texts into speech, to Icelandic and advertised a position for a linguist for Icelandic. University professors were enthusiastic: Icelandic, which is spoken by 330,000 people and constitutes the identity of the people on the island, is increasingly being replaced by English in everyday life. Of all things, language computers, which actually only "understand" programming language, could now save regional languages and dialects from extinction. Alexa has also been answering questions in Hindi, which is more than half a billion in India Of all things, language computers, which actually only "understand" programming language, could now save regional languages and dialects from extinction. Problems with the sociolect But where voice control preserves local color on the one hand, it increases social inequalities on the other: According to a study by Stanford University, automatic speech recognition systems understand blacks much more poorly than whites. The researchers examined voice recognition systems from Amazon, Apple, Google, IBM and Microsoft on the basis of almost 20 hours of audio material. Result: The system misunderstood around 19 percent of the words of the white speakers. In contrast, the error rate for blacks averaged 35 percent. The reason for the misunderstandings: the sociolect, the African-American Vernacular English spoken by African-Americans in urban areas. Instead of "I mean" (I mean), the speakers often say the synonymous "me mean". Scientists are therefore calling for the African-American Vernacular English to be given more consideration in the training data (which can only be achieved with diverse development teams). In order for the language assistants to be able to answer our questions, we have to reveal a great deal of our lives and provide a great deal of data for processing - or have it subtracted somehow. Got it ? Do not loose an opportunity to sharpen your language skills !