Natural Language Processing
What does it do?
Natural language processing is the study of human language as it can be understood by computers - words and their grammatical context being put into a strict
logic, but also more importantly strict logic that allows for the understanding of more fuzzy logic words and their grammatical context.
Natural language processing has its challenges, in that the languages that real people use do not often conform to formal language structure
(Kamp & Reyle 1993). If they did, then translating them into data that a computer can process for its own tasks would, hypothetically, be roughly as simple as
converting a program from one language to another. Attempts to push real-life language into unambiguous, perfectly rational and logical form, have mostly failed so far
(JONES 2012), and so natural language processing is the endeavour to move in the other direction, and form a digital understanding of the language(s) that people already use,
that can then be converted into a formal language.
Natural language processing has a rich history dating back to the 1950s, but currently uses neural natural language processing(Gomez-Perez, Denaux & Garcia-Silva 2020),
which involves feature learning and machine learning methods in the style of deep neural networks to process terabytes of examples of human text.
This would logically be supervised learning (Zhou 2018), meaning it requires manual correction for mistakes, because unsupervised learning would be circular in logic -
it would require a machine that already knows valid language in order to validate the language.
Natural language processing is found most commonly in daily life in personal virtual assistants (Ana Berdasco et al. 2019), such as Alexa or Google Assistant.
These apps listen at all times for a command which is intended to be difficult to encounter in ordinary conversation. For Alexa, it's "Hey Alexa," and for Google Assistant,
it's "Okay Google." They send the recording over the internet to their respective servers for a two-step process: First the voice is run through voice recognition (Kamath,
Liu & Whitaker 2019), and then the voice, now text, is run through the natural language processing of the home servers. The additional computing power allows it to be done
with minimal turnaround time. The natural language processing allows the server to guess with increasing certainty what it was that the user meant for their virtual assistant
to hear, and then to guess again what it was they wanted their virtual assistant to do. The virtual assistant then executes the orders sent by the home server.
Virtual assistants, in this manner, have a variety of functions, including playing music, turning on or off devices, setting timers and other reminders, as well as many others
(Ana Berdasco et al. 2019). In this manner, and using only their voice, a user can enact various kinds of utilitarian tasks that can be done by appliances and other machinery in their
home.
Natural language processing is also used by Facebook (‘Facebook wants you to put a Portal camera and microphone in your home’ 2018) for advertising purposes. While users have
the Facebook app installed, the Facebook app listens for all words said in proximity to it, and sends that over the internet to its home servers. Similar to Alexa and Google
Assistant, Facebook's high computing power allows it to use a two step process, that of converting the speech to text, and the text to language data, rather efficiently.
Facebook corporate then stores this data as market research, which it uses for targeted advertisement placement as well as sellable market research for other companies.
What is the likely impact?
A major way that NLP could impact the world is that as computers become better at understanding complex “human language”, such as differentiating between contexts, e.g. when one yells “fire” at a building that is on fire versus “fire” when one instructs a marksman to shoot (Kamath, Liu & Whitaker 2019). Another example would be when one says the word “sheep” and it looks in its language database and finds the word “sheep”, but will probably then act on the assumption that a sheep is black, because the colour of sheep is commonly listed as black. It won’t induce that the sheep is white because it’s rare that we have to say a sheep is white, as it is perceived as ‘common-sense’.
It is when that computer can understand ‘common-senses’ and important contexts in when words are said that we will be able to have seamless conversations about deep subjects and with emotions. Robots will be able to have AI that can talk in a manner indistinguishable from a human. We could see NLP integrated even more into our everyday life; it could lead to a major upheaval of the service sector, as we could potentially see AI with advanced NLP replace the need for actual humans.
How will this affect you?
There are two potential ways that NLP in combination with AI will affect both the individual and the labour force as a whole.
(Frey & Osborne 2017; Frank et al. 2019; Arntz, Gregory & Zierahn 2016; Pajarinen, Rouvinen & Ekeland 2015; Brzeski & Burk 2015;
‘The computerisation of European jobs’ n.d.; James Doubek 2017; Acemoglu & Restrepo 2017)
- Optimists suggest that technology may substitute for some types of labour, but that efficiency gains from technological augmentation outweigh transition costs, and, in many cases, technology increases employment for workers who are in not direct competition with it. Furthermore, the skill requirements of each job title are not static and actually evolve over time to reflect evolving labour needs. For example, workers may require more social skills because those skills remain difficult to automate. Even if technology depresses employment for some types of labour, it can create new needs and new opportunities through “creative destruction”. For instance, the replacement of equestrian travel with automobiles spurred demand for new roadside amenities, such as motels, gas stations, and fast food.
- Technology improves to make human labour more efficient, but large improvements may yield deleterious effects for employment. This obsoletion through labour substitution leads many to worry about “technological unemployment” and motivates efforts to forecast NLP and AI’s impact of jobs. One study (Acemoglu & Restrepo 2017) assessed recent developments in NLP and AI to conclude that 47% of current US employment is at high risk of computerisation, while a contrasting study, using a different methodology, concluded that a less alarming 9% of employment is at risk (Arntz, Gregory & Zierahn 2016). Similar studies have looked at the impact of automation on employment in other countries and reached sobering conclusions: Automation will affect 35% of employment in Finland (Pajarinen, Rouvinen & Ekeland 2015), 59% of employment in Germany(Brzeski & Burk 2015), and 45 to 60% of employment across Europe (‘The computerisation of European jobs’ n.d.). Critics have complained that prospective studies lack validation, but retrospective studies also find that robotics are diminishing employment opportunities in US manufacturing (Frey & Osborne 2017).