Humans, machines and semantics
According to the Oxford Dictionary, king is defined as “the male ruler of an independent state that has a royal family”.
If we were now to change just one word in this definition from male to female, the female ruler of an independent state that has a royal family, what would that mean? Queen, correct! It is also the first word that comes to mind. Semantically speaking, a king can be defined as a combination of a man and royalty in an abstract form. Likewise, a queen is a combination of a woman and royalty. If we know the abstract formulation, we can easily define that a prince is a boy plus royalty and a princess is a girl plus royalty.
Thanks to years of evolution and interaction with languages from the day we are born, we have managed to communicate even in abstract forms. For us, finding such semantic relationships is self-evident. But what if we ask the same question of an artificial intelligence (AI) system? The problem then becomes incredibly difficult, and the answer lies in the rapidly growing subfield of AI, natural language processing, commonly known by the abbreviation NLP. NLP enables modern computers to interpret human language, establish semantic relationships between pieces of text and thus communicate with humans.
The roots of NLP go back to the 1900s, when the Swiss linguistics professor Ferdinand de Saussure developed the idea of the patterns and functions of a language. Saussure argued that a language is a social phenomenon, since it is a structured system that can be viewed synchronically and diachronically. In other words, a language exists at a specific point in time and changes over time. He held that meaning arises within language, between relationships and contrasts through humor, sarcasm, irony, etc., where the literal meaning is far removed from the intended meaning. The Oscar-nominated film “The Imitation Game” (2014) is based on the biography of the British mathematician and early computer scientist Alan Turing, who is widely regarded as the father of theoretical computer science and AI . In 1950, he wrote a paper describing a test for a “thinking” machine. He developed a test to evaluate the machine’s ability to exhibit intelligent behavior equivalent to that of a human. Turing proposed that a human examiner should conduct a conversation with two players, a machine and a human. If the examiner could not tell which player was a machine, the AI system had successfully passed the Turing test . Turing’s work laid the foundation for today’s NLP systems, as can now be seen in the advances in the chatbot industry.
“Machines may have passed the Turing test, but they still have a long way to go before they can start generating thoughts.”
NLP: science and industry
Between 1950 and 1980, progress in the field of NLP was limited by computing power and the availability of data. Today we have both in abundance; consequently, algorithms have achieved significant improvements in text processing. The 2019 data from the “Annual Conference of the Association for Computational Linguistics”, one of the most important conferences in linguistics, show a dramatic increase in the number of submitted research papers on NLP. The focus of AI research groups almost doubled from 2018 to 2019.
Today there is a substantial number of software packages and code libraries that enable the development of effective NLP algorithms. Although NLP has been around for several decades, it is fair to say that it has only just begun to make small advances!
In view of the unprecedented growth of NLP in academia, industry is also rapidly deploying machine learning systems to take advantage of NLP. This can be used for various tasks such as sentiment analysis, topic extraction, text summarization, semantic similarity, analysis of unstructured documents, content classification, categorization of news articles, analysis of product descriptions, etc. NLP has become a bridge of communication between computers and humans. Machines now have a framework that enables them to understand us better. With the introduction of advanced AI methods such as Deep Learning the industry is recording exponential growth driven by big data, computing power and increasing interest in communication between humans and computers. The global market for natural language processing is estimated at 34,80 billion by 2025 and is recording an annual growth rate of 21,5 % in the period 2020-2025.
“AI will become our most important companion in everyday life, similar to our mobile phones, only more intelligent and smarter.”
The relationships between words
To return to our question: How can an AI system understand that royalty plus a woman equals a queen? The answer lies in “word embeddings”, a type of word representation that allows words with similar meanings to receive similar representations. It is a representation of words in a high-dimensional vector space that captures semantic relationships between words based on their position in a sentence. Bengio et al. (2003) published a research paper, in which they introduced the concept of word embeddings, which today forms the basis for various NLP algorithms. The following illustration explains the answer to our question. The semantic relationship “the king is to the queen as a man is to a woman” is encoded in word embeddings. Similarly, word embeddings can also capture verb-tense relationships, “Walking is to walk as Swimming is to swim”, and country-capital relationships “Wien is to Austria as Rome is to Italy as Berlin is to Germany and so on…”.
Summary
The focus is increasingly shifting to the development of online applications that can process text information. The benefits of using AI methods are well proven across the industry and increase a system’s efficiency. We want to develop AI systems that understand the nature of human communication. It is easier to teach an AI our habits of life than the other way around. Although a language is a structured system, it is extraordinarily complex in itself. To understand a language, one must know numerous variables such as culture, dialects, history, science, art, etc. A perfect example of this is the moment that marked the friendship between Charlie Chaplin and Albert Einstein at the premiere of Chaplin’s film „City Lights“. A conversation is said to have taken place between the two thought leaders:
„What I admire most about your art“, said Albert Einstein, „is its universality. You do not say a word, and yet … the world understands you“.
That is true, replied Chaplin: „But your fame is even greater… the world admires you, even if no one understands you.“
Today, NLP has only just begun to demonstrate its capabilities, while the future will see an overwhelming integration of NLP into business scenarios. At STRG, we keep up to date with advances in NLP. Send us an email at office@strg.at to learn more about how NLP can be integrated into your company, or visit our website https://www.strg.at/ to get an overview of our services.
by Dadhichi Shukla