Back to overview

Article

Why Natural Language Processing is the best solution

Consider these two phrases:

"I like films."
and
"I love movies."

What do we know about the person or people who produced these phrases? Nothing. At least, on the face of it, we know nothing. For all, we know these are just random phrases conjured from thin air. However, if we look more deeply, we can, with varying degrees of accuracy, make some inferences about the people behind the language.

First, let's look at the similarity of the phrases. They both have the same syntactic structure: pronoun, verb, and noun. And semantically they both mean the same thing: a person is saying they enjoy the cinema. So, at the surface level, we have the same information from the two phrases.

Let's go deeper and look at how the phrases differ. Take the verbs "like" and "love". Love is certainly a much stronger, and more expressive, word than like. So, perhaps, the person who used "love" is more extraverted than the person who used "like". And the nouns: "film" vs "movie". What can we say about the person behind the noun choice? Well, "film" is more commonly used in Britain and "movie" is more commonly used in America. Sobe, the person who said "I like films" is British. And the person who said "I love movies" is an extraverted American.

English, of course, has many words that mean the same thing. Notice how I used a different verb and noun, "enjoy" and "cinema", to describe the two example phrases. If a third person was to say "I enjoy cinema" what would you picture in your mind about the person behind the phrase?

Welcome to the world of "Natural Language Processing" (NLP). It's "Natural Language" because the language is produced by humans. And we "Process" the language through a system, usually via software, to produce some output. In the case of Pera our NLP systems output predictions of a candidate's competencies, which in turn can be used to predict how well a candidate fits for a particular job.

We're, usually, not thinking about the process of writing or speaking. We are just communicating, subconsciously producing language. By language I mean not just the words we use but also looking deeper into how the language is formed. This includes properties such as the order of the words, the syntactic structure and length of sentences, choice of verbs or adjectives, how concrete or not our language is and many other features. Taken together the features we extract from a person's language form a linguistic fingerprint. We don't, consciously, notice a person's linguistic fingerprint in day-to-day conversations just like we don't, normally, look at a person's actual fingerprints. But Pera's algorithms, based on 6 years of R&D, do notice and extract hundreds of features from a person's language to produce their unique linguistic fingerprint. And from their linguistic fingerprint, we can make predictions about the person behind the language.

In human-to-human interactions as soon as we hear or see another person, we start to make conscious and sub-conscious assumptions about the person. We may read a CV, notice the university is not top-ranked, and assume about intelligence or motivation. Also, my assumptions will be different from yours and probably different from my own tomorrow. Humans are changeable and, no matter how hard we try, biased.

At Pera, we only process natural language produced by candidates. We don't process candidate CVs, videos or speeches. That is, we don't hear or see a candidate. This removes most of the bias found in traditional assessment methods. A candidate is only assessed on the language they produce as answers to three open-ended questions. There's no time pressure on candidates to answer and it's the same questions for all candidates. It's unique assessment candidates love - we consistently score 95% on satisfaction. Everyone gets a fair chance to produce their best answers and is compared to their peers on a level playing field. Further, our algorithms, and our assessors, can't have a bad day. The same assessment is consistently applied to all candidates.

Our AI is not only unbiased it is also extremely quick and working 24/7. Each natural language assessment is processed in a matter of seconds. So our HR clients always have the most up-to-date results to review and never miss a good candidate. Consider the case of a recruiter with one vacancy to fill but 1,000 applications. How can the recruiter fairly assess all 1,000 applications in a reasonable amount of time? They can't. Besides the obvious solution of throwing all the CVs in the air and interviewing based on where they land, what can they do? They can use Pera Skope to quickly and fairly assess each candidate. Pera Skope identifies the top candidates, gives the recruiter insights into each candidates' strengths and weaknesses, and even suggests questions to ask and areas to probe as preparation for an in-person interview. So, by using Pera Skope our recruiter has greatly improved the efficiency and fairness of their recruitment process. And they've used a unique technology that candidates love.

In summary, by using a natural language assessment for recruitment we produce results which are quick, and fair and give the candidates a unique experience they enjoy.

I hope I have piqued your interest in natural language assessment. If you'd like to explore NLP, here are some suggestions for further reading:

  • Daelemans, Walter. "Explanation in computational stylometry." International conference on intelligent text processing and computational linguistics.
  • Hirsh, Jacob B., and Jordan B. Peterson. "Personality and language use in self-narratives." Journal of research in personality
  • Pennebaker, James W. "The secret life of pronouns: What Our Words Say About Us" Bloomsbury