Home » Blogs » AI Recruiting Technology » The AI Recruiter Who Uses ...

The AI Recruiter Who Uses Natural Language to Really Get It

By Eric Garnick, Software Engineer


Understanding natural language with a computer has many inherent challenges. Natural language AI goes far beyond what traditional word matching can do. Depending on your goals, you can start with simple keyword matching, but words, and combinations of words – phrases and sentences – do not always have an obvious or straightforward meaning when taken at face value:

I’ll go ahead and let you chew on that for a while.

Really? Where are you “going”? Is it really “ahead” of anything? Is there anything to “chew on”? How long is “a while”?  

In order to approach the necessary level of linguistic comprehension for a sentence like this, we need a much more sophisticated approach than keyword matching.  This is part one of a two-part article, so I’m going to start out by exploring some of the ways humans understand ambiguous language. Then in part two, I’ll consider how these techniques might be helpful for a computer’s understanding of natural language.

Basic rules of our language

One of our best sources of comprehension comes from our knowledge of the basic rules of our language.  For most speakers, this knowledge is more implicit than explicit, but it’s still there even when you can’t identify the rules that apply.  Consider the following simple cases:

  1. Coordination: I can repair electronics and operate hand tools and heavy machinery.
  2. Negation: I have worked in a cafe, never in a restaurant.
  3. Spelling: I have on year of experience.

Without some basic linguistic understanding, naïve word-matching may not be able to determine that the first sentence refers to skills including {repair electronics, operate hand tools, operate heavy machinery}, or that the second sentence indicates work experience including only {cafe work}, or that “on year” is probably a misspelling of “one year” in the third sentence.


Context and prior information

Slightly more advanced, we can use context and prior information to facilitate language comprehension.  In the following examples, some information would be lost without being able to connect knowledge from separate sentences or parts of a sentence, or having some knowledge about the relevant vocabulary for a given topic:

  1. I was a sales lead at ABC Corp. from 2014 to 2016 and have been doing the same thing at XYZ Inc. since then.
  2. I received my bachelor’s in finance in 2010.  
  3. I’ve been a financial analyst at National Realty Group since college.
  4. I was in charge of the servers at a bank. – (Applying for an IT position)

Disregarding the fact that the first two sentences require some conception of time (a whole separate challenge), it would be impossible to fully understand what the first speaker did or for how long without being able to recover sales lead from “the same thing”, or 2016 to the present from “since then”.  Similarly, the second example requires connecting receipt of a bachelor’s degree in 2010 from the first sentence with “since college”.  And while the there are other reasons why most people would only interpret example 3 as overseeing web-servers at a financial institution, knowing that there is an IT job being considered helps in eliminating the possibility that the sentence refers to managing waiters and waitresses by a river.

Real-word knowledge

Finally, real-world knowledge is an indispensable part of understanding language.  This may be as basic as knowing that the ice is cold, to something as subtle as knowing that if someone says “brrr” in room with the windows open, they may be cold and want you to close the windows.  The following examples are a little more relatable to conversations around job applications:

1. Q: Do you have a reliable means of transportation?

    A: I don’t have a car, but my apartment is along the bus route.

2. I’d like a job in the Bay Area

As obvious as these are to any competent reader, once you break the meaning down, they are not so obvious to a computer.  The answer in example 1 would not be understood as “yes” without understanding that living along a bus route implies a reliable means of transportation and that describing an accessible, reliable means of transportation is equivalent to affirming that such transportation is available.  In example 2, one needs regional knowledge to know that the Bay Area refers to San Francisco, CA and the surrounding cities rather than the area of the Massachusetts Bay, or the Saginaw Bay in Michigan, or any of the thousands of other bays in the world.

In this first article, I’ve given you a some idea of the challenges of understanding natural language, and how we as humans can overcome these challenges.  Part II of the article will look at how these types of understanding can help an automated system, and how we’re making sure our AI recruiter at AllyO really gets it.




Be sure to catch AllyO at this year’s HR Technology Conference! 

AllyO At HR Tech


Stay up-to-date with the latest insights and trends from AI recruiting brought to you by AllyO Blog!

Schedule Demo Widget

Ready to transform your HR communications?

Learn More

AllyO Virtual Hiring Events™Learn More