NLP: A Time for Reflection or a Way to Shape Decades of Hyperbole and Handwaving?

August 2, 2020

The most unusual online information service published “The Field of Natural Language Processing Is Chasing the Wrong Goal.” The article comments about the Association for Computational Linguistics Conference held in July 2020.

The point of the write up is to express concern about the whither and why of NLP; for example:

My colleagues and I at Elemental Cognition, an AI research firm based in Connecticut and New York, see the angst as justified. In fact, we believe that the field needs a transformation, not just in system design, but in a less glamorous area: evaluation.


Yep, the discipline appears to be chasing benchmarks. DarkCyber believes this is a version of the intra-squad rivalries as players vie to start the next game.

The write up raises this question:

How did the NLP community end up with such a gap between on-paper evaluations and real-world ability? In an ACL position paper, my colleagues and I argue that in the quest to reach difficult benchmarks, evaluations have lost sight of the real targets: those sophisticated downstream applications. To borrow a line from the paper, the NLP researchers have been training to become professional sprinters by “glancing around the gym and adopting any exercises that look hard.”

The answer, in part, is for NLP developers to follow this path:

But our argument is more basic: however systems are implemented, if they need to have faithful world models, then evaluations should systematically test whether they have faithful world models.

DarkCyber’s view is that NLP like other building blocks of content analysis and access systems have some characteristics which cause intra-squad similarities; that is, the players are more similar than even they understand:

  1. Reliance on methods widely taught in universities. Who wants to go in a new direction, fail, and, therefore, be perceived as a dead ender?
  2. Competing with one’s team mates, peers, and fellow travelers is comfortable. Who wants to try and explain why NLP from A is better than NLP from B when the results are more of the same?
  3. NLP like other content functions is positioned as the big solution to tough content challenges. The reality is that language is slippery and often less fancy methods deliver good enough results. Who wants to admit that a particular approach is “good enough.” It is better to get out the pink wrapping paper and swath the procedures in colorful garb.

NLP can be and is useful in many situations. The problem is that making sense of human utterances remains a difficult challenge. DarkCyber is suspicious of appeals emitted by the Epstein-funded MIT entity.

Jargon is jargon. NLP is one of those disciplines which works overtime to deliver on promises that have been made for many years. Does NLP pay off? This is like MIT asking, “Epstein who?”

Stephen E Arnold, August 2, 2020


Got something to say?

  • Archives

  • Recent Posts

  • Meta