Computational Verbal Behavior

To set the stage:

  1. There is no field, or specialty within behavior analysis, “Computational Verbal Behavior”, as far as I can tell.
  2. I am not a behavior analyst, though I played one in graduate school.
  3. I know very little about Skinner’s treatment of Verbal Behavior, though I did read the book once 40+ years ago
  4. I have been developing software for 40+ years, 20 commercially, and 20 on my own.
  5. I know little about AI or machine learning.
  6. But lacking skills or qualifications has never deterred me from doing something that I found interesting.

Okay, now that I have totally disqualified myself, we can begin.

For more information regarding how far I have lived from this topic, see my LinkedIn profile.

Natural language processing

I have been doing Natural Language Processing (NLP) of one sort or another on and off since the early 1990s when I did the internationalization/localization work on this search engine product: Callable Personal Librarian (CPL). The web pages recorded in the Wayback Machine are the only record of this product now. NLP can be contrasted to computational linguistics. Where NLP uses computational techniques to do useful work, computational linguistics uses the same techniques to study linguistic questions. I generally use the term “NLP”, incorrectly, to refer to both.

If we are creating analogous terms for behavior analytic work with verbal behavior, we might in fact want a similar split between the science and the technology. However, there is so little work findable on the web in any way related that it really does not much matter, I think.

There is a LOT of software, free and otherwise, out in the world for doing NLP. It is all based on traditional linguistics and cognitive science. For example these: 12 open source tools for natural language processing. Note that for a lot of such work, and AI work, Python is the preferred language. The APA seems to be favoring Python as a “tool” language: A brief introduction to Python for psychological science research.

Why take a behavior analytic approach?

The question is: what could a behavior analytic approach provide that these legacy approaches do not?

I don’t know. I am assuming that as in many other areas, behavior analysis would bring more effective techniques than those currently available, or would in some way advance the science/technology. But what are those techniques, or in what situations might it be useful to develop software to bring behavioral techniques to natural language processing? Or do we just continue to cede this area to cognitive sciences?

I have discussed (directly or indirectly) the issue of defining or creating a field of computational verbal behavior with a few folks who are familiar with both verbal behavior and computing. They are (or may be) doing work that might fall into the category of computational verbal behavior, but it is within a proprietary commercial setting (I think). We discussed doing work together as part of my day job, but I probably bungled that opportunity. So it is entirely possible that there is work being done in the area of computational verbal behavior of which I am unaware. A lot of haziness there.

In our discussions, there seemed to be no clear consensus regarding the value of software based on Skinner’s treatment of Verbal Behavior, per se. Their efforts are more focussed on artificial neural nets, and not of the behavior analytic variety. In contrast, see this “bio-plausible” neural net approach in JEAB: A selectionist approach to reinforcement (Donahoe, Burgos, and Palmer, JEAB, 1993).

An Exploration

This will be a sequence of blogs exploring the possibility of bringing behavior analysis to the software treatment of verbal behavior, especially in a freely available open source manner (or as close to it as possible).

I have been meaning to re-read Verbal Behavior. But work, demands of life, plus I am old and of low energy. But recently I learned that Andy Lattal is offering a verbal behavior course at WVU. Might I take that course, non-credit? Yes, I might. The senior citizen price to take any course at WVU non-credit is a flat fee of $50. Whoo-hoo! Oh, but in person classes, and not starting until 17:00. By that time of day I am pretty much a vegetable (I get up between 3-4:00, and I’m old). Not going to work out for me.

Andy did supply me with a syllabus. Hmmm, think I will follow along in the readings and occasionally bug Andy with questions and comments. Geez, I’m already way behind.

For the foreseeable future, these VB blogs will be my blathering about verbal behavior and computational aspects. Don’t expect too much scholarship here. Feel free to lambaste me in comment replies to the postings. Start your replies with something like, “Tom, you ignorant …”.

A resource

This webinar on Friday, September 17, 2021, looks promising. I’ll continue trying to register, but as a computer illiterate I am having great difficulty with the site:

Natural Language Processing: Quantitative and Computational Analyses of Verbal Behavior

FYI: the speaker, David Cox, is also the owner of this Slack channel:

Behavior Analysts Who Code

Blog Category

On the main page, find postings under category “Computational Verbal Behavior“, starting with posting “I’m Ba-a-a-ack.