Who or what is NLP?

NLP is not a who, at least not in the context of this reflection. NLP refers to the what of Natural Language Processing. In this context, the target of the NLP is writing analytics (Gibson & Shibani, 2022) and educational discourse (communication in all forms, chat, prose, etc.) (Dowell & Kovanović, 2022).

This is a new approach to the application of NLP for me. In using voice-activated digital assistants for years (Indigo (then Lyra now Teneo.ai), Dragon, Cortana, Siri, Google Assistant, Bixby, etc.), my understanding was that NLP was centered around the processing of the spoken word. That was obviously a narrow view of the technology’s capabilities. While I still believe that the pacing, inflections, and nuances of the spoken word are more complex for software to decipher, I should have realized that that same powerful analytical engine could be pointed at text to begin with, and not as a result of some voice-to-text conversion. Why not just skip that first step and cut to the chase?

What is Writing Analytics?

Using NLP to analyze writing constitutes writing analytics (WA) which is a sub-field of learning analytics (LA) which is in turn a sub-field of artificial intelligence (AI).

Descriptive WA vs. Evaluative WA

There are many ways to evaluate and analyze anything. For example, the performance of a sports car can be outstanding in acceleration and cornering and garner high praise for on-track nimbleness while it may have a harsh suspension and zero cargo room which would garner low praise for the practicality of taking a road trip. Same car, yet two quite different measures of success.

Writing analytics works in much the same way. Descriptive WA evaluates the writing sample based on technical aspects such as word count, frequency of repetitive word usage, or number of sentences overall. This is much more like quantitative analysis. These metrics can ensure that a certain specification for a writing prompt is met, but this truly does little to inform on the quality of the writing itself. For example, just because you included 500+ words in the piece does not mean you stayed on topic or were even on the correct topic.

Evaluative WA, on the other hand, is a qualitative analysis that checks on the writing context and provides the writer with feedback on the written sample’s alignment with the writing prompt’s goals. Of course, evaluative WA then can return significantly different feedback depending on the context and specific purpose it is being used for. This feedback is often used to guide the writer to improve their content where this loop leads to continuous improvement via the feedback to action to new feedback to new action cycle iterates.

Writing to learn vs. learning to write

While both descriptive and evaluative WA have their place in LA, they are both ultimately just tools. For the learning environment to get the highest return on this investment, a properly optimized pedagogy or andragogy must inform the overall process. Without this oversight, the LA could give feedback that is irrelevant at best, and detrimental at worst. For example, quality instruction and coaching for how to swing at a pitched baseball (step toward the pitcher’s mound, start the swing from your back foot, etc.) will absolutely corrupt a swing at a golf ball on a tee.

In writing assignments, depending on the classroom prompt the assignment itself can be an exercise in mastering the art of writing – how to create proper sentence structure, gather and express thoughts, spell words correctly, use proper punctuation, etc. This immersive exercise would be classified as “learning to write” and here, the subject matter is not that important since the technical execution of the writing itself is the goal.

However, as the student progresses through the educational process, it will be assumed that the skill of writing no longer needs to be practiced. At this point, the student is now “writing to learn” in whereby putting their thoughts to the written word, documenting research, and exploring ideas are the focus of the writing prompt.

Whether learning to write or writing to learn, each of the reasons for the writing assignment have a different purpose. Because of this difference, the WA applied to provide feedback must inherently understand the reason for the assignment and adopt the feedback accordingly. This adaptation of the feedback generated could provide a combination of descriptive and evaluative WA. In this way, the LA needs to be as dynamic as the learners themselves.

We’re in this together!

Self-regulated learning (SRL) and socially shared regulated learning (SSRL) are used in some NLP applications aimed at understanding the relevance of the diverse word choices used when learners are working in isolation and when learners are working collaboratively (Dowell & Kovanović, 2022). By overlaying individual traces of educational discourse with group traces of educational discourse where the individual was a part of the group, and comparing the learning outcome achievements, it was shown that collaborative work had better results than groups that did not collaborate and interact.

Further evaluation of the data indicated that collaborative work leads to deeper inclusivity and acceptance of learners’ discourse across ethnic, gender, and cultural demographics.

Lasting impression

A key takeaway from these selections for me is that despite the rapid move to digital adoption forced on everyone (professionals and students alike) by the COVID-19 pandemic (LaBerge et al., 2020), is that even through the bits and bytes of the internet, collaboration is important in the overall development of the well-rounded learner’s experience. Digital assistance from AI can certainly help learners with many aspects of their education, but that human connection, even if over social media or collaboration software, is the icing on the cake that can elevate the success of a learning endeavor to the next level. Rick Springfield was right, we all need that Human Touch (Springfield & Springsteen, 1983).

References

Dowell, N., & Kovanović, V. (2022). Modeling Educational Discourse with Natural Language Processing. In C. Lang, G. Siemens, A. F. Wise, D. Gašević, & A. Merceron (Eds.), Handbook of Learning Analytics (Second ed., pp. 105-119). New York: Society for Learning Analytics Research (SoLAR). https://doi.org/10.18608/hla22

Gibson, A., & Shibani, A. (2022). Natural Language Processing – Writing Analytics. In C. Lang, G. Siemens, A. F. Wise, D. Gašević, & A. Merceron (Eds.), Handbook of Learning Analytics (Second ed., pp. 96-104). New York: Society for Learning Analytics Research (SoLAR). https://doi.org/10.18608/hla22

LaBerge, L., O’Toole, C., Schneider, J., & Smaje, K. (2020). How COVID-19 has pushed companies over the technology tipping point—and transformed business forever. McKinsey & Company. New York: McKinsey & Company. https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/our-insights/how-covid-19-has-pushed-companies-over-the-technology-tipping-point-and-transformed-business-forever#/

Springfield, R., & Springsteen, B. (Writers). (1983). Human Touch (Official Video) [Motion Picture]. https://youtu.be/yo0uTu2uLtI?si=xBTuVJadA7qpTEwG