Language Analyzation Methods
Chapter 11 Discussion
Language Analyzation Methods
In this article, it was interesting to start understanding how language is analyzed in Natural Language Processing. Our text describes discourse analysis methods most commonly used: The Bag-of-Words approach, dictionary-based approaches, document-term matrix, probabilistic topic modeling algorithms, and group communication analysis.
The Bag-of-Words approach disregards grammar and word order and treats each word as its own entity. So, this method will count the number of times a unique word, or unigram, is mentioned and quantify the data based on those observations. This approach is commonly used for the “purpose of analytical model development”. Dictionary-based approaches use a catalog of predefined text to make comparisons to language. Our text mentions this is useful in determining cognitive load in a text. Document-term matrix is a method used to categorize language in a series of tables where word frequencies are attributed to rows of information. This method is useful for organizing clusters of information across different texts. Probabilistic topic modeling algorithms move away from 0=0 observational analysis and move into predictions by assigning probabilities to words belonging to various topics. With the advancement in deep learning, probabilistic topic modeling can make even more intelligent analyses.
And, finally, the group communication analysis which takes the focus to “multi-party discourse” (or in many instances, Social Media, online forums, or chat). I find this the most interesting and surprising where it lands in the trajectory of this type of machine learning. Multi-party discourse is a type of technology is ever-present in how social networks feed us recommendations and what they market to us. This is where this type of analysis intersects with our everyday lives and with much of my professional career up until this point. While I conceded that this type of technology can render many adverse outcomes( ie, I read about a new running exercise plan, and then I am fed with diet-culture propaganda) it also has the ability to bring me to concepts, products, human groups that I would not have the connection to otherwise. This is what I find so exciting about this type of work- the ability to harness data and information and use it to make even the most mundane elements of human-like more informed, thus more potent or productive.
Scaling of Discourse Analytics
The application of NLP tools has been used to “quantify aspects of learner-generated posts, as well as learners’ cognitive, affective, and social processes.” With the data this type of analysis yields, researchers have been able to identify correlations between student affect in how it reads in their written engagement and make predictions on their likelihood to drop out or be retained. With this information, they are able also to offer strategies to intervene, like when confusion is detected, intervening to reduce frustration.
With class sizes increasing and teachers leaving the profession left and right, the integration of these types of tools can increase the efficacy of student observation for a personalized needs assessment, giving educators more time to do the work of intervention and let fewer students fall through the gaps.
Resource
Lang, C., Siemens, G., Wise, A. F., & Gašević, D. (2017b). Handbook of Learning Analytics. In Society for Learning Analytics Research (SoLAR) eBooks (2nd ed.). https://doi.org/10.18608/hla17
