Rhetorical structure of writing

See also Writing.

• Cotos, E., Huffman, S., & Link, S. (2015). Furthering and applying move/step constructs: Technology-driven marshalling of Swalesian genre theory for EAP pedagogy. Journal of English for Academic Purposes, 19, 52–72. doi:10.1016/j.jeap.2015.05.004

  Proposes a schema for tagging “moves” and “steps” in IMRD academic reports. Moves are specific rhetorical tasks, like identifying the niche the study fills, establishing credibility, and framing the new knowledge identified in the study. Steps are specific components of these moves, like highlighting a gap in previous work or describing the data analysis to be done. The schema is based on a review of real papers from a bunch of disciplines, and they propose using the schema pedagogically by teaching students the moves and steps explicitly. They also built a tool that tries to classify text based on the schema and shows feedback to students (somewhat like Docuscope Write & Audit), but little detail is given on how the classification is done.

• Fiacco, J., Cotos, E., & Rosé, C. (2019). Towards enabling feedback on rhetorical structure with neural sequence models. In LAK19: Proceedings of the 9th international conference on learning analytics & knowledge (pp. 310–319). doi:10.1145/3303772.3303808

  Builds a neural network model to automatically classify sentences using a move/step model. The network layers:

  1. Word embeddings using GloVe
  2. A bidirectional LSTM applied to each sentence’s word embeddings, to produce a single output for each sentence
  3. A bidirectional LSTM applied to the sentences, to produce new outputs for each sentence in the document, which are context-dependent (i.e. depend on previous and future sentences)
  4. A conditional random field layer to model transition probabilities between tags.

  There’s then some work trying to identify which neurons contribute most to discrimination between steps, though I’m not sure what this is useful for. But ultimately they suggest giving feedback to students by comparing “the expected progression of moves and steps” to those used by the students, suggesting the specific modification to make to improve the structure.

• Fiacco, J., Jiang, S., Adamson, D., & Rosé, C. (2022). Toward automatic discourse parsing of student writing motivated by neural interpretation. In Proceedings of the 17th workshop on innovative use of NLP for building educational applications (BEA 2022) (pp. 204–215). doi:10.18653/v1/2022.bea-1.25

  A more complex neural network approach for analyzing rhetorical structure, this time using “Rhetorical Structure Theory”, which looks at structure within sentences as well as across the document. This forms trees of structural components within sentences and across the document.