- "be it in" awkward language
- ! Suggest starting by giving a (brief!) sentence about the "what
  people say" aspects of understanding, so that people know what
  you're distinguishing against.
- "At the same time, humans do not" is awk transition, because it
  suggests that "keywords" occur /in opposition to/ what you say in
  the first paragraph. Suggest the following sentiment: "When humans
  analyze rhetorical purpose, they do so deeply, not merely relying on
  the presence/absence of key words".
- Fig 1 caption could benefit from a concrete example of what a
  hypothesis looks like. ("e.g. that the author's word choice suggests
  blame for a particular group.")
- +In this abstract, I describe+ "The RASHI system is a computational
  model of this human ability. It recognizes and syst..." Suggest this
  so as to explicitly connect RASHI to the previous paragraph about
  what humans do.
- Decisions/choices spade.
- Summarized hypothesis? maybe "forms a hypothesis of the author's
  overall intent which it describes to the user as a textual
  brief/summary."
- to shorten: "complex nested representations, all of which involve
  humanly-plausible operations."
- As a first step -> In the first step (stylistic prefrence)
- comprised of -> composed of (though there's evolving usage)
- "simple English, for example the following news story titled"
- [The typeface switching is a presumably temporary visual crime. I
  suggest using color instead, or not having running threads
  throughout---they might be more effective in speaking.]
- "RASHI processes the textual input using the START parser, yielding
  hierarchical semantic parses."
- "This inner language allows ... presence in the text" could be made
  more clear. Perhaps something about taking into account
  subject/object roles in the sentence.
- to find instances of -> to collect all potentially meaningful cues
  about [style]
- At this stage, RASHI ... -> At this stage, potential evidence is
  merely collected, not evaluated. The collected evidence is assembled
  into a /composition story/ for further processing.
- from the above excerpt -> with the above excerpt [stylistic]
- "it must interpret it as indicative of the author’s intention to
  modulate meaning." is a little awk. Suggest "After gathering textual
  evidence, RASHI must evaluate which evidence indicates the author's
  intent."
- "subjectivity model, a natural language"
- !! following elaboration graph goes missing. Better to say
  "generates the elaboration graph shown in Fig 4." What's an
  elaboration graph?
- As the last step -> Having determined the significance of the
  evidence, RASHI must distill potentially conflicting cues to get an
  overall portrait of authorial intent. [The key is describing the conflict-resolution]
- Maybe add a line like "RASHI counts various instances for or against
  individuals in the story, then uses a hand-coded decision tree to
  yield its final summary of authorial intent."