In my last post, I asked ChatGPT-3.5 to revise a bunch of “loose” sentences. I would have preferred to ask ChatGPT-4.0, but I had used up the daily allotment of GPT-4 interactions that OpenAI allows non-paying customers.
So the next morning, the first thing I did was run those sentences past GPT-4. The results, interestingly, are closer to the original sentences than the GPT-3 versions are. And, unlike the GPT-3 versions, they don’t introduce any new infelicities. On the other hand, while GPT-3 tightened up all three sentences by eliminating their modifiers, GPT-4 only does this with the third.
- Original: In Temple Grandin’s Thinking in Pictures, it discusses how autistic people can be very visual in their thought processes.
- GPT-3: Temple Grandin explores how autistic individuals often rely on visual thinking in her book Thinking in Pictures.
- GPT-4: In her book Thinking in Pictures, Temple Grandin discusses how autistic people often have highly visual thought processes.
- Original: From talking with the student’s mother, it seems as though she is very satisfied with the accommodations he receives at school.
- GPT-3: Talking with the student’s mother suggests she is very satisfied with the accommodations he receives at school.
- GPT-4: From talking with the student’s mother, it seems she is very satisfied with the accommodations he receives at school.
- Original: For those individuals that are included with their regular education peers, they struggle more with accessing classroom reading materials because they are reading below grade level
- GPT-3: Those individuals who are included with their regular education peers struggle more to access classroom reading materials due to reading below grade level.
- GPT-4: Individuals included with their regular education peers often struggle more with accessing classroom reading materials because they are reading below grade level.
Instead, in the first two GPT-4 revisions, what we get are smaller edits that tighten things up in other ways—i.e.:
- From “In Temple Grandin’s book Thinking in Pictures, it” to “In her book Thinking in Pictures, Temple Grandin”
- From “can be very visual in their” to “often have highly visual”
- From “it seems as though” to “it seems”
This has me wondering whether Chat is getting better, when prompted to revise, at simply revising what’s there rather than regenerating. This would be a good development, because sometimes all we want is copy-editing and proofreading; not rewrites that potentially introduce new problems.
So I went my latest book, Students With Autism, and selected a passage that has nagged at me ever since shortly after its publication. That’s because an early reader noticed an embarrassing typo that somehow slipped past me, my editor, the copy editor, all my other early readers, and Grammarly. Today I asked ChatGPT-4 to proofread this passage (this time I said “proofread” rather than “revise”), and lo and behold, GPT-4 caught the typo and fixed it. Here is the passage as it appears in the book:
Besides this, written discourse has several specific syntactic characteristics that require readers to draw on their background knowledge and/or inferencing skills to fill in missing linguistic information. First, compared to spoken discourse, written discourse involves more frequent use of passive sentences, and these often omit the agent responsible for the action (as in “mistakes were made”). Second, written discourse tends to be more compact—connecting more ideas with fewer words. Much of this compactness comes from the use of nominalizations (or complex noun phrases) in place of clauses. Both passive voice and nominalizations make the relationships between concepts less explicit. Compare:
- They regret that the damn flooded the valley.
- They regret that valley was flooded.
- They regret the flooding of the dam.
In example 1, an active clause specifies which entity caused the flooding (the dam); in 2, the passive structure omits this entity. In 3, where a noun phrase (“the flooding of the dam”) replaces a clause, three pieces of information are lost. First, while the dam is mentioned explicitly, the valley goes unmentioned. Second, there is no longer a tensed verb (“flooded”) to indicate that the flooding occurred in the past. Third, the relationship between the dam and the flooding is no longer explicit: Is the dam responsible for the flooding or a victim of it? Out of context, “the flooding of the dam” could denote a current or future event in which the dam is the thing being flooded. In general, the more the discourse relies on noun phrases like “the flooding of the dam,” as opposed to clauses like “the dam flooded the valley,” the more work the reader must do to figure out what’s going on. Sometimes the broader context of the discourse will suffice. But sometimes background knowledge—for example, about the effects of new dams on water flow—is also helpful, or even essential.
Given the near-complete absence in our Brave New World of meticulous human copy editing, this is an excellent development.
Catherine & Katharine 