Skip to main content
22 events
when toggle format what by license comment
Mar 20, 2023 at 8:18 comment added Shoe Here is Geoff Pullum's explanation of the CGEL's "because = preposition" stance. I agree with you that most native speakers would not say "...because him" except when trying to be funny. languagelog.ldc.upenn.edu/nll/?p=9494
Mar 20, 2023 at 8:15 comment added Shoe @Mari-Lou A. Yes, ChatGPT3.5 makes some surprising mathematical mistakes. One of the videos I watched about its maths capabilities had an example of ChatGPT3.5 agreeing that 10+9=20, because the user said: "My wife says 10+9=20 and my wife is always right." Maybe that was ChatGPT displaying a sense of humour.
Mar 19, 2023 at 22:27 comment added Mari-Lou A I knew it. The machine doesn't know how to count. It happened to me also. The essay I wrote in response to prompt 5 is 298 words long it is instead 387 words long. I would not trust it to do complex mathematical problems.
Mar 19, 2023 at 22:17 comment added Mari-Lou A G.23 R2 “According to CGEL, "because" can be analyzed as a preposition when it is followed by a noun phrase or a pronoun, as in the sentence "I did it because him", where "because" is followed by the pronoun "him"” … I know the part in bold can be said in very informal English, due to ellipsis, but wouldn't it be considered ungrammatical by many speakers? I would say "because of him”.
Mar 19, 2023 at 22:13 comment added Mari-Lou A I think repeating and repairing the same phrases poses a challenge for our computer friend. It confuses it because it doesn't understand what is syntax, what is grammar. Unless the human identifies the mistake, it will continue merrily on its way. There is never a moment of self-reflection, when ChatGPT says, without being first challenged or corrected, "Wait a moment, can we go back to when I said ZXY? It should have been XYZ.” BTW I spotted a few classic EL&U questions!
Mar 19, 2023 at 20:57 comment added Shoe @Mari-Lou A. Thanks for pointing out this error (alternatively: pointing this error out). I already amended the page to take account of the other issues you alerted me to above. And I will wield the red pen on this latest ChatGPT gaffe tomorrow.
Mar 19, 2023 at 20:33 comment added Mari-Lou A Whip out the red pen for G4, Response 8.2 the Chat bot gives a muddling answer when it says “For example, "take off" is inseparable [not if it means to undress], so it cannot be separated by an object ("take off your shoes" is correct, but not "take your shoes off") [Yes, it is!], while "look up" is separable, so it can be separated by an object ("look up the definition")." But where's the separation?
Mar 18, 2023 at 13:35 comment added John Lawler They built a language generator. That's all it does. The meaning of the language is effectively random, the same as it is in real life when you listen to random people. But, also just like real life with real people, talking good sounds good and one tends to believe. This is not a novel effect.
Mar 18, 2023 at 7:24 comment added Shoe @Xanne. Thanks for the feedback. My interactions of ChatGPT3.5 have left me very impressed with the quality of the texts it produces but a lot less impressed with the quality of its responses to questions about grammar or to requests for lists of sentences exemplifying a particular grammatical feature. However, I read recently that Duolingo are partnering with ChatGPT4 to provide feedback on grammar to their users. I assume from this that Duolingo is convinced that OpenAi have rectified some of the problems that ChatGPT3.5 has with grammar analysis.
Mar 18, 2023 at 5:20 comment added Xanne @Shoe Absolutely fascinating. Thanks for doing this and sharing the results.
Mar 17, 2023 at 19:11 comment added Mari-Lou A No, no I'm dipping back in now, just in a more leisurely fashion. You did a very good job of analysing the bot's strengths and weaknesses. In particular, how it can help a learner increase their vocabulary knowledge. The questions are interesting to begin with.
Mar 17, 2023 at 19:08 comment added Shoe @Mari-Lou A. Thanks a lot for all the feedback. Bedtime reading - as in, I need something to send me to sleep?
Mar 17, 2023 at 17:36 comment added Mari-Lou A I feel one of the bot's weaknesses are in V15, and V16. Identifying typical vocabulary errors by German ESL speakers. I might ask the bot to do the same for vocabulary errors by Italian speakers. I'd be able to say if the list is accurate or not. V18 is interesting, are there different pronunciations of bank depending on its meaning? You missed an error in G3.6, ChatGPT got their=they're mixed up. Very interesting analysis–I didn't finish. It will make for good bedtime reading :)
Mar 17, 2023 at 16:53 comment added Shoe Mari-Lou A. Its human interlocutor was me, Well-spotted that ChatGPT does occasonially punctuate idiosyncratically.
Mar 17, 2023 at 16:53 comment added Mari-Lou A V12. Give me a list of twenty English-German cognates. Is where the bot got terribly "confused" but on the 3rd attempt fixed it. Would a non-native speaker have been aware? Possibly, e.g. foto = photo, Komfort = comfort.
Mar 17, 2023 at 16:49 comment added Mari-Lou A You said it was wrong because of the preceding sentence In American English, "pissed off" is typically used to mean "very angry," whereas in British English, it can also mean "drunk Funnily, the bot didn't use closing speech marks.
Mar 17, 2023 at 16:49 comment added Mari-Lou A Interesting. So far I spotted a few typos by its human interlocutor: “Perform the same task but this this the correct answer is the word that fills the gap in the question sentence.” // “I mean an errors such as "She is a well child" and "He is an afraid boy". But when the bot says In fact, in British English, "pissed" is often used as a slang term for being intoxicated it is correct.
Mar 17, 2023 at 15:57 comment added Shoe @Mari-Lou A, As Araucaria says, ChatGPT3.5 reconstitutes the chunks of text it accesses from its data store into coherent and grammatically perfect English responses. But its responses to various questions and tasks in respect of English grammar are much less good. If you are interested, you can read my analyses of over 80 interactions with ChatGPT here: internationalschooltutors.de/English/advice/learners/info/…
Mar 17, 2023 at 13:45 comment added Araucaria - Him @Mari-LouA It only has data about how to put language together in chunks. It doesn't have data about other facts in the world. It just spews out what it has read.
Mar 17, 2023 at 13:33 comment added Mari-Lou A Out of curiosity, didn't Wikipedia have an equivalent English page? Why would ChatGPT mistake "mammals" and "pets" as verbs? Ahh…because it analyses a word that ends in -s to denote the 3rd person singular? On the other hand, to pet is a verb whereas "to mammal" does not exist in any dictionary that I looked up. Isn't that rather interesting that such a simple mistake was made, and, repeated? What does that tell us about Chat robots? That it has its data and will not deviate from it?
Mar 17, 2023 at 12:56 history edited Shoe CC BY-SA 4.0
deleted 1 character in body
Mar 17, 2023 at 12:42 history answered Shoe CC BY-SA 4.0