Column: Why Should I Learn English When ChatGPT Can Do My Work?
Column: Why Should I Learn English When ChatGPT Can Do My Work?
  • Professor Soo-Ok Kweon (HSS)
  • 승인 2023.06.15 08:50
  • 댓글 0
이 기사를 공유합니다

Professor Soo-Ok Kweon (HSS)
Professor Soo-Ok Kweon (HSS)

  Ever since the impressive debut of ChatGPT, numerous English teachers have expressed concerns that their jobs have become increasingly challenging, as students may simply rely on the AI system to generate text or use AI-powered translators to convert their Korean writing into English. As a result, teachers may find it difficult to manage writing classes effectively and to evaluate students’ writing skills accurately.
  While it is true that advanced AI systems like ChatGPT can generate impressive pieces of writing, it is important to remember that they are not a replacement for human creativity and critical thinking. As an AI language model, ChatGPT is designed to learn patterns and generate text based on those patterns. However, it does not have the same level of understanding of context, nuance, and cultural references that a human writer would have.
  Therefore, it is crucial for English teachers to emphasize the importance of originality and critical thinking in their writing classes. While AI systems can provide a starting point or inspiration for students’ writing, they should not be used as a substitute for genuine intellectual engagement and creative expression.
For that reason, teachers should encourage their students to engage with English on a deeper level, by reading and analyzing literature, watching English-language films, documentaries, and TV shows, and practicing conversation and writing with native speakers. Fortunately, POSTECH offers courses such as Intermediate and Advanced Reading, Intermediate and Advanced Listening and Speaking, and Intermediate and Advanced Writing, which comprehensively address these crucial aspects of language learning.
  One disadvantage of using AI in learning English is that AI systems can sometimes be limited in their ability to provide context-specific feedback. For example, an AI system may be able to identify grammatical errors in a sentence, but it may not be able to fully understand the intended meaning of the sentence or provide nuanced feedback on the style or tone of the writing. In contrast, human teachers can provide more personalized feedback that takes into account the specific needs and goals of each individual student.
  To be more specific, if you ask ChatGPT to create sentences with multiple possible meanings in English, it will provide you with a lengthy list of such sentences, including “I saw the man with the telescope,” which can be interpreted in two different ways: “I saw the man who had the telescope” or “I saw the man through the telescope.” The human mind has the ability to resolve the ambiguity without explicit learning, and ChatGPT appears to perform well up to this level of ambiguity resolution. I then decided to test the AI's abilities using a well-known English sentence in psycholinguistics literature that is used to illustrate the way humans process sentences, known as a garden path sentence. A garden path sentence is a sentence with an ambiguous part that leads the reader to initially assume a certain interpretation of the sentence, until they reach a point where the ambiguity is resolved and this initial interpretation is shown to be wrong.
  Essentially, when you read a garden path sentence, you encounter an ambiguous part that has multiple possible interpretations, one of which is significantly more likely to be true than the others, which causes you to select that interpretation for the information that you have read so far. However, as you continue reading, you suddenly realize that this initial interpretation is not valid, since it would cause the sentence to be ungrammatical. This forces you to re-process the sentence, in order to identify its correct interpretation.
  The sentence “The horse raced past the barn fell” is a well-known example of a garden path sentence, which can be better understood when rephrased as “the horse which was raced past the barn fell.” It is called a garden path sentence because, when a reader starts reading it, they typically assume that "raced" is an active verb. However, upon reaching the word “fell”, they realize that “raced” cannot be active because the sentence would be ungrammatical (since the verb “fell” would not have a subject - a sentence should have only one subject and one verb). Then, the reader needs to reprocess the sentence to identify the correct interpretation, where the verb “raced” is, in fact, passive. After reprocessing, “The horse raced past the barn fell" is not ambiguous, but "I saw the man who had the telescope” still is. In short, people can differentiate between garden path sentences and ambiguous sentences without pre-training.
  As I anticipated, ChatGPT was unable to handle the garden path sentence. When prompted with the sentence "The horse raced past the barn fell," its response was as follows.
  This sentence is syntactically ambiguous, as it can be interpreted in two ways depending on the reader's interpretation of the sentence structure:
  1. The horse that was racing past the barn fell.
  2. While the horse was racing past the barn, the barn fell.
  Sentence 1 corresponds to the way that humans typically understand the sentence, but sentence 2 is not a plausible interpretation that any human would arrive at. It appears that ambiguity resolution, which is believed to be a fundamental characteristic of the human language acquisition system, is an area where AI still needs to improve. Unlike AI, humans possess an innate ability to resolve ambiguity without the need for pre-training in language acquisition.
  In summary, AI systems like ChatGPT have made impressive progress in natural language processing, but they still have limitations in understanding and resolving ambiguous sentences. Human teachers are still essential in language learning, and AI systems should be used in conjunction with traditional language learning methods.