There is no evidence to support plagiarism allegations against computer science courses

staff reporter

The News found no evidence to support claims circulated among students that Professor Arman Cohan copied the Stanford University course curriculum for his Natural Language Processing course.

A post on the anonymous campus chat app Fizz claimed that Cohan had “copied the entire Stanford CS224N curriculum,” including lectures and homework, without specifying credits. The post received 1,600 upvotes as of Monday, and four follow-up posts from other users also received more than 1,000 upvotes.

“No, I did not create a syllabus based on Stanford’s cs224n or any other existing course,” Cohan told the News. “NLP is a fairly well-established course, several textbooks are available, and there is no overlap in course content with other universities, as the subject matter we teach is well-established and we have a common core of I would like to emphasize that it stems from the fact that we have knowledge.”

The News reviewed lecture slides from Stanford University’s CS224N course called Natural Language Processing with Deep Learning, which will be offered in winter 2023, and found no notable similarities between Stanford’s material and Cohan’s material. . All verbal and visual materials included in Kohan’s slides were original.

None of the users who posted the plagiarism allegations on Fizz responded to requests for comment in time for the news to be published.

Professor Christopher Manning, who taught CS224N at Stanford University in the winter of 2023, reviewed Cohan’s slide deck and compared it to the Stanford version when interviewed by Sunday News.

“The content of the lectures does not appear to have been specifically copied from cs224n,” he wrote. “They look quite different.”

Kohan’s course has only released one homework assignment so far. The assignment includes three parts, the first two of which are unlike any material in the Stanford course. The third part asks students to reimplement the word2vec algorithm (a natural language processing technique published in 2013 that involves obtaining vector representations of words), which is part of his second work at Stanford University. It’s similar to the process required for assignments.

But Manning said he “doesn’t care” about any similarities in the assignments.

“There is some overlap in the missions and that may be what they are aware of,” he wrote. “In Part 3, we have someone implementing word2vec, which is what we have students do in Assignment 2.” However, in both courses he asks students to reimplement the same word2vec algorithm from 2013. He explained that neither of them are “original” because they are.

Cohan said implementing word2vec is a precursor to understanding neural networks and is the standard for teaching the subject.

The three readings recommended for Kohan’s class are also listed on the Stanford University syllabus as recommended reference texts. One of his books, “Speech and Language Processing”, was co-authored with a professor at Stanford University. At least two of the three readings are suggested in the syllabus for natural language processing courses offered at Princeton and Berkeley.

Another post claimed that Yale’s course website is a copy of Stanford’s website. Each website includes a home page and sections for course schedules and assignments. However, both Princeton and Berkeley course sites contained similar structures.

Another user claimed to have written a letter to the dean about the matter and encouraged others to do so as well. Cohan said he first learned of the allegations through an investigation by The News and was “shocked and surprised” when he heard about Fizz’s post.

Yirun Zhao GRD ’29, a teaching assistant for Mr. Kohan’s course, said that after two hours of review, he could not identify any significant content overlap between Mr. Allman’s course and the Stanford version.

“There are some common topics, such as a discussion of neural networks, but I think this overlap is to be expected in NLP courses, which often cover a core curriculum that is recognized across the field,” Zhao said. I did.

Zhao noted that Kohan’s slides are more similar to those of former Yale University professor Dragomir Radev, who taught the course from 2017 to 2023. However, he said this “seems reasonable” as it is common for different teachers to use it. Use similar course materials when teaching the same course with different terminology.

Kejian Shi GRD ’24, another teaching assistant for the course, said that Kohan and his teaching assistants “spent days preparing for the first assignment,” and that the course was designed specifically for Yale students. He said it was designed. He also pointed out that natural language processing, like physics and biology, has a common order of introductory topics that are likely to be taught in any course.

Cohan also highlighted some important differences between his course and Stanford’s CS224N course. His courses include topics in areas such as pre-training his data implications, evaluation details, search-based language models, self-alignment and interpretability, none of which are covered in the Stanford edition. not.

While his course also focuses on traditional methods of natural language processing, such as Naive Bayes classification, the Stanford course focuses exclusively on methods involving neural networks.

“Many courses in our curriculum are standard introductory courses and are widely available across the university,” wrote Jeffrey Block, dean of Yale’s School of Engineering and Applied Sciences. “Professor Cohan is a globally recognized expert in the field of natural language processing, and we are confident that he has carefully selected the material for this introductory course that falls within his research expertise.”

Yale University’s Department of Computer Science was founded in 1969.