Speaker 1: Hello, my name is Oscar Moreno Arias. My research proposal is entitled Automated Closed Captioning and Accessibility in Online College Lectures. This proposal is part of my current work in Ling 4900 Special Problems in Linguistics. A major motivation behind this research project is to focus on issues of accessibility and inclusion in our society. The Americans with Disabilities Act is a civil rights law that prohibits discrimination against individuals with disabilities. We see the realization of this law in everyday life. One example would be wheelchair ramps outside of buildings. Under this law and Section 504 of the Rehabilitation Act, universities are required to ensure that all electronic course materials are made accessible to students with disabilities. As more and more courses have moved fully online at many universities, there is a growing need to ensure that these courses are ADA accessible and inclusive. One example of accessibility in an online course is providing closed captioning on lectures, which in turn supports students with hearing impairments. This is where I have decided to focus my research. My motivation for this project is also partially drawn from an observation I had in a class where I noticed that the automated closed captioning provided by Zoom didn't quite match with what the professor was actually saying. When I surveyed the literature, I actually learned that closed captioning benefits all students in the classroom, not just those with a hearing impairment. Research has shown that these captions on lectures support language development for students that speak English as a second language. Research also shows that captioning increases student engagement in lectures and increases their ability to retain and recall the subject matter. While captioning has many benefits for students, the process of captioning a lecture is very time-consuming for faculty. In fact, one 20-minute lecture can take hours to caption. Thus, faculty need to be able to rely on automated closed captioning software in order to expedite the captioning process. In this research, I am focusing on the use of automated closed captioning in lectures. Automated captioning is produced by Automatic Speech Recognition Technology, or ASR, which is the same technology that powers your Alexa or Siri. Closed captioning, on the other hand, would imply manually created or edited closed captioning, so specifically a professor might manually add or edit the captioning to their lecture, which as I said can be very time-consuming. When I stated earlier that the captioning in my course didn't quite match with what my professor was actually saying, this is because they were relying on automated captioning. Literature supports that automated captioning software isn't perfect, so for this research, I wanted to look at its implications from the viewpoint of college lectures, where they focus on how ASR errors impact the teaching of disciplines, specifically linguistics. My research questions are twofold. Identify challenges with automated captioning as it relates to topics and illustrations in linguistics, and assess which ASR technology provides more accurate automated captioning. For example, does Panopto perform better than YouTube? To accomplish this, I am collecting video lectures from two linguistic classes at UNT Dallas, Ling 2050 and Ling 4030. There are a total of 10 lectures from two different faculty members, and these lectures were pre-recorded, so only the faculty members are speaking in them. I am also collecting the automated closed captioning of these lectures from both Panopto and YouTube, and I am comparing these automated captions line-by-line with the correct closed captioning that the faculty and I have produced. While I plan to review my data from many angles, I am first going to classify the errors I find with this classification system. Homophones, lexical omissions, lexical insertion, and lexical substitutions, which should account for a large portion of the errors that I find. Here are some examples of the types of errors we will likely find in the data. Homophones are words that sound the same, but are spelled differently, such as here and here or by and by. It will likely be challenging for technology to know which word is being used in the context of a lecture. Lexical omission is where a word is being completely left out of the captioning. For example, we will blank teaching, as opposed to we will be teaching. Here, the ASR omitted the word be. Lexical insertion is when a word is added to the captioning. For example, we will go a over, as opposed to we will go over. Here, the ASR inserted the word a. Finally, here are two examples of lexical substitution errors that are actually from the data that I hypothesize will be the most common error I find in this process. The first line indicates what the professor actually said, and the second line indicates what the automated captioning produced. So here we have socialex versus sociolex. The second syllable of the word was substituted with the word looks. In the second example, the phoneme r is substituted with the word are. Both of these lexical substitutions are going to interfere with a student's ability to understand the important linguistic components that the faculty member is illustrating in these lectures. There are actually several applications for this study. An intriguing question that I kept asking myself regarding the errors that I have discussed with you all today, is automatic captioning good enough for a linguistic course? Some aspects of automatic captioning are useful for online lectures, but does that usefulness outweigh the potential confusion that errors can cause in student understanding, especially with the complex nature of linguistic topics and illustrations? With that said, through this research, we can identify and predict social problems regarding captioning and teaching linguistics online. For example, if faculty are referencing words from languages other than English as illustrations in their lectures, can this research help them identify foreign language words that ASR technology can and cannot handle so they can modify their lectures or their captions accordingly? As I pointed out earlier, we also want to discover if one platform was better than the other in terms of providing automated captioning for these lectures. Since faculty are encouraged to edit the audio captions, it is important to know which platform provides a more accurate initial transcription. Automated captioning is not only important for assisting hearing-impaired students, but it turned out to be beneficial for each and every student viewing these online lectures. Not to dismiss that this is a very time-consuming process for faculty. However, we hope this research helps awareness about the benefits and conclusiveness of captioning lectures. This concludes my presentation. Thank you very much for your time.