Jonathan Porter and Dr. Mark Tanner
Purpose
The purpose of this project was to identify the degree to which prosodic features in English (primarily pausing, word stress, and sentence-final pitch) influence native speakers (NS) comprehensibility of non-native speech. Identifying the extent to which particular features of speech influence comprehensibility could have major ramifications on the type of pronunciation instruction that is provided to non-native speakers. It can also have impact on the development of classroom and published materials for non-native speakers.
One major reason that ESL learners are willing to commit significant amounts of time and money to learning English is so that they can better communicate in those contexts in which they find themselves. This includes everyday social encounters with native speakers as well as in their employment. Pronunciation is often identified as a critical stumbling block for many learners of English in achieving their goal of fluent and intelligible English.
Background
A key objective of the study was to see if there was a correlation between the frequency of specific types of prosodic error and native English speakers judgments of comprehensibility. We used listening passages taken from Tanner and Landon’s 2009 study that included 75 ESL speakers. Participants in that study recorded themselves reading aloud a short passage that was later analyzed for prosodic errors in pausing, word stress, and sentence-final pitch. From that collection of 75 speakers, a matrix was created in which ESL learners were identified that had a high frequency of prosodic errors to those who had a low frequency. A total of 16 ESL speakers were identified whose error counts fit the context of the matrix. These oral readings were then randomized and placed on CDs so that NSs could perform paired listening tasks aimed at identifying a rank order of which passages were most comprehensible.
ORCA Project
n a 2009 study 40 NSs listened to the CDs and the results were analyzed. For the ORCA project an additional 40 NSs as well as 80 non-native speakers NNSs rated the 16 ESL speakers. Each of the raters was given a CD with 12 paired passages. They listened to each of the pairings and then identified which of the two ESL learners was more comprehensible, and then were asked to briefly describe the reasoning behind their decision. Using a statistical model that has previously been used to predict the outcome of sporting events, we rank ordered the 16 passages from most comprehensible to least comprehensible. Qualitative comments for all participants (NSs and NNSs) were also analyzed to see what features (in the NNSs’ speech) influenced the NS and NNSs’ raters judgments.
Results
A multiple regression analysis was performed separately for both sets of raters that was based off of the rankings that were determined by the NS and NNs raters. The regression found that both sets of raters (NS and NNs) considered word stress and sentence final pitch significant in comprehending NNSs speech. It is interesting that even the raters that are not native speakers were influenced by the same factors that the native speakers were. Pausing errors in speech were not as significant as word stress and sentence final pitch for both sets of raters.
Much of my contribution to the ORCA project was in the gathering and analyzing of the comments that were provided by both sets of raters. After looking at all of the comments that were given by the raters, Dr. Tanner and I classified the comments in order to determine more fully which factors (positive and negative) were influencing the raters the most. After classifying the comments, we then compared the frequency of the comments that were given by NNS raters and NS raters to see if the same factors were influencing both sets of raters. The most frequent negative factor for both sets of raters was the inability to enunciate sounds and words. For NS raters other factors mentioned (in order of frequency given) were slow speech, background noise and improper stress. The NNS raters mentioned factors (in order of frequency given) such as: strong accent, background noise, hesitations or pausing, and too rushed didn’t pause. The qualitative comments helped us understand that prosodic elements of speech (stress, pausing, pitch) are not the only factors that are significant in comprehending NNS speech.
Looking through our classifications, Dr. Tanner and I were a little confused by the comment that background noise was a negative factor in determining comprehensibility. The night before our presentation at the 2010 TESOL Conference we were looking through our results to verify their accuracy and looking to find the answer to why background noise was mentioned often. As we were looking through the CDs trying to find which passage contained background noise on it, we noticed that the randomized pairings were in a different order then we previously had determined. It was frustrating because the regression analysis results could not be presented with confidence the next day, because our results were based off of pairings that were not accurate. However this did not influence the comments that both sets of raters provided so we adapted our presentation to highlight the valuable information that we gained from the comments provided by the raters, as well as highlighting parts of our analysis that could be presented accurately.
Since the TESOL conference, Dr. Tanner and I have been busily going back through the CDs and making sure that the pairings are correct so that we are more accurate in our analysis. After correcting the pairings, we have since run another multiple regression analysis and are currently in the process of complying the results. We sent in an abstract for the Utah Conference on Undergraduate Research (UCUR) that was accepted, so we are looking forward to present our results in February. This has been a great opportunity to help Dr. Tanner with some important research, and we are working towards eventual publication of our results.