Obtaining New Data
Due to scheduling conflicts and a lack of data, we have been unable to make significant progress on our mouth movement research for the sign language avatar. Instead, we have shifted focus to data that is more readily avaiable. The same annotated database we were using for mouth movements has several other tiers of annotations. We are now focusing on the amount of time between main glosses. AT first, I thought we could isolate the spaces by subtracting the main gloss tier from the English translation tier. Through ELAN, we can process subtractions on multiple files at once, so it seemed like the perfect route. It initially appeared that the English translations spanned the entire length of the videos, however, upon further inspection, I found gaps between the annotations. Thus, I needed a new method to retrieve the gaps between the main glosses.
ELAN has another tool that allows you to create a new tier from the gaps in another tier. Unfortunately, this is only available to process one file at a time, meaning I’d need to go into all 870 videos and create this new tier for each one. I spent a day trying to find a way to automate the work, and I settled on the Macro Recorder 2.0 which records the activity of the mouse and keyboard, and then activates that same sequence.
There are two main issues that I need to address. The first and last annotations are the dead spaces of the video outside of the actual signing. I was able to conquer the first annotations by including their deletion in my recorded sequence of actions. Unfortunately, I can’t do the same for the last annotation because the videos are all different lengths. For some, I must scroll to the end to find the last annoation, but in many others, the length of the video does not even reach the border of the window. I have yet to decide on a way to remove the last annotations. ELAN has multiple file editing tools that may help me, but I need to investigate further.
Once I am able to obtain all the information about the time in between glosses across all 870 videos, we’ll be able to use this data to inform our avatar’s timing. What is the average time between glosses? How much does the time deviate from sign to sign?
For future research, I propose that emotions place a significant role in the time between signs, but wWhat other factors increase and decrease the time? Are there rules we can define?