Original link: http://vis.pku.edu.cn/blog/dehumor/
While humor is an important communication skill, understanding it can often be challenging – using it successfully requires a combination of engaging content construction and appropriate oral delivery (such as pauses). Previous work emphasizes the textual and audio features of punchlines, but ignores the build-ups for punchlines in longer contexts. In addition, these theories are often too abstract to make sense of each specific piece of humor.
In this study, the authors developed DeHumor, a visual analysis system for analyzing humorous behavior in public speaking.
This work [1] synthesizes techniques for quantifying humor elements in existing works, adds disassembly and supplements to humor theory, and summarizes five core design requirements.
- R1: Simultaneously analyzing the sound and content of humor to reveal their relevance
- R2: Provides a video-level overview showing humorous speech and language styles and their distribution
- R3: Provide a context-level overview, showing the humorous elements in the context and their relationships
- R4: At the sentence level, support highlighting the pairwise relationship between a word and its phonetic features
- R5: Support intuitive interaction, help users shuttle between different levels, and reveal details at different levels.
As shown in the image below, users can use the dashboard (A) to find lectures of interest. Humor Spotlight (B) helps users further focus on humorous clips with specific language and sound styles. Humor Exploration (C) guides users through a multi-level exploration via an augmented temporal matrix (left) for summarizing humor characteristics and a contextual link map (right) for analyzing the context of humor along with its textual content and oral delivery. Users can click on the sentence to display and play the original video clip in the humor focus (B).
Figure 1: The user interface of DeHumor.
The authors use case studies to verify the effectiveness of the system. In one case, user E1 uses DeHumor to explore the “stand-up comedy” dataset and examine how effectively comedians use interesting events in their lives to set up humor. In particular, Researcher E1 was interested in the use of words in jokes and wanted to see how it helped create humor.
First, E1 filtered videos with the keyword “personal experience” in the control panel. She then became interested in the videos for recurring moments of humor, sorted the videos by the total number of laughs, and chose the #1 video, “Obviously you can’t pretend you’re a cop.”
At the video level, E1 wanted to explore the voice and language styles of comedians, and she found that “repetition” and “break” were two techniques that were frequently used in jokes. E1 raised two questions, one is what words did the actors use to create a sense of incongruity and how was it conveyed (corresponding to R1, R4)? The second problem is that when browsing the reinforcement time matrix, E1 found that the dark gray lines representing the jokes have very short intervals between them, which means that the jokes are very close, and she wants to know how the actor builds the humor in the brief context Sensitive (corresponding to R2, R3).
At the context level, E1 finds that a joke at the beginning has rich humorous features, so he clicks on the time matrix to find the corresponding humorous segment. in the code snippet. She noticed that the phrase containing the word “police” was repeated 3 times before the joke. By examining the corresponding sentences, she realized that the actor was present at the crime scene, where he was asked if he was a police officer. To avoid explicitly claiming to be a cop, he invoked a common trope of cops in film and television to mislead people into believing he was a cop.
At the sentence level, E1 refers to the humorous feature annotation of the sentence and finds that the actor raises his voice at the first few words of the joke. The speaker then paused before revealing the substance of the content. “I’ll ask you questions!” Finally, he intensified his anger with a particle word “okay.” E1 clicks on this sentence and jumps back to the original video clip in the humor focus view.
E1 then clicks on the second highlighted fragment and finds that there are multiple repeated phrases in the context summary, by observing the red rectangles (“to the ground”) and purple rectangles (“going to kill you”, “kills ”), which she found to be an actor recounting a conflict between herself and a student. The student said that if he lost, he would be “ready to die”. Later, E1 found that the phrase “ready to die” appeared again. By exploring the context of this phrase, E1 found that the actor said that he was “ready to die” because of the low salary. The joke is that the actor contrasts the low-paying job with the effort it takes to teach students. The actor inserts a pause and raises his voice after the words of pay to emphasize his complaints about his challenging job and low pay.
Figure 1: E1 uses humor context to explore the context of sentences.
After reviewing the original video, E1 confirmed that these humorous feature annotations help reveal both the content and expression of humor. During the exploration, E1 concluded that the comedian set up dialogue scenes to tell his funny personal experiences. He excels at using contextual repetition to connect parts of a story and using words to create incongruity. In addition, he adjusts his voice (for example, using pauses and raising the volume) to express his emotions and enhance his sense of humor.
references:
[1] X. Wang, Y. Ming, T. Wu, H. Zeng, Y. Wang, and H. Qu, ‘DeHumor: Visual Analytics for Decomposing Humor’, IEEE Trans. Visual. Comput. Graphics, pp. 1 –1, 2021, doi: 10.1109/TVCG.2021.3097709.
This article is transferred from: http://vis.pku.edu.cn/blog/dehumor/
This site is only for collection, and the copyright belongs to the original author.