A few days ago I stumbled across an old episode of Talking Machines which was on Real Human Actions and Women in Machine Learning. I was particularly interested in the research by Lillian Lee over at Cornell, whose talk was about how memorable language distinguishes itself from non-memorable language. Lillian and her team explored this by creating a data set comprising memorable movie quotes (with reference to IMDb’s memorable quotes) and non-memorable quotes, which were not identified as memorable but had the same speaker and similar point of context within a movie. They found that memorable quotes were distinctive from non-memorable quotes in several ways.
Primarily, they found that the memorable quotes did not significantly differ in their syntactic structure to non-memorable quotes, the difference was in the unusual word choices for that specific point in dialogue. The other key finding was that the memorable quotes were more generalisable, these quotes had the ability to be applicable in a larger span of contexts, as they involved less contextual constraints.
The generality of memorable language
Specifically, memorable quotes had:
- Fewer third person pronouns
- More indefinite articles
- Fewer past tense verbs and more present tense verbs
These features do seem to lend themselves strongly to the characteristic of generality. Fewer third person pronouns would lessen the constraints on character reference, whereas “I” and “you” seem to be more malleable in real-time conversation (there’s no need for a third party). More indefinite articles allow us to refer to the general idea or object stated while more present tense verbs compared to fewer past tense verbs may have to do with language employment as-the-context-arises. Of course, these are just suppositions: at a glance these features appear fitting for generality.
Memorability for marketing applications
One of the key areas in which understanding memorable language seems useful and profitable, would be within marketing applications. This would include creation of slogans for advertising campaigns to even the conscious phrasing of Tweets in order to gain an ideal number of likes and retweets. In the study, slogans were found to be the most general when compared to memorable and non-memorable quotes, to which they argued was an example of memorability in language across domains (and outside the realm of amusing movie quotes).
Memorability for the adult language learner
Aside from marketing applications, the memorability of language has also been looked at within the domain of second language-learning. Notably, research on what aids adult language learners in effectively capturing a new language include the examination of the impact of gesture and visual aids on language retention. What ties in well with the interest of Lillian’s study would be the language-learning aids on a phrasing level. That’s to say, how phrases are constructed in order to aid language retention. One key example that comes to mind is this study by Boers and Lindstromberg, which found that phrases that included alliteration were more memorable to the language learner and thus were easier to learn. Could the characteristics of memorable quotes of Lillian’s study also be translated to a language-learning domain?
Many language-learning software applications implement a classic combination of words, phrases, sound clips and visual references to illustrate the concepts and objects to be learned. In a sense, the language learner undertaking the basic lessons is guided by the situation and set up, whether it be “the boy is eating an apple” or “she likes to wear green dresses”. Some exercise call for an exact translation and a guess using evidence from images, sounds and the native-language equivalent phrases. In some ways, I do think memorability and specifically, this idea of generality, can be implemented in this space.
Although Lillian’s study states there are certain aspects of memorable language that gives it its generality, I think we can tap into the general picture of portability when it comes to language-learning. For phrases to be portable, one must have the freedom to use a given phrase within different contexts. I think this is what is missing in many online beginner language-learning environments. First-time learners are asked to make direct translations and word-naming using standard visual aids. It would be interesting to see how learners would do if given situational tasks where, as an example, they must choose a word, or a phrase in order to co-operate within a given lesson.
Situation: “How are you feeling today?”
Choose your response:
“I’m well, thanks”
“I could do with a drink”
“I’m the king of the world!”
In my example above, notice how I used an idiom and a Titanic quote. Notice also how all options abide by those generality features outlined by Lillian and team. I think a mix of unexpected word choice, a dash of humour while showcasing generality features may keep the first-time learner engaged. Although ideally I’d like to summarise that teaching English learners memorable movie quotes is imperative, I do think Lillian’s study showcases a very interesting mix of NLP and social analysis which explores a very interesting aspect of how humans are affected by their own language use. Memorable language for the domain in language-learning may provide ideas to emphasise the generality/portability of language use in order to foster engagement.
You can read the study here!