What makes certain movie quotes memorable?
Simple tricks to write catchy, memorable lines
Why are some movie quotes indefinitely etched in your brain and not others? What makes them more memorable? Are there any intrinsic semantic patterns that makes them memorable? These are the questions that Danescu-Niculescu-Mizil et al. posits in their paper “You had me at hello: How phrasing affects memorability”.
The premise of the paper is simple: they want to know whether, and how, the way information is phrased and structured can affect how much people remember them. This has application ranging from better script writing to advertisement and politics. For example, a memorable slogan goes a long way keeping the public engaged with your election campaign.
The authors posit that there are two properties of memorable quotes: lexical distinctiveness and generality. Memorable quotes use uncommon words in a common way, meaning they use words which are less frequently used but in a common sentence structure. Memorable quotes are also very general, which makes them applicable in a wide range of situations.
Their data set was full scripts of around 1000 movies. To judge whether a quote from the script is memorable or not, they used the ‘Memorable Quote’ section of IMDB, which is sourced from users i.e. crowdsourced. In addition, they used Google search to search for “quote” + “movie title” and see the result count. Higher the count, higher the memorability.
One of the challenges of the experiment is to separate out the factors that influences memorability. The authors needed to make sure that when they calculate memorability, phrasing of the quotes is the only independent variable. They achieved this by pairing each memorable quote with a non-memorable quote of same word length delivered by the same character in a very close proximity. This made sure that the memorable and non-memorable quotes did not have difference between who delivered the quote and in which context it was delivered. They completed a pilot study with human subjects to make sure that the labeling of memorable and non-memorable quotes were statistically correct.
To study the lexical distinctiveness of the quotes, the authors employed a Language Model (LM) study. They created a “common language model”, language that is commonly used by an average rational human using the Brown corpus. Then they created six LMs with the quotes: 1-gram, 2-gram and 3-gram word models and 1-gram, 2-gram and 3-gram parts-of-speech models. Now that we have two LMs, we can compare them. We can say how similar are our quotes to the common language model. The results show that memorable quotes are more different when compared to the common language model than non-memorable quotes, ergo, memorable quotes are more distinct. They employ words that are less frequently used in the common language models, but they follow the syntactic patterns of the model.
To study how general the memorable quotes are, they adjudicated them on the following criterion:
- Fewer 3rd person pronouns — Third person pronouns are harder to apply to newer contexts. It’s easier to apply “I eat ice cream” than “He eats ice cream”, because saying “he” means that another person has to exist. But we can say “I” all the time as it is not dependent on another entity.
- More indefinite articles — They refer to general concepts.
- Fewer past tense and more present tense verbs — Using past tense usually means you are referring to concrete actions/situations that have already taken place. Those situations might not be portable to general scenarios.
The results show that memorable quotes meet the aforementioned criteria significantly better than non-memorable quotes.
How can we apply this to write better? One of the obvious application is in advertisement, to write ‘catchy’ quotes. Another obvious application is in political campaigns. But I think we should use these criterion whenever we want the words to stick to the readers. For example, let this two criterion guide you when you are writing the first and last sentences of a paragraph, so that readers remember the gist of it.
The caveat to this whole thing is that this is from one research paper only. I am not sure if this replicated elsewhere or someone disproved the work. But, intuitively, it makes sense! I would love to get some A/B testing data with this approach.
