[Paper Review] You had me at hello: How phrasing affects memorability

Cristian Danescu-Niculescu-Mizil, Justin Cheng+2 more|arXiv (Cornell University)|Mar 28, 2012

Language, Metaphor, and CognitionPsychology39 references107 citations

TL;DR

This paper investigates how phrasing influences memorability in language by analyzing movie quotes in a controlled corpus where memorable and non-memorable quotes are paired by speaker and context. It finds that memorable quotes are more lexically distinctive, use common syntactic structures, and are more general and portable, enabling broader reuse across contexts.

ABSTRACT

Understanding the ways in which information achieves widespread public awareness is a research question of significant interest. We consider whether, and how, the way in which the information is phrased --- the choice of words and sentence structure --- can affect this process. To this end, we develop an analysis framework and build a corpus of movie quotes, annotated with memorability information, in which we are able to control for both the speaker and the setting of the quotes. We find that there are significant differences between memorable and non-memorable quotes in several key dimensions, even after controlling for situational and contextual factors. One is lexical distinctiveness: in aggregate, memorable quotes use less common word choices, but at the same time are built upon a scaffolding of common syntactic patterns. Another is that memorable quotes tend to be more general in ways that make them easy to apply in new contexts --- that is, more portable. We also show how the concept of "memorable language" can be extended across domains.

Motivation & Objective

• To isolate the effect of linguistic phrasing on memorability by controlling for speaker, setting, and situational context.
• To identify intrinsic textual features that distinguish memorable from non-memorable quotes beyond contextual factors.
• To validate the hypothesis that memorable language is both distinctive in word choice and generalizable in application.
• To extend findings on linguistic memorability to other domains, such as slogans and marketing phrases.
• To develop a computational framework for measuring and predicting language memorability based on linguistic form alone.

Proposed method

• Constructed a controlled corpus of movie script lines, pairing memorable quotes (from IMDb lists) with non-memorable lines spoken by the same character at similar narrative points.
• Used external web search frequency and IMDb listing status as proxy metrics for public memorability.
• Applied natural language processing to extract linguistic features: lexical distinctiveness (rare word usage), syntactic structure, word complexity (syllables), phonetic features (front vs. back sounds), and coordination frequency.
• Trained binary classifiers to predict memorability using bag-of-words, syntactic, and combined feature sets, with accuracy compared against random baseline.
• Conducted a pilot human recognition study to validate that perceived memorability aligns with computational labels.
• Evaluated generalization by testing the model on a separate set of slogans, demonstrating transferability of findings.

Experimental results

Research questions

• RQ1 To what extent does linguistic phrasing affect memorability independent of speaker, setting, and narrative context?
• RQ2 What specific linguistic features distinguish memorable from non-memorable quotes in a controlled setting?
• RQ3 How do lexical distinctiveness and generality (portability) contribute to a quote’s memorability?
• RQ4 Can the principles of memorable language identified in movie quotes be generalized to other domains, such as slogans?
• RQ5 Do phonetic and phonological features (e.g., front sounds, syllable count) correlate with perceived memorability?

Key findings

• Memorable quotes use significantly less common (more distinctive) words while relying on common syntactic patterns, indicating a balance between novelty and familiarity.
• Memorable quotes are more general and portable, allowing them to be applied across diverse contexts, which enhances their reuse potential.
• After controlling for context, memorable quotes show statistically significant differences in phonetic features: they use more front sounds (e.g., /i/, /y/) and fewer back sounds (e.g., /u/), suggesting sound symbolism may contribute to memorability.
• Memorable quotes have higher word complexity (more syllables per word) and use fewer coordinating conjunctions, supporting the hypothesis of linguistic distinctiveness.
• Classifier performance using combined linguistic features significantly outperformed bag-of-words baseline (p < .01), with accuracy improvements in both movie quote and slogan prediction tasks.
• A human pilot study confirmed that people can reliably distinguish more memorable quotes from less memorable ones, even when unfamiliar with the source film, validating the external validity of the memorability labels.