Slogans Are Not Forever: Adapting Linguistic Expressions to the News

Lorenzo Gatti, Gözde Özbal, Marco Guerini, Oliviero Stock, Carlo StrapparavaIJCAI 2015

Introduction

In the contemporary digital landscape, the demand for creative, attention-grabbing content is relentless. While human copywriters excel at crafting memorable slogans, they struggle to maintain pace with the 24/7 global news cycle. Traditional contextual advertising often falls short, relying on rigid keyword matching that lacks creative flair. This research introduces a system designed for "automated linguistic creativity," specifically focusing on "micro-blending." By taking a familiar linguistic anchor—such as a famous movie title or advertising slogan—and subtly modifying it to reflect current news, the system generates "parasitic" slogans that are both novel and recognizable, adhering to the Optimal Innovation Theory.

Methodology

The proposed system utilizes a four-stage pipeline to transform static, well-known expressions into contextually relevant news taglines.

The four-stage pipeline for generating news-aware linguistic expressions.
  1. News Retrieval & Keyword Extraction: Daily headlines are fetched from BBC News RSS feeds. The system identifies "newsy" keywords based on low corpus probability and expands these terms using WordNet for synonyms and Freebase for related entities (e.g., mapping "Beyoncé" to "Bee").
  2. Semantic Pairing: To ensure relevance, a word2vec skip-gram model calculates the cosine similarity between slogans and news articles. Modifications are only attempted when a slogan is semantically close to a news topic.
  3. Syntactic Contextualization: Moving beyond simple word replacement, the system employs dependency statistics from the LDC GigaWord corpus. It evaluates the grammatical fit of a new keyword within the original expression's structure, ensuring that substitutions maintain linguistic integrity.
  4. Ranking and Selection: Final outputs are ranked based on a composite score of news similarity and grammatical dependency, with the top-rated blend selected for the user.

Experimental Setup

The system was tested against a diverse library of linguistic expressions, including 287 clichés, 64 top-rated movie titles, 112 famous song titles, and over 2,000 commercial slogans. The LDC GigaWord 5th Edition, containing over 23,000 documents, served as the reference corpus for calculating word probabilities and dependency statistics.

Results and Discussion

Evaluation via crowdsourcing compared the "smart" dependency-based strategy against a baseline of simple Part-of-Speech (PoS) replacement. The results demonstrated significant improvements in both quality and impact.

Examples of original expressions, news headlines, and the resulting system output.

Key findings include:

  • Grammaticality: Dependency statistics (Dep) outperformed PoS replacement in 81% of cases.
  • Meaningfulness: The combined Dependency and Similarity (Dep+Sim) approach was rated superior to the baseline in 81% of instances.
  • Effectiveness: The full system (Dep+Sim) consistently achieved higher Likert-scale ratings for headline effectiveness compared to simpler strategies.

Qualitative analysis revealed that replacing nouns and verbs is particularly effective, as these parts of speech carry the primary semantic weight. For instance, transforming "Smells like teen spirit" into "Inquests like teen spirit" successfully pivots a familiar cultural touchstone to address a specific news event.

Conclusion and Future Work

While the system successfully generates creative taglines, several challenges remain. A primary concern is the "black humor" risk, where a lighthearted slogan might be inappropriately paired with a tragic event. Future iterations will incorporate sentiment analysis to align the "mood" of the expression with the news content. Additionally, improving parser accuracy and handling the complex dependency relations of verbs will allow for more sophisticated sentence restructuring and even greater creative flexibility.

Summary updated

BibTeX
@inproceedings{gatti2015slogans,
  author = {Lorenzo Gatti and Gözde Özbal and Marco Guerini and Oliviero Stock and Carlo Strapparava},
  title = {{Slogans Are Not Forever: Adapting Linguistic Expressions to the News}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}

Loading related content...

Figure