Recontextualizing Famous Quotes for Brand Slogan Generation

Slogans are concise and memorable catchphrases that play a crucial role in advertising by conveying brand identity and shaping public perception. However, advertising fatigue reduces the effectiveness of repeated slogans, creating a growing demand for novel, creative, and insightful slogan generation. While recent work leverages large language models (LLMs) for this task, existing approaches often produce stylistically redundant outputs that lack a clear brand persona and appear overtly machine-generated. We argue that effective slogans should balance novelty with familiarity and propose a new paradigm that recontextualizes persona-related famous quotes for slogan generation. Well-known quotes naturally align with slogan-length text, employ rich rhetorical devices, and offer depth and insight, making them a powerful resource for creative generation. Technically, we introduce a modular framework that decomposes slogan generation into interpretable subtasks, including quote matching, structural decomposition, vocabulary replacement, and remix generation. Extensive automatic and human evaluations demonstrate marginal improvements in diversity, novelty, emotional impact, and human preference over three state-of-the-art LLM baselines.

💡 Research Summary

The paper addresses the growing need for fresh, memorable advertising slogans in the face of advertising fatigue, where repeated exposure to the same catch‑phrases diminishes their impact. While recent work has leveraged large language models (LLMs) to automate slogan creation, the authors identify two persistent problems: (1) generated slogans often lack a distinct brand persona and become stylistically redundant, and (2) they frequently feel overtly machine‑generated, offering limited originality and emotional resonance.

To overcome these issues, the authors propose a novel paradigm that recontextualizes famous quotes for brand slogan generation. Famous quotes naturally align with the typical length of a slogan, are rich in rhetorical devices (metaphor, contrast, parallelism), and are already familiar to a broad audience. By remixing a persona‑aligned quote with a brand name or product, the method aims to blend novelty (the new brand‑specific content) with familiarity (the well‑known quote), thereby enhancing memorability and emotional impact.

Technically, the approach decomposes the generation task into four interpretable subtasks, forming a modular pipeline:

1. Quote Matching – Given a target brand and a persona (e.g., Pride, Anticipation, Fear, Joy, Trust), the system prompts an LLM to retrieve multiple candidate quotes that match the desired tone and emotional resonance. This step ensures that the source material is already aligned with the brand’s intended personality.

2. Structure Breakdown – Each selected quote is parsed into “fixed” segments (the core syntactic and rhetorical skeleton that must be preserved) and “editable” segments (parts that can be swapped out for brand‑specific content). For example, Shelley’s “If winter comes, can spring be far behind?” is abstracted to the pattern “A comes, and B cannot be far behind.”

3. Vocabulary Replacement – Within the editable segments, the system performs minimal word substitution. Constraints enforce that replacements have the same length and grammatical role as the original words, and that the total number of changes is limited (typically just inserting the brand name). This preserves the original cadence and rhythm, preventing the remix from sounding forced.

4. Remix Generation – The LLM combines the preserved structure with the newly inserted vocabulary to produce a complete slogan. An additional refinement pass checks for grammatical correctness, logical coherence, semantic clarity, and safety, discarding any remix that fails these criteria.

The authors fine‑tune a QwQ‑32B model using LoRA on a mixed corpus of existing slogans, movie dialogues, and famous quotes, then apply the modular pipeline on top of this tuned model.

For evaluation, 40 diverse consumer brands spanning categories such as beauty, appliances, clothing, nutrition, and electronics are paired with five personas, yielding 200 brand‑persona combinations. Three strong baselines are used: GPT‑4o, DeepSeek‑R1‑Distill‑LLaMA‑70B (DS‑L), and DeepSeek‑R1‑Distill‑Qwen‑32B (DS‑Q). Automatic metrics assess diversity, novelty, and emotional impact, while human judges rate preference and emotional punch on a 5‑point scale.

Results show that the quote‑recontextualization approach consistently outperforms the baselines, achieving modest but statistically significant gains (approximately 3–5 % improvement) across all automatic metrics. Human evaluations reveal a clearer advantage: participants prefer the remix slogans and rate them higher on emotional impact, especially in the “novelty” and “insightfulness” dimensions. Qualitative examples illustrate how the method preserves recognizable structures (“A is B, and the other is also B”) while inserting brand‑specific content, producing slogans such as “Work is coming; Coke cannot be far behind.”

The paper also discusses limitations. The current quote database is limited in size and cultural coverage, potentially restricting applicability across languages and regions. Legal considerations around quote copyright and the risk of unintended meaning shifts when remixing culturally loaded statements are acknowledged. Moreover, the observed performance gains are modest (“margin” level), indicating room for improvement through richer quote repositories, more sophisticated structural parsing, and adaptive persona modeling.

In summary, this work introduces a creative, controllable, and interpretable framework for brand slogan generation that leverages the inherent memorability of famous quotes. By modularizing the remix process, the authors achieve better alignment with brand personas and modest improvements over state‑of‑the‑art LLM baselines, while opening avenues for future research on multilingual quote integration, ethical remixing, and interactive human‑in‑the‑loop refinement.