Introduction
In the study of language evolution, researchers often seek to identify "selection"—the process by which certain linguistic variants are preferred over others due to inherent utility, efficiency, or cognitive ease. However, a significant challenge arises because word frequencies are not solely determined by linguistic fitness. They are deeply intertwined with shifting cultural, social, and political landscapes. A word might surge in popularity not because it is a "better" word, but because the topic it describes has become more prominent in the public consciousness.
This paper, originally presented at the 12th International Conference on the Evolution of Language (EVOLANG), introduces the topical-cultural advection model to decouple these two forces and provide a clearer picture of how language actually evolves.
The Challenge: Selection vs. Advection
When a term like "streaming" or "blog" enters the common lexicon, is its success driven by linguistic superiority over previous terms, or is it simply "riding the wave" of a technological trend? In evolutionary linguistics, "selection" implies a preference for a variant (e.g., a more efficient irregular verb becoming regular). Conversely, "advection"—a term borrowed from physics describing the transport of a substance by the bulk motion of a fluid—refers to a word being carried along by the momentum of a popular topic.
Without a mechanism to control for these topical "weather patterns," researchers risk misidentifying temporary cultural fads as permanent evolutionary changes. This leads to a high rate of false positives in language evolution studies, where cultural drift is mistaken for genuine linguistic innovation.
The Topical-Cultural Advection Model
The authors propose a model that treats words as "passengers" on topical ships. If the ship (the topic) moves, all associated passengers (words) move with it, regardless of their individual fitness. The model operates through three primary steps:
- Topic Definition: For any target word, the model identifies its "topic" by analyzing its most frequent co-occurring words.
- Measuring Advection: The "advection value" is calculated as the weighted mean of the log frequency changes of these associated neighbor words.
- Distinguishing Signals: If a word’s frequency rises in tandem with its topical neighbors, it is likely driven by advection. However, if a word’s frequency increases while its topical neighbors remain stable or decline, this provides a much stronger signal of genuine linguistic selection.
Methodology and Datasets
To validate the model, the researchers utilized two distinct datasets:
- Corpus of Historical American English (COHA): A massive dataset spanning 20 decades, used to evaluate the model's descriptive power on real-world historical shifts.
- Artificial Corpus: To establish a "ground truth," the authors synthesized a 26-period corpus by mixing 'spoken' and 'academic' genres from the Corpus of Contemporary American English (COCA). By gradually shifting the ratio from academic to spoken content, they simulated changes in topic and content without introducing any actual linguistic selection.
Key Findings
The study yielded several critical insights into the dynamics of the English lexicon and the necessity of controlling for context:
1. Explaining Frequency Variance
In the historical COHA dataset, the model demonstrated that topical changes account for up to 31% of the variance in noun frequency changes. This suggests that nearly a third of the movement we see in word usage is attributable to cultural drift rather than linguistic fitness.
2. Filtering False Positives
The artificial corpus experiment highlighted the dangers of "naive" modeling. Without controlling for advection, a naive model flagged 46% of nouns as undergoing selection, even though the researchers knew no selection was occurring. The advection model successfully eliminated 32% of these false positives, providing a much more accurate baseline for evolutionary study.
3. Innovation and Communicative Need
The model also sheds light on linguistic innovation. By analyzing 133 new nouns that entered the English language in the late 20th century, the researchers found that 55% of these innovations appeared during periods when their associated topics were trending upward. This suggests that new words are often "pulled" into existence by an immediate communicative need created by cultural shifts.
Conclusion
The topical-cultural advection model represents a significant advancement in distinguishing between linguistic selection and cultural drift. While the model does not eliminate all false positives—suggesting that cultural drift is a complex phenomenon requiring further study—it provides a necessary baseline for future research. By filtering out "topical noise," linguists can more accurately identify the selective pressures that truly drive the evolution of language and better understand how our vocabulary adapts to a changing world.
Summary updated