SynFlow: A Multidimensional Diachronic Semantic Analysis Toolkit
The paper introduces SynFlow, an open-source toolkit that unifies diverse linguistic dimensions—such as syntax, morphology, and constructional patterns—into a single workflow to enable interpretable, multidimensional diachronic semantic analysis, as demonstrated through a case study of the German adjective *viral* and evaluation on SemEval-2020 Task 1.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Language is not a static museum exhibit; it is a living river that constantly reshapes its banks. When a word changes its meaning over time, linguists have long tried to understand not just that the change happened, but exactly how it happened. Did the word start being used in new sentences? Did it begin to pair with different kinds of words? For decades, researchers have relied on mathematical models that treat words as points in a vast, invisible space. These models are powerful at spotting that a shift has occurred, but they often act like a black box: they can tell you a word has moved, but they cannot easily explain which specific habits of usage caused that movement. To truly understand the evolution of language, scientists need to look at the concrete details of how words behave in the real world, examining the specific patterns of grammar and the company words keep.
A team of researchers at KU Leuven and other institutions has built a new tool called SynFlow to solve this problem. Think of it as a specialized microscope for the history of language. Instead of relying on abstract mathematical distances, SynFlow breaks down a word's life into several distinct, observable dimensions. It looks at how a word connects to other words in a sentence, the specific grammatical roles it plays, the shapes of the phrases it appears in, and even the morphological features attached to it. The system takes a collection of texts from different time periods, counts how often a specific word appears in each of these different contexts, and then compares the results. It does not just say "the word changed"; it calculates exactly which specific connections grew stronger, which ones faded away, and which new combinations of words began to appear.
To see this tool in action, the researchers applied it to the German adjective viral. They tracked the word across thirty years of news texts, from 1995 to 2025. In the early years, the word was almost exclusively used in medical contexts, describing infections that spread like a virus. As time passed, the meaning shifted to describe anything that became wildly popular on the internet. SynFlow did not just detect this shift; it mapped the journey step by step. The system showed that the change was not a sudden explosion but a gradual restructuring of the word's habits. It revealed that the word began to appear more frequently in specific grammatical slots where it modified verbs, rather than just nouns. It showed that the company the word kept changed, moving away from medical terms like "infection" and toward words related to marketing, videos, and social media platforms.
The power of SynFlow lies in its ability to connect these different layers of evidence into a single, coherent story. The researchers found that the shift in meaning was reflected simultaneously in the word's grammatical behavior, the specific phrases it formed, and the types of words it was paired with. For instance, the tool highlighted how the word began to appear in a specific construction with the verb "to go," creating the phrase "to go viral." This was not a random occurrence; the system showed that this specific combination of words grew steadily in frequency, replacing older patterns. By breaking the change down into these individual, observable parts, the researchers could see that the new meaning was built on a foundation of new grammatical habits, rather than just a vague shift in usage.
The researchers also tested SynFlow against existing methods used to detect word changes. They found that their approach was just as effective at identifying words that had changed meaning, but with a crucial advantage: the results were transparent. While other systems might flag a word as changed without explaining why, SynFlow provided a clear list of the specific linguistic evidence that drove the conclusion. The tool is designed to be open and accessible, allowing other scientists to apply the same rigorous workflow to any word in any language. By turning the complex history of a word into a set of clear, countable patterns, SynFlow offers a new way to watch language evolve, turning abstract theories of change into concrete, observable facts.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.