Why Language Exceptions Remain the Rule
The past tense of “sweep” is “swept,” not “sweeped.” It would be easier if all verbs followed the regular rule of adding “ed,” so language researchers have wondered why exceptions, like “swept,” resist this simplification as language evolves. A new model shows that interactions among speakers can enable irregular past tenses to survive, even as children continually enter the speaking population with a preference for regularity. Established speakers can keep irregular verbs going if they are spoken frequently, according to the model, which agrees with the observation that the most common English verbs have irregular past tenses.
Since language rules “have an obvious cognitive advantage in terms of [information] storage, we can ask why so many rules have exceptions” that persist, says Christine Cuskley of the Institute for Scientific Interchange in Turin, Italy. A commonly studied example is the past tense of English verbs. Most verbs form a regular past tense ending in “ed,” but roughly irregular verbs do not. Recent work by Cuskley and her colleagues looked at the evolution of verbs in historical texts spanning 160 years . They found a fairly sharp distinction between less common verbs, which tended to be regular, and frequently used verbs, which were predominantly irregular. The exceptions were a handful of verbs in the middle of the frequency range, like “sneaked” and “snuck,” that came in both forms.
To gain some insight into these language dynamics, Cuskley and her colleagues have developed a model in which language evolves through interactions between speakers. Their idea is based on the “naming game,” a model that addresses how a group of interacting speakers settles on a name for a new object or idea (for example, “podcasts”) . For regular and irregular verbs, the model defines each speaker to be in a “word state.” Some fraction of the population might start off in the regular R-state, using “sweeped” as the past tense of sweep, while the rest start off in the irregular I-state, using “swept.” However, speakers can change state if they interact with someone in a different state. The researchers proposed several different rules for these interactions and computed analytically how the population evolved over time, depending on the interaction rate, which is set by how frequently speakers use a given word.
Unlike previous interaction-based language models, the model Cuskley and her colleagues used includes a rate for replacing older speakers with new ones as children supplant their parents. Studies have shown that children often apply standard rules to unfamiliar words, creating past tenses such as “buyed” or “slided.” So the team stipulated that replacement speakers start off in the R-state. However, this assumption pushed all speakers toward the R-state, for all verbs.
The team could offset this bias by choosing interaction rules that favored retention of irregular forms, but that led to another problem: Rather than a sharp distinction between high-frequency irregular verbs and low-frequency regular verbs, the model instead generated a broad distribution of verbs with significant populations of both I-people and R-people—contrary to what the team found in their study of texts. To solve this problem, they added a third state, M, representing a person who uses the regular and the irregular forms interchangeably. The team found that if they tuned the interactions so that M-people tended to convert to the fixed state (I or R) they most often interacted with, the model roughly reproduced the observed pattern of past tenses. The relationship of the M state to real speakers remains obscure, however.
The model demonstrates that the persistence of irregular verbs can be explained simply by the influence speakers have over one another’s vocabulary, say the researchers. This influence is strongest for the most frequently spoken words and can be strong enough to overcome the preference of children and other new speakers for regular verbs. The model also generates approximately the correct pattern of usage, where most verbs are either regular or irregular but not both.
“The strength of this paper is that it focuses on the essential ingredients of simplified models for the evolution of language,” says Eduardo Altmann of the Max Planck Institute for the Physics of Complex Systems in Dresden, Germany. Because the new model can be solved analytically, it provides more direct access to the conditions that produce the different language states, he says. In addition to irregular verbs, Altmann imagines the model could also work for instances where two languages (or ideas or beliefs) compete.
This research is published in Physical Review E.
Michael Schirber is a Corresponding Editor for Physics based in Lyon, France.
- C. F. Cuskley, M. Pugliese, C. Castellano, F. Colaiori, V. Loreto, and F. Tria, ”Internal and External Dynamics in Language: Evidence from Verb Regularity in a Historical Corpus of English,” PLoS ONE 9, e102882 (2014)
- A. Baronchelli, M. Felici, V. Loreto, E. Caglioti and L. Steels, ”Sharp Transition Towards Shared Vocabularies in Multi-Agent Systems,” J. Stat. Mech. P06014 (2006)
“An Introduction to Agent-based Modeling and Simulation” (PDF) by Charles Macal and Michael North of Argonne National Lab
List of common irregular verbs from the Capital Community College Foundation