TL;DR
Subject-verb agreement errors (e.g., "The cost of the improvements have...") are driven by the presence of a plural "local" noun, but are surprisingly immune to semantic features like animacy or the physical distance between the subject and the verb. This suggests that the brain's agreement mechanism operates on abstract syntactic relations in a highly modular fashion, largely isolated from general cognitive constraints like memory load or non-linguistic meaning.
What Problem Does This Paper Address?
In English, we usually get subject-verb agreement right, but we occasionally "derail." The most common error is proximity concord (or attraction), where the verb agrees with a nearby noun rather than the actual grammatical subject.
The authors investigate a fundamental debate in psycholinguistics: Is language production an interactive process where meaning (semantics) and memory (cognition) constantly influence grammar, or is it autonomous, consisting of specialized modules that do one job without looking at the "big picture"? Specifically, they test whether agreement errors are caused by:
- Memory decay: Forgetting the subject because the sentence is too long.
- Semantic interference: Being distracted because a nearby noun is "subject-like" (e.g., it's a person/animate).
- Notional number: Thinking of a singular subject as plural because it refers to multiple things (e.g., "the label on the bottles").
Methods
The researchers conducted three experiments using a sentence completion task. Participants listened to "preambles" (sentence fragments) and had to repeat them and finish the sentence as quickly as possible. This pressure elicits natural speech errors.
- Experiment 1: Varied the length of the material between the subject and the verb and tested "single-token" vs. "multiple-token" subjects (e.g., "the bridge to the islands" vs. "the picture on the postcards").
- Experiment 2: Varied the animacy of the local noun (e.g., "the author of the speeches" vs. "the speech of the authors") to see if a human-referring noun is more likely to "steal" the verb's agreement.
- Experiment 3: Used complex relative clauses (e.g., "The king that the colonies oppose...") where the local noun is actually the subject of the first verb, to see if the brain prioritizes linear position or structural role.
Datasets
The study used custom-designed sets of sentence preambles (32 experimental items and 56 fillers per experiment). The authors also compared their experimental results against naturalistic error corpora—collections of real-world speech slips—donated by researchers like Anne Cutler and Jane Oakhill.
Key Insights and Ideas
- The Plurality Bias: Errors are almost exclusively "one-way." A plural local noun frequently causes a singular subject to take a plural verb ("The key to the cabinets are..."), but a singular local noun rarely causes a plural subject to take a singular verb ("The keys to the cabinet is..."). Plurality seems to be a "marked" feature that is easily misapplied.
- Syntactic Autonomy: The agreement engine is "blind" to animacy. Even though humans are more likely to be subjects than objects, an animate local noun didn't cause more errors than an inanimate one.
- Memory is Not the Culprit: Increasing the length of the sentence fragment made it harder for people to repeat the preamble correctly, but it did not increase the rate of agreement errors. This suggests agreement is calculated before the sentence is "mapped" to its final linear length.
- Subject Assignment vs. Agreement: Animacy does matter for deciding who the subject is (Experiment 3), but once that choice is made, the agreement process itself ignores animacy.
Key Results
- Proximity Effect: In Experiment 1, 90.5% of agreement errors occurred when the head and local nouns mismatched in number.
- Asymmetry: 88% of mismatch errors occurred with a singular head and a plural local noun.
- Structural Priority: In Experiment 3, when the local noun was the structural subject of the verb (despite being the second noun heard), participants got the agreement right 93% of the time. This proves that agreement isn't just about agreeing with the most recent noun; the brain is actively tracking the hierarchy of the sentence.
- Null Results for Length/Semantics: There were no significant effects for the length of the postmodifier or the "token" status (notional number) of the subject.
Limitations and Future Work
The authors acknowledge that their task relies heavily on the verb "to be" (is/are, was/were), which is irregular. They note that it remains to be seen if regular verbs (which only mark the third-person singular with an "-s") follow the exact same error patterns.
Additionally, the study focuses on English, a language with relatively impoverished agreement. Future work could explore whether these findings hold in "agreement-rich" languages (like Bantu languages or highly inflected European languages) where nouns might have many more categories requiring agreement.
Finally, the authors suggest that the "plurality bias" might be a morphophonological issue—perhaps the brain has a mechanism that "checks off" an affix (like the plural -s) and then fails to apply it again to the verb, leading to a mismatch. This remains a hypothesis for further testing.
Summary updated