TL;DR
This paper investigates how changes in language use over time, specifically in how verbs are used with prepositional phrases, affect how people of different ages process sentences. They found that verbs whose usage patterns have shifted over decades are interpreted differently by older adults compared to younger adults, suggesting that our lifelong language experience shapes our real-time comprehension.
What Problem Does This Paper Address?
Language is not static; it evolves. New words emerge, and the way existing words and grammatical structures are used changes over time. This means that different generations experience slightly different versions of a language. The central question this paper tackles is whether these diachronic (across time) changes in language usage are reflected in how people of different ages process language now. Specifically, they focus on how the attachment of "with" prepositional phrases (PPs) to verbs or direct objects changes and how this relates to age-based differences in sentence comprehension. For example, "Slice the cake with a knife" uses "with a knife" as an instrument for the verb "slice," while "Slice the cake with the pink frosting" uses "with the pink frosting" to modify "cake." The preference for one type of attachment over the other, known as "verb bias," can vary depending on the verb.
Methods
The researchers employed a multi-pronged approach:
- Diachronic Corpus Analysis: They analyzed two large historical corpora (Syntgram and CCOHA) to track the frequency of "with"-PP attachments to verbs versus direct objects for 27 specific verbs across several decades (1940s-2000s). This allowed them to quantify how verb biases have changed over time. They used dependency parsing to identify the attachment sites of the "with"-PPs.
- Behavioral Experiment: They conducted an online experiment where participants (aged 25-75) were presented with sentences containing ambiguous "with"-PPs and asked to interpret them by clicking on corresponding images. This measured their real-time sentence processing preferences.
- Bayesian Multilevel Modeling: Statistical models were used to link the diachronic changes observed in the corpora with the age-related differences found in the behavioral experiment. This allowed them to test whether changes in verb bias over time predict differences in interpretation across age groups.
Datasets
- Syntgram Corpus: A corpus of verb-specific syntactic annotations based on Google N-grams.
- CCOHA (Clean Corpus of Historical American English): A large historical corpus of American English.
- Participant Data: Behavioral data collected from 209 participants recruited via Amazon Mechanical Turk, aged 25-75.
Key Insights and Ideas
- Language Change is Reflected in Processing: The study provides strong evidence that changes in language usage over decades are not just abstract linguistic phenomena but have tangible effects on how individuals process language in real-time.
- Lifelong Experience Matters: Older adults' language processing is influenced by the language statistics they encountered earlier in life, not just the current state of the language.
- Verb Bias is Dynamic and Age-Dependent: Verb biases are not fixed. They change over time, and these changes interact with age, leading to different interpretation patterns in younger versus older adults.
- Decreasing Instrument Bias is a Key Trend: A common diachronic trend observed was a decrease in the frequency of instrument-based "with"-PP constructions over time for certain verbs.
Key Results
- Diachronic Verb Bias Shifts: For some verbs, the "instrument" versus "modifier" bias for "with"-PPs changed significantly between the 1940s and 2000s. A common pattern was a decrease in instrument bias.
- Age-Related Interpretation Differences: In the behavioral experiment, older adults showed different interpretation preferences compared to younger adults, particularly for verbs that had undergone diachronic change.
- Predicting Interpretation from Diachrony: The most striking result was that verbs whose instrument bias decreased over time were more likely to elicit instrument interpretations from older adults than from younger adults. This suggests that older adults' processing reflects the stronger instrument bias that was prevalent in earlier decades.
- Model Fit: Bayesian models showed significant effects of verb bias categories and age, with a notable interaction effect for verbs that had decreasing instrument bias. For these verbs, older adults were more likely to favor instrument interpretations.
Limitations and Future Work
- Limited Syntactic Scope: The study focused on only one syntactic alternation ("with"-PP attachment). Future work could explore other syntactic changes (e.g., genitive and dative alternations) and their relationship with lifespan processing differences.
- Corpus Size and Composition: The CCOHA corpus, while valuable, had fewer relevant verbs and instances than Syntgram, potentially affecting the robustness of its diachronic trends. Differences in genre balance between corpora might also play a role.
- Dependency Parsing Limitations: The dependency parsers used might not fully capture the nuanced semantic plausibility of "with"-PP attachments (e.g., the appropriateness of a "sponge" for "feeding"). Future studies could employ more context-sensitive models like BERT.
- Norms Based on Young Adults: The initial categorization of verb biases was based on norms collected from young adults. Future research should consider age-specific norms or collect production data across the lifespan to better understand age-related biases.
Summary updated