TL;DR
This paper investigates why the optional infinitive marker "to" is sometimes more likely to appear after a preceding "to" (priming) and sometimes less likely (inhibition). They propose and test two hypotheses, finding that the avoidance of self-embedding in grammatical structures is the key factor driving this difference, rather than simple proximity.
What Problem Does This Paper Address?
In English, the "to" before a verb in an infinitive construction is often optional. For example, "All they do is (to) report gloomy things" and "We helped (them) (to) clear the table." Previous research has shown that the presence of an earlier "to" can influence the likelihood of this optional "to" appearing. Specifically, in constructions like the "do-be" construction (e.g., "All they do is..."), a preceding "to" increases the chance of the optional "to" appearing (priming). However, in constructions involving the verb "help" (e.g., "We helped them (to) clear..."), a preceding "to" decreases the chance of the optional "to" appearing (inhibition). This paper seeks to explain this seemingly contradictory behavior: why does "to" prime in one case and inhibit in another?
Methods
The authors conducted a large-scale corpus study using the Corpus of Contemporary American English (COCA). They programmatically extracted sentences containing the verb "help" and analyzed them using a constituency parser to understand their grammatical structure. They then built a mixed-effects binary logistic regression model to predict the presence or absence of the optional "to" before a complement verb. This model controlled for a wide range of linguistic factors, including the form of "help," the presence of a direct object, the animacy of the subject, and various phonological and syntactic properties.
To test their hypotheses, they also analyzed a broader set of verbs that can take infinitival complements. They compared the rate of infinitival complements following an infinitival main verb versus a non-infinitival main verb, looking for evidence of self-embedding avoidance.
Datasets
The primary dataset used was the Corpus of Contemporary American English (COCA), a large collection of spoken and written English text. The authors also drew upon the Carnegie Mellon Pronouncing Dictionary (CMUdict) for phonetic information and WordNet for animacy classifications.
Key Insights and Ideas
- Locality vs. Function: The paper directly contrasts two potential explanations for the priming/inhibition difference:
- Locality Hypothesis: Interference effects are strongest when elements are close together. The "help" construction might have a shorter distance between the preceding "to" and the optional "to" site than the "do-be" construction.
- Function Hypothesis: The grammatical function of the preceding "to" matters. In "do-be" constructions, "do" acts as an elliptical placeholder, and the preceding "to" is part of the expected parallelism in ellipsis. In "help" constructions, when both verbs are infinitival, it creates self-embedding, which is grammatically disfavored.
- Self-Embedding is the Culprit: The authors' findings strongly support the Function Hypothesis. They demonstrate that self-embedding (an infinitive within an infinitive) is a significant factor that speakers avoid, leading to the inhibition of optional "to" in the "help" construction.
- Broader Implications for Verb Complements: The study extends beyond "help" to show that this avoidance of self-embedding is a general phenomenon across many verbs that take infinitival complements, not just an idiosyncrasy of "help."
Key Results
- The regression model confirmed that a preceding infinitival "to" significantly inhibits the use of optional "to" in "help" constructions (β = -2.072, z = -33.4, p < 0.0001). This is the opposite of the priming effect observed in "do-be" constructions.
- The presence of a direct object NP following "help" had the largest contribution to the model's fit, highlighting its crucial role in determining the syntactic interpretation and subject of the complement verb.
- When testing the self-embedding hypothesis across a wider range of verbs, they found that infinitival main verbs disfavored infinitival complements in 82.1% of cases, with an aggregate ratio of 0.292, strongly supporting the idea that self-embedding is avoided.
Limitations and Future Work
- Spoken vs. Written: The study notes that written language generally shows less optional "to" than spoken language, which they attribute to processing demands.
- Exclusion of Certain Tokens: For the spoken data, tokens where "help" was preceded by "want to," "have to," or "going to" were excluded, as these might represent contracted forms ("wanna," "hafta," "gonna") that don't clearly involve an optional "to."
- Future Work: The paper suggests that exploring the effect of overall corpus frequency of complement verbs (as opposed to their frequency in the "help (to)" construction) could be a fruitful area for future research, as it showed a counter-intuitive promoting effect in their analysis. They also mention that while center self-embedding is known to cause processing difficulty, the effect of self-embedding on the edge of a constituent, as seen in infinitival complements, has received less attention.
Summary updated