Skip to main content

Filling the gap: Strategies for the processing of unbounded dependencies cross-linguistically

Final Report Summary - FILLING THE GAP (Filling the gap: Strategies for the processing of unbounded dependencies cross-linguistically)

Understanding language is what makes us human. How can we translate a series of sounds (or gestures) into a meaningful proposition, and what enables us to do this so accurately, quickly, and effortlessly? The project "Filling the gap: Strategies for the processing of unbounded dependencies cross-linguistically" ( aimed to explore how listeners comprehend syntactically complex sentences in real time, and in particular, whether the mechanisms involved in online sentence processing are affected by properties of the grammar of specific languages, or whether they are universally shared by all listeners.
To do this, the project focused on resumptive pronouns (RPs) – pronominal elements that overtly mark the end of long-distance (also termed filler-gap) syntactic dependencies, and which appear in certain languages (e.g. Hebrew), but not in others (e.g. English). For example, sentence (1) in English is acceptable without the final pronoun, but English speakers judge it as unacceptable with the pronoun. In contrast, its Hebrew equivalent in (2) can appear either with or without the pronoun (the underlined noun phrase in the sentences is the filler; the gap position is marked by an underline):
(1) This is the boy that I thought that Dan liked __ / him.
(2) ze ha-yeled še-xašavti še-dan mexabev __ / oto.
The project addressed two main research questions (RQs):
(RQ1) Does the grammatical existence of RPs in a language cause readers/listeners to adopt different processing strategies than those adopted by readers/listeners in languages where these elements are unavailable?
(RQ2) Do these elements facilitate processing of complex sentences in the languages in which they appear, as argued by some authors, or alternatively do they hinder processing, as argued by others?
To investigate these questions, and others related to them, we have run about a couple dozen experiments using various methods: acceptability judgments, sentence completion, self-paced reading (SPR), cross-modal lexical priming, and event-related potentials using electroencephalography (EEG). Based on their results, we were able to make substantial progress in understanding these issues, providing the following answers (As) to the two research questions posed above:
(A1) The existence of grammatical resumption in Hebrew leads comprehenders in this language to adopt unique processing strategies that have not been observed in languages such as English. Specifically, in a series of SPR experiments, we found that Hebrew readers actively predict the resolution of a long-distance dependency in positions which can host RPs in this language, extending beyond the positions in which comprehenders of other languages where shown to predict dependency resolution.
In addition to what this teaches us about resumption, this result has more important general implications. It shows that the grammatical properties of a language determine and shape processing choices and strategies adopted by comprehenders of this language in real time. This is a significant contribution to our currently very limited knowledge of cross-linguistic differences in processing routines.
(A2) In the overwhelming majority of cases, resumptive pronouns have no processing function whatsoever, namely, they do not aid processing. In fact, there is strong evidence that in many cases, they hinder processing, causing interference and slow-down. This has been witnessed in several of our experiments using different linguistic stimuli and different methods.
These results are significant since they contradict a very widespread intuition that RPs are helpful for understanding long or complex dependencies. The results rule out this option, pointing instead to one of two possible alternatives: i. Resumptive pronouns are a purely grammatical phenomenon; or ii. Resumptive pronouns have a role for the producer, rather than for the comprehender, of the language. Further research can decide between these two hypotheses.
In addition to our two major RQs, early results from the project led us to pursue several additional paths, which we did not anticipate. Two main themes were investigated:
1. The work on the processing function of RPs raised general questions with regard to the use of working memory resources during sentence comprehension, and in particular during the comprehension of filler-gap dependencies. Specifically, we asked what information about the filler, if any, is maintained during the processing of these structures, and what are the implications of this maintenance. These questions were addressed in two series of SPR and acceptability judgments experiments. We learned that: 1. Certain features of the filler are actively maintained throughout processing of the dependency; 2. One such feature is the animacy feature of the filler; 3. Maintained information causes interference with similar material, reflected in reading slowdown as well as a decrease in acceptability; 4. Maintained information plays an important role in determining when an attempt will be made to resolve the dependency.
These results provide us with a very fine-grained characterization of the memory processes involved in processing filler-gap dependencies, which was not available thus far.
2. As filler-gap dependency formation entails an active prediction for the resolution of the dependency, we have also started exploring syntactic as well as lexical prediction processes more generally. We studied this topic using SPR, production, and event-related potentials. Our main novel results are: 1. Not all predictions are equal: comprehenders commit to some predictions to a higher degree than to others; 2. Not all comprehenders are equal: some (specifically, those with a large working memory span) tend to commit to predictions more easily than others; 3. Failed predictions lead to active inhibition of the predicted content, to allow integration of the actual input.
During the period of the project, we have disseminated our results and conclusions widely. The results culminated thus far in five papers in leading peer-reviewed journals (Lingua, Language, Journal of Psycholinguistic Research, Journal of Experimental Psychology, and Natural Language and Linguistic Theory), with four more manuscripts currently in preparation. The results were also presented in the top international conferences on sentence processing, as well as in Israeli conferences, and in a number of departmental seminars in Linguistics and Psychology departments in various Israeli universities (Tel Aviv University, The Hebrew University, Bar-Ilan University, Ben Gurion University).
Importantly, since research on sentence processing in Hebrew is scarce and resources are limited, we decided to make the results of all our pretests and norming studies publicly available through the Psycholinguistics and Neurolinguistics Research Lab's website ( These include by now reaction times for dozens of Hebrew nouns and verbs, associate norms for Hebrew nouns, cloze probability norms, and the coded results of a large-scale (200 participants) online sentence production survey probing subcategorization preferences for 200 Hebrew verbs.
Work on the project during the reporting period was importantly accompanied by setting up and consolidating the operation of the Psycholinguistics and Neurolinguistics Research Lab, led by Dr. Meltzer-Asscher. This included renovating a dedicated space and equipping it for running psycholinguistic experiments, both behavioral and electrophysiological. In addition, we built a team of excellent research assistants and graduate students working on the project, and on psycholinguistic research in general, with many of them planning on pursuing this type of research in the future.