17 Май, 2019


Evidence for Prediction-based Processing in Native and Non-Native Speakers of

Russian and English

The prediction-based approach to processing puts together the linearity of word-for-word presentation of a sentence in real life and its syntactic parsing based on phrase structure rules (Croker 1999, Schneider & Phillips 2000). The study investigates how the top-down structural prediction is built in native and non-native sentence processing. In a self-paced reading task (Linger), native speakers (NS) of English and Russian and second language (L2) adult learners of these languages read sentences and answered comprehension questions. The stimuli manipulated perception and non-perception matrix verbs as well as the type of anaphora (1).

(1) Bill saw / arrested the mother of the woman that was speaking about herself / her in the yard.

This person was speaking about:

(a) the mother              (b) the woman

The possibility to attach the RC [that was talking about herself / her in the yard] high in Russian and low in English (Fodor 2002) yields two corresponding language-specific binding domains. In Russian the nearest c-commanding element to the anaphor is the noun phrase (NP) [NP the mother], in English – [NP the woman]. In the comprehension task, the Russian-like pattern of anaphor resolution is ‘herself = the mother // her = the woman’, if binding principles are observed. In English, the pattern is reversed (Table 1).

A perception verb saw in (1) makes the reader anticipate an eventive complement and triggers a projection that modifies the matrix verb: Bill saw (what) [the mother of the woman’s talking in the yard]. Grammatically, the doer of talking can only be the higher NP [NP the mother] (Grillo & Costa 2014). When a perception verb is placed in the matrix clause of a restrictive RC, its eventive projection takes the upper-hand over the ambiguity of the RC and favors HA across languages (Grillo et al 2015). In top-down structural processing, a perception verb in the matrix clause facilitates RC attachment to the higher noun, which, in its turn, binds the reflexive. As a result, the preferred pattern for anaphor resolution is ‘herself = the mother // her = the woman’ in the sentences with a perception verb in the matrix clause in both languages.

The data were analyzed with software R. The results show that adult NSs of Russian, as well as L2 learners of this language, co-reference the reflexive with the higher NP and the pronoun with the lower one (herself = mother // her = woman). In English a reversed pattern is preferred. The factors that influence the interpretation decision are the language of testing (p ˂ .001) and the type of the matrix verb (p = .05). Crucially, bilingual parsing is not influenced by the L1, but by the language currently in use. The matrix perception verb influences binding resolution and increases the reading time of the embedded verb (p ˂ .001), the area where the eventive projection is inhibited. Across languages, reflexives are processed faster than pronouns (p ˂ .001), and comprehension decisions are made faster in sentences with reflexives than with pronouns (p ˂ .001). This suggests that Pirnciple A is easier for processing in both native and non-native languages than principle B. Additionally, the pronoun may cause an anticipation for a possessive phrase. This anticipation is dismissed when the parser comes across the following preposition, but a temporary garden path effect complicates processing and increases reading times.

In sum, the data show that different exterior interpretation patterns in English and Russian result from the application of similar parsing mechanisms in both monolinguals and bilinguals. The underlying processing mechanism is prediction-based, as parsing decisions of higher processing cycles shape parsing preferences of the lower processing cycles.


Вам понравиться