Reading is an inevitable part in sight translation, as it’s the only source of input.
This task involves the procedure of extracting information from written words and constructing a mental representation of the whole text or comprehending the meaning conveyed. Within this process that is cumbersome in nature but rather quick in relation to time, a few fundamental subtasks have to be completed before we can determine whether the reading task is done successfully.
First of all, word identification is a core element and a prerequisite of reading that we encounter early on. Information is purely embedded in written symbols, and therefore orthographic decoding is the first and foremost step (de Groot, 2013), among which letter recognition plays the central role. Individual features such as horizontal or oblique lines are preserved in readers’ mental storage, and the number of matching features decides whether a certain letter will be activated or not (Rayner, Pollatsek, Ashby, & Clifton Jr., 2012). Under this efficient framework, our brain can save a huge amount of cognitive resources and quickly identify letters to trigger meaning access in later stages (for further discussion, see Grainger, Rey, & Dufau, 2008).
After letters and letter strings are activated or “fired”, the information goes one step further to activate phonological and even semantic information. The information passes through two routes for meaning extraction. One is through the
Grapheme-to-Phoneme route, in which phonological coding is first activated and then the semantic information is retrieved, while the other is through the lexical route, in which orthographic information directly leads to semantic meaning. The latter is especially needed for words with irregular pronunciation (Coltheart, 2000; and see Jobard, Crivello, & Tzourio-Mazoyer, 2003 for relevant information). Researchers in reading found that these two routes are probably not contradictory, but rather
co-existent, and they compete as morphological information is fed to both routes and whichever processes faster wins the “meaning retrieval game”.
Phonological information is important for word identification and even for comprehension during reading (Frost, 1998). Inner speech can be a persuasive support for this point of view. It’s been found in many electromyogram (EMG) studies (the tool which measures the electrical activity of muscles) that less-skilled readers tend to rely on subvocalization more than skilled readers do, and the phenomenon of
subvocalization becomes more prevalent as text difficulty increases. In addition, comprehension suffers when difficult texts are encountered and subvocalization eliminated (Rayner et al., 2012).
The phenomenon of inner speech does seem to make some sense, since children start learning a language through speaking and listening, while reading comes later as a relatively unnatural activity. In order to facilitate learning how to read, children are inclined to resort to sounding out words (Rayner, Foorman, Perfetti, Pesetksy, &
Seidenberg, 2001, 2002). Aside from this fact, phonological information is preserved
better in short-term memory (Rayner et al., 2012), which is absolutely needed for information processing and meaning integration.
Generally, as the activation of lexical representation passes a certain “threshold”, a word is recognized, and the meaning and syntactic properties become available for further processing (de Groot, 2013). However, the intended meaning of a word does not naturally “emerge” every time. For words with multiple meanings, a selection mechanism has to be in place to ascertain the suitable meaning that is plausible and consistent with the whole context or the previously constructed scenario.
Rayner, Cook, Juhasz, and Frazier (2006) clearly demonstrated through fixation durations the effect of meaning selection and the cognitive effort induced. In their experiment, a word in each sentence was selected as the ambiguous stimulus. In one condition, the dominant meaning of the ambiguous word would be suitable for interpreting the sentence; in the other, the subordinate meaning would be a better fit.
In addition, sentences were classified into two categories: one with a modifier to prime the meaning, and one without. The researchers found that when there was a modifier that prompted the subordinate meaning, fixation time became longer than in the neutral context and in conditions that went with the dominant meaning.
Interestingly, the ambiguous word with the dominant meaning was fixated longer when not preceded by a dominant-meaning prompting modifier. The above findings suggest that a context, even just one modifier preceding the target, can affect the meaning retrieval process, not to mention the influence of the global context, which
clearly sways the amount of time spent on resolving meaning issues (see Duffy, Morris, & Rayner, 1988; Kambe, Rayner, & Duffy, 2001).
In Rayner, Warren, Juhasz, and Liversedge’s (2004) experiment, sentences that were 1) normal, 2) possible but somewhat implausible, and 3) anomalous were employed to examine their effects on eye movements. The results showed that, starting from gaze duration, eye movement indices were obviously affected in the anomalous condition, meaning the comprehension process was seriously disrupted, while in the implausible condition, go-past time was the earliest index to “take a hit”
but did not reach statistical significance. In Rayner et al.’s (2004) experiment, the syntactic structure was fixed, confining the possible thematic roles or features of the ensuing constituent, therefore limiting the discussion to plausibility of the matching between the verb and the noun that followed.
In addition to semantic information, syntactic properties also impact the comprehension process in their own way. Frazier and Rayner (1982) successfully proved the influence of syntactic parsing. In their experiment, two sentences as stated below were shown to participants:
The girl knew the answer by heart. (Direct object)
The girl knew the answer was wrong. (Sentence complement)
Participants were found to have more difficulty reading the second sentence. Eye fixations were longer, and the number of regressions increased, suggesting that readers originally adopted a certain way of assigning roles to each constituent, which
was suitable for the first sentence, but not for the other. When they found the grammatical structure of the following part did not correspond to their assumption, they had to reanalyze the sentence.
Garden-path sentences are another kind of stimulus most favored by reading scholars. McDonald and Carpenter (1981) successfully drew out the influence of idiom interpretation on participants’ comprehension. In their experiment, sight translators in the study returned to former regions (in the form of regressions) when encountering disambiguating information that exhibited inconsistency with their earlier interpretation of the idiom further down the sentences. Though this experiment is more related to semantic processing, when combined as a whole with
aforementioned studies, it shows that sentence processing is incremental in its nature (see Frazier & Rayner, 1982; Rayner et al., 2006).