Learning to read proceeds in different ways for deaf and hearing readers. The most obvious difference, associating sounds with letters, is just the tip of the iceberg. Deaf readers typically do not have proficiency in the language represented by print prior to learning to read. Signed languages, such as American Sign Language (ASL), are not simply visual analogues of spoken languages, but fully independent languages that arise naturally in deaf communities (Klima & Bellugi, 1978; Padden & Humphries, 1988; Stokoe, 1980). Thus, the visual–auditory mappings argued to be a key route to breaking the orthographic code for hearing readers may be of little help to deaf readers (Coltheart, Rastle, Langdon, & Ziegler, 2001; Gough, Hoover, & Peterson, 1996; Harm & Seidenberg, 2004; Perfetti & Sandak, 2000).Footnote 1 To develop literacy skill, deaf readers must master the vocabulary and grammatical principles of a novel language, one that is distinct from their primary means of communication. They must simultaneously master the orthographic code that maps visual symbols onto meaning. This contrasts with hearing readers, who have already developed a great deal of knowledge of the language, and whose primary task is to discover how the system of visual symbols maps onto this well-developed knowledge base. To capitalize on visual skills among deaf students, teachers have attempted to use visual analogues of spoken English (manually coded English, or MCE) to promote reading skill (Goldin-Meadow & Mayberry, 2001). However, little or no evidence has indicated that such training methods actually promote acquisition of literacy skill in deaf readers (Luckner, Sebald, Cooney, Young, & Goodwin Muir, 2005; Schick & Moeller, 1992).

Deaf readers vary greatly in reading proficiency (for reviews, see Goldin-Meadow & Mayberry, 2001; Kelly, 2003; Mayberry, 2010; Musselman, 2000; Schirmer & McGough, 2005), possibly because different deaf readers apply different strategies to map orthographic forms to meaning, because of differences in instruction methods, differences in first-language experience, or other factors. Although some deaf readers attain high degrees of skill, the average deaf student gains only one third of a grade equivalent each school year, and deaf students on average have a fourth-grade reading level at high school graduation; research indicates that this has been the case since the early twentieth century (Gallaudet Research Institute, 2002; Holt, 1993; Wolk & Allen, 1984). Despite 100 years of research and interventions, no approaches to reading instruction with deaf readers have been developed that have resulted in consistently high reading levels.

To fully comprehend English sentences, readers (including deaf readers) must undertake morphosyntactic processing. Those morphosyntactic processes reveal aspects of lexical meaning, such as tense and aspect for verb interpretation, as well as revealing how words in sentences relate to one another. Morphosyntactic cues will often be supplemented by other kinds of cues (e.g., visual and semantic context, animacy, etc.), but some sentences can be successfully interpreted only after detailed syntactic computations have been undertaken (Altmann & Steedman, 1988; Ferreira, 2003; Nicol & Swinney, 1989; Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy, 1995; Zurif & Swinney, 1995). For example, syntactic cues must be analyzed to establish appropriate thematic relationships in reversible passives, such as Example (1a), whose meaning contrasts with the active-voice Example (1b):

  1. (1a)

    The girl was chased by the boy.

  2. (1b)

    The girl chased the boy.

Adult native speakers of English respond robustly to these morphosyntactic cues during online interpretation (Gilboy, Sopena, Clifton, & Frazier, 1995; MacDonald, Pearlmutter, & Seidenberg, 1994; Traxler, Pickering, & Clifton, 1998; see Pickering & van Gompel, 2006; Traxler, 2011, 2012, for reviews). Syntactic parsing theories differ with respect to the details of syntactic analysis, but nearly all of them subscribe to two general tenets: (1) Recovering sentence meaning involves syntactic computations, and (2) comprehenders eventually construct or recover a single description of the syntax of a sentence.

Only a handful of studies have attempted to determine whether difficulties in reading comprehension among deaf readers result from disruptions in syntactic processing (Kelly, 1996, 2003; Lillo-Martin, Hanson, & Smith, 1992; Miller, 2000, 2010; Robbins & Hatcher, 1981). These studies have produced conflicting results. Both Kelly (2003) and Lillo-Martin et al. (1992) found that deaf ASL–English bilinguals showed patterns of comprehending English relative-clause sentences similar to those of hearing native English speakers, whether they were skilled or poor readers. These authors suggested that reading comprehension difficulties are related to low levels of processing automaticity (Kelly, 2003) or a phonological processing deficit (Lillo-Martin et al., 1992). By contrast, Miller (2010) compared performance on phonological and orthographic awareness and sentence interpretation tasks by Israeli deaf and hearing readers, and concluded that reading deficits among Israeli deaf students are a product of syntactic processing and not phonological decoding. In his study, deaf participants were compared with controls on a phonological awareness task and a sentence comprehension task involving sentences of differing complexity and plausibility. The results suggested that, by college age, reading comprehension outcomes did not correlate with scores on the phonological awareness task. Adult deaf readers were less accurate than hearing readers on the sentence comprehension task, especially for sentences expressing implausible outcomes. Although syntactic complexity varied across the set of sentences, the relationship between syntactic complexity and comprehension outcomes was not reported. Furthermore, no processing-time data were reported.

In the present study, we investigated how deaf readers respond to syntactic structure cues as they read English sentences. Prior studies have often conflated changes in sentence structure with changes in lexical content, thus making it difficult to disentangle the independent contributions of the multiple sources of meaning in sentence understanding (e.g., Miller, 2010). In the present study, we manipulated structural complexity without changing lexical content. This is particularly important in investigating sentence processing in a population that is heterogeneous in language experience and that may exhibit processing differences in multiple domains (phonological, lexical, and syntactic). Our study is unique in measuring the time course of comprehension in order to gain insights to parsing. This study was designed to begin to answer two very basic questions: (1) Does syntactic complexity affect deaf readers in the same way that it affects hearing readers? (2) Do deaf readers respond to helpful semantic cues that supplement syntactic structure cues? Answering these two questions can help us begin to develop an account of the online parsing processes that deaf readers undertake while reading sentences in English.

The influence of syntactic complexity and semantic cues in hearing readers

Psycholinguistic research on parsing processes in hearing readers has relied heavily on chronometric (reaction time) methods. One such line of research has focused on the comparison between sentences that are more complex syntactically versus those that are less complex (in light of representational assumptions that are supported by linguistic analysis; Gordon, Hendrick, & Johnson, 2001; King & Just, 1991; MacWhinney & Pleh, 1988; Mak, Vonk, & Schriefers, 2002; Traxler et al., 2002; Wanner & Maratsos, 1978). Several studies in this line have focused on the contrast of sentences with subject-relative clauses [e.g., Example (2a)] and object-relative clauses [e.g., Example (2b)]:

  1. (2a)

    The lawyer that phoned the banker filed a lawsuit. (subject relative)

  2. (2b)

    The lawyer that the banker phoned filed a lawsuit. (object relative)

In (2a), the subject of the sentence (lawyer) also serves as the subject of the relative clause that (trace) phoned the banker, as in the lawyer phoned the banker.Footnote 2 In (2b), the subject of the sentence serves as the direct object of the verb inside the relative clause that the banker phoned (trace), as in the banker phoned the lawyer.

Generally, sentences with object-relative clauses take longer to read than sentences with subject-relative clauses, with difficulty manifesting during reading of the relative clause and continuing during reading of the main verb.Footnote 3 Processing accounts variously attribute this difficulty to the dual grammatical function of the subject noun in object relatives (Keenan & Comrie, 1977; parallel function hypothesis); changes in the reader’s perspective at various points in the sentence (MacWhinney & Pleh, 1988); the potential for confusion between the different role-players in the sentence (Gordon et al., 2001); the number of discourse referents that intervene between a head and the trace, or other factors that may increase working memory load (Gibson, 1998; Wanner & Maratsos, 1978; but see Traxler et al., 2012; Traxler et al., 2005); or a general tendency to treat the subjects of sentences as subjects of embedded clauses (Traxler et al., 2002).Footnote 4

The difficulty of object-relative clauses can be reduced when helpful semantic cues are available to the reader (Mak et al., 2002; Traxler et al., 2002; Traxler et al., 2005). In particular, object relatives are processed almost as quickly as subject relatives when the subject of the sentence is inanimate and the noun within the relative clause is animate, as in Example (3a):

  1. (3a)

    The pistol that the cowboy dropped remained in the saloon.

When the subject of the sentence is animate and the noun within the relative clause is inanimate, as in Example (3b), the sentence is much more difficult to process:

  1. (3b)

    The cowboy that the pistol injured remained in the saloon.

In subject relatives, such as Examples (3c) and (3d), the positions of animate and inanimate nouns have little or no effect on processing difficulty:

  1. (3c)

    The pistol that injured the cowboy remained in the saloon.

  2. (3d)

    The cowboy that dropped the pistol remained in the saloon.

Reducing the semantic confusability of the critical nouns does not, by itself, eliminate processing difficulty that attaches to object-relative clauses. If it did, (3a) and (3b) would be equally difficult to process. For the same reason, integration across intervening discourse elements does not provide a complete explanation for the object-relative penalty.

Given the multiple factors involved in object- and subject-relative clause interpretations, it remains an open question whether deaf readers respond to sentences with subject- or object-relative clauses in the same way that hearing readers do. One goal of the present study was to determine whether deaf readers experience the object-relative penalty and, if so, whether the penalty decreases when helpful semantic cues are present.

Experiments investigating syntactic complexity have also assessed comprehenders’ responses to passive-voice and active-voice sentences such as those in (1a) and (1b) (Christianson, Williams, Zacks, & Ferreira, 2006; Ferreira, 2003):

  1. (1a)

    The girl was chased by the boy.

  2. (1b)

    The girl chased the boy.

Some linguistic analyses treat passive-voice sentences as being essentially equivalent to adjectives (e.g., The girl was tall; Townsend & Bever, 2001). Other analyses treat them as examples of extraction and movement (Chomsky, 1981, 1995; see also Jackendoff, 2002). If so, passive-voice sentences like (1a) entail more complex representations and potentially more complicated processes than do active-voice sentences like (1b). For example, Jackendoff (p. 47) described a generative account under which passive voice involves an extracted element [girl in (1a)], a trace located at a gap site [between chased and boy in (1a)]. The passive voice is comprehended via a derivational process, involving the application of transformations. These transformations build a mental representation in which the surface element girl ends up in its canonical location, immediately adjacent to and following chased. On this account, Example (1b) does not entail such covert transformations, and should impose a lower processing load.

Comprehension of passive-voice sentences is more susceptible to impairment by brain damage (Grodzinsky, 1986; Zurif, Swinney, Prather, Solomon, & Bushell, 1993). Furthermore, interpretations based on fully specified syntactic structure representations are more likely to be overruled by lexical–semantic features in passives than in actives (Ferreira, 2003). As a result, readers are more likely to assign an incorrect interpretation to The mouse was eaten by the cheese than The cheese ate the mouse.

We do not have much information about how deaf readers interpret passive-voice sentences (Quigley, 1982). Power and Quigley (1973) showed deaf middle school and high school students index cards with a written English sentence. Participants were asked to demonstrate the meanings of sentences, such as “The car was pushed by the tractor,” using toys to represent the subject and object. Even the oldest age group, 17–18 year olds, completed only three or four of the six trials correctly. In agent-deleted passives, such as “The car was pushed,” participants completed only a third of the trials correctly, on average. Performance on a production task was poorer than on the comprehension task. In this case, participants were shown a picture and then asked to fill in a blank from a set of provided words. The 17- to 18-year-olds completed only two or three of six trials correctly. Quigley and colleagues, who systematically investigated a range of syntactic structures, concluded that deaf school-aged children are not sensitive to many syntactic cues, but rely instead on a general assumption that English sentences follow SVO word order (Quigley, 1982; Quigley, Wilbur, Power, Montanelli, & Steinkamp, 1976). This poses a particular problem for passives, since the order of the subject and object are reversed. Power and Quigley reported that students who were most successful in their study appeared to rely on the presence of the word “by” to identify the structure as a passive (p. 9). This word was more consistently included in the production task than were the appropriate auxiliary and past tense form of the main verb, and comprehension performance was below chance on agent-deleted passives, suggesting that the students did not detect the passive when the word “by” was not present as a cue. Note that these investigators did not assess language proficiency in ASL or English, nor did they report the hearing status of the deaf students’ parents. Thus, there is no way to know what proportion of the students had been exposed to a first language in early childhood.

The present study comprised two experiments assessing deaf readers’ responses to two sentence types. In the first experiment, we tested responses to sentences containing subject- and object-relative clauses. In the second, we tested responses to active-voice and passive-voice sentences. If deaf readers make early syntactic commitments similar to those of hearing readers, object relatives should take longer to read than subject relatives. If deaf readers are sensitive to semantic cues to sentence structure, the object-relative penalty would be reduced when the subject of the sentence was inanimate and the noun within the relative clause was animate. If deaf readers interpret passives via a transformation-like process of affiliating the subject of the sentence with a gap located after the main verb, they should take longer to read passives than to read active-voice sentences.

Experiment 1

In Experiment 1, we assessed deaf readers’ responses to sentences like those in Examples (3a)–(3d):

  • Object Relative, Inanimate Subject

  • (3a) The pistol that the cowboy dropped remained in the saloon.

  • Object Relative, Animate Subject

  • (3b) The cowboy that the pistol injured remained in the saloon.

  • Subject Relative, Inanimate Subject

  • (3c) The pistol that injured the cowboy remained in the saloon.

  • Subject Relative, Animate Subject

  • (3d) The cowboy that dropped the pistol remained in the saloon.

If deaf participants respond like hearing, native English readers do, then sentences containing object-relative clauses [(3a) and (3b)] should be read slower than sentences containing subject relatives [(3c) and (3d)]. Furthermore, if deaf readers are sensitive to semantic cues, the object-relative disadvantage should be reduced when the sentence has an inanimate subject [e.g., (3a)].

Method

Participants

Three groups of participants were recruited: deaf readers (deaf), monolingual hearing English speakers (English), and hearing bilingual readers whose first language was not English (bilingual). A total of 68 participants were included in the deaf group, 31 in the English group, and 34 in the bilingual group. The deaf participants were recruited from the Sacramento, San Francisco, and Los Angeles regions through community contacts and community service centers for the deaf. The present experiments were conducted as part of a larger project investigating language processing as a function of age of exposure to ASL. In the deaf group, 22 deaf readers were classified as native signers if they had begun to learn ASL from birth. These readers’ median age was 34 years old (range: 22–45). Another 24 early signers had learned ASL before puberty, but not from birth (median: 35, range 21–65). The 22 late signers had learned ASL after puberty (median age: 47, range 27–69).

The hearing monolingual and bilingual readers were recruited from the UC Davis Psychology Department participant pool. The bilingual participants all had English as their second language and had a wide variety of first languages. Their self-reported English proficiency averaged 3.0 on a scale from 1 (poor) to 7 (native proficiency), with a range of 1–6. They averaged 66 % correct on the Nelson–Denny vocabulary test (range 24 % to 100 % correct). The monolingual English group provided us with a baseline. The bilingual group resembled the deaf group, in that they were reading the experimental sentences in their second language. Hence, the bilingual group would provide an indication whether differences in performance between the deaf and English groups were due to language status (first vs. second language) or hearing status.

Stimuli

The stimuli consisted of 28 quadruplets of sentences similar to those in Examples (3a)–(3d). These stimuli were the same as those used in previous experiments with hearing readers (Traxler et al., 2002; Traxler et al., 2005).

Each participant also read 24 sentences from Experiment 2, as well as 50 filler sentences with a variety of simple sentence structures (active-voice sentences, active intransitives, etc.). One version from each quadruplet was assigned to one of four lists of items using a Latin square design. Each participant saw only one version of each item, and every participant read equal numbers of items from each of the four subtypes. Each participant read a total of 102 sentences: 28 sentences from the present experiment, 24 from Experiment 2, and 50 filler items (Appendix).

Apparatus and procedure

Participants read each test sentence one word at a time. The participants were instructed to read at a normal, comfortable pace in a manner that would enable them to answer comprehension questions.Footnote 5 Sentences were presented with a self-paced moving window procedure using a PC running custom software. Each trial began with a series of dashes on the computer screen in place of the letters in the words. Any punctuation marks appeared in their exact position throughout the trial. The first press of the spacebar replaced the first set of dashes with the first word in the sentence. With subsequent spacebar presses, the next set of dashes were replaced by the next word, and the preceding word was replaced by dashes. Yes-or-no questions followed 45 of the sentences, and participants did not receive feedback on their answers. The computer recorded the time from when a word was first displayed until the next press of the spacebar.

Analyses

We analyzed reading times for the relative-clause region (the three words between the word “that” and the main verb of the sentence) and the main-verb region (the main verb of the sentence) using hierarchical linear modeling (Blozis & Traxler, 2007; Raudenbush & Bryk, 2002; Traxler et al., 2005). Separate models were run for each scoring region. At the first level of the model, reading times were considered as a function of clause type (subject vs. object relative), animacy (the subject of the sentence was inanimate or animate), and the interaction of clause type and animacy. Group (deaf vs. English vs. bilingual) was entered into the second level of the model. At this level, parameter slopes and error terms were allowed to vary freely between individuals.Footnote 6 The Level 1 and Level 2 parameters were estimated simultaneously.

Results and discussion

Comprehension question results

Comprehension questions were asked after a subset of the sentences in the experiment, with most questions being presented after filler sentences. Due to the sparse data, it was not possible to compare comprehension outcomes across the different conditions. A multilevel model testing for between-group accuracy differences (deaf vs. English vs. bilingual) showed that the English group had the highest accuracy, at 89 % [significantly higher than the deaf group, at 72.3 %; t(130) = 6.52, SE = 021, p < .001]. The bilingual group (84 %) also had higher accuracy than the deaf group [t(130) = 3.24, SE = .024, p < .01]. A multilevel model testing for between-group effects that divided the deaf readers into three subgroups (native, early, and late signers) showed that the native signer group had greater accuracy than did the early [81 % (range: 51 %–97 %) vs. 68 % (range: 47 %–90 %); t(128) = 2.66, SE = .037, p < .01] and late [68 % (range: 46 %–96 %); t(128) = 2.52, SE = .040, p = .01] signer groups, and lower accuracy than the English group [t(128) = 2.71, SE = .026, p < .01], but that they did not differ from the bilingual group [t(128) < 1, SE = 0.028, n.s.].

Reading time results

Table 1 shows the mean self-paced reading times and standard errors for the relative-clause and main-verb scoring regions by condition for Experiment 1.

Table 1 Mean self-paced reading times and standard errors (in parentheses) for the relative-clause and main-verb scoring regions by condition and group for Experiment 1
Full size table

Relative-clause region

Multilevel models for the relative-clause region suggested that baseline reading time (in the inanimate, subject-relative condition) was greater in the deaf group than in the English group [significant by participants, with t 1(130) = 2.30, SE = 90.5, p < .05; but not by items, t 2(25) = 0.55, SE = 90.2, n.s.]. The analyses also produced an animacy by sentence type interaction, indicating that sentences with animate subjects and object-relative clauses were harder to process than the other three sentence types [t 1(130) = 1.83, SE = 27.8, p = .06; t 2(25) = 2.13, SE = 50.3, p < .05]. Follow-up analyses conducted separately for the deaf, English, and bilingual groups showed a significant interaction of animacy and clause type in the deaf group [t 1(67) = 2.43, SE = 49.1, p < .05; t 2(27) = 2.36, SE = 82.2, p < .05] but not the bilingual group [t 1(33) = 1.06, n.s.; t 2(33) < 1] or the English group [all ts < 1, n.s.].

Main-verb region

Models of data from the main-verb region also produced an interaction of animacy and sentence type [t 1(130) = 5.26, SE = 17.4, p < .001; t 2(25) = 2.44, SE = 42.5, p < .05]. Follow-up analyses conducted separately for the deaf, English, and bilingual groups showed a significant interaction of animacy and clause type in all three groups [deaf, t 1(67) = 2.66, p = .01; t 2(27) = 3.83, p < .001; English, t 1(30) = 2.90, SE = 29.2, p < .01; t 2(27) = 2.87, SE = 25.7, p < .01; bilingual, t 1(33) = 3.11, SE = 31.8, p < .01; t 2(27) = 1.95, SE = 65.1, p = .06].

Individual-differences analysis for deaf readers

A further set of multilevel models were conducted to assess group differences between native, early, and late signers. To assess these differences, group (native vs. early vs. late signer) was added as a categorical variable at the second level of a two-level model. The same text variables from the preceding analyses were entered at Level 1. Cross-level interactions indicate that Level 1 parameters differ significantly across the three groups.

These multilevel models indicated that baseline reading time (reading time in subject relatives with inanimate subjects) differed across the three groups in both the relative-clause and main-verb regions. Native signers read the relative clauses at a faster rate than early signers [1,306 vs. 1,622 ms, t(65) = 2.06, p < .05], and late signers [1,306 ms vs. 1,748 ms, t(65) = 2.24, p < .05]. Similar results occurred in the main-verb region [native vs. early, 424 vs. 513 ms, t(65) = 1.75, p = .09; native vs. late, 424 vs. 593 ms, t(65) = 3.50, p < .001]. Group (native vs. early vs. late) did not interact with clause type (object vs. subject relative). Similarly, group did not moderate the size of the clause type by animacy interaction. Thus, the results do not indicate that the three groups responded differently to the clause type and animacy manipulations.

The comprehension data indicate that the deaf group comprehended the sentences to a lesser degree than the English or bilingual group. Comprehension in the native signer group, however, was comparable to the bilingual group, and higher than the other two groups of deaf participants. A cross-language transfer hypothesis, according to which English sentences are comprehended by mapping them to ASL equivalents could account for the difference between the deaf reader subgroups. Such an account would appeal to differences in the quality of ASL representations. One possibility is that native signers have more robust and more finely differentiated conceptual representations. Another possibility is that English-to-ASL mapping processes are more robust and reliable in native signers. Alternatively, it is possible that these comprehension differences reflect different degrees of mastery of English syntax. However, such an account would predict differences in sensitivity to English morphosyntactic cues across the native, early, and late signers. If such differences were present, they did not lead to different patterns of reading time results in the three groups.

For the reading time results, multilevel models in which participant group (deaf vs. English vs. bilingual) was included as a predictor did not produce cross-level interactions (the exception: one model indicated an overall reading speed difference between the deaf and English groups). Thus, no strong evidence emerged that patterns of processing time differed between the three groups. All groups showed the commonly found interaction of animacy and sentence type. All three also experienced more difficulty processing object relatives with animate subjects, as compared with the other three sentence types.

The absence of the animacy-by-clause type interaction in the relative-clause region for the hearing readers is a departure from previous findings (e.g., Traxler et al., 2002; Traxler et al., 2005). Given the robust nature of this interaction in English readers and cross-linguistically, one plausible possibility is a Type II error.

The three subgroups of deaf readers responded to the syntactic and semantic manipulations very much like hearing readers do. Previous studies have established that English object relatives are relatively difficult to process. Previous studies have also shown that semantic cues can reduce the object-relative penalty (in English and other western European languages). Specifically, having an inanimate sentence-subject and an animate relative-clause subject greatly reduces processing load. This pattern occurred to about the same degree in all three groups of deaf readers, indicated by the significant clause type by animacy interaction in the multilevel models in combination with the absence of by-group interactions (native vs. early vs. late signers). Overall, the results indicate that deaf readers undertake similar parsing processes to hearing readers to interpret subject and object-relative clauses. Deaf readers respond differently to different relative-clause types, mediated by the specific semantic properties of the nouns in the sentence. This result contradicts prior accounts suggesting that deaf readers are insensitive to English syntax or that they ignore word order as a cue to syntactic structure.

Experiment 2

Experiment 2 assessed deaf and hearing readers’ response to active-voice and passive-voice sentences. We measured participants’ reading times on the main verb and the following noun phrase. These are two areas where influences of syntactic complexity would be expected to emerge. In the test sentences used here, the main verb codes for two argument slots, but the thematic roles assigned to the pre-verbal and post-verbal arguments differ. In the active voice, the preverbal argument also serves as a thematic agent, whereas in the passive voice, that same constituent would serve as thematic patient. The unusual ordering of thematic roles could induce greater load during processing of the main verb. Similarly, the postverbal argument in the passive-voice sentences would constitute a thematic agent occupying a position more normally occupied by a patient, theme, or experiencer. Assigning the postverbal argument an agent thematic role could require additional syntactic computations (relative to those undertaken for the active-voice sentences; see, e.g., Jackendoff, 2002). However, this assumes that prior syntactic cues (the presence of an auxiliary verb, the preposition by) have not sufficiently prepared the participants to cope with unusual ordering of thematic roles by the time they encounter the postverbal noun phrase. If participants predicted a passive after encountering the auxiliary verb was, then we might observe little or no processing difficulty in the passive-voice sentences.

Method

Participants

The participants were the same individuals who participated in Experiment 1.

Stimuli

The test sentences included 24 pairs of active- and passive-voice sentences, such as those in Examples (4a) and (4b):

  1. (4a)

    The farmer tricked the cowboy into selling the horse. (Active voice)

  2. (4b)

    The farmer was tricked by the cowboy into selling the horse. (Passive voice)

Plausibility norming

Because the lexical content of the scoring regions was kept the same between the active- and passive-voice conditions, the meanings of the two sentences differed. Although we intended the test sentences to be fully reversible, it is possible that the active-voice sentences were significantly more plausible than the passive-voice sentences. To see whether this was the case, we had 16 participants rate the plausibility of the active-voice sentences [e.g., (4a)] and active-voice paraphrases of the meanings of the passive-voice sentences (e.g., The cowboy tricked the farmer). One version of each item (either the original active-voice sentence or the paraphrase of the passive counterpart) was assigned to one of two lists of items. The critical items were randomly interspersed with three other types of items: implausible, impossible, and plausible sentences. The implausible sentences were active-voice sentences like The burglar arrested the policeman, which expressed an unlikely event. The impossible sentences were active-voice sentences like The student believes the stairwell, which expressed impossible events. The plausible sentences were included to fill out the lists to a total of 44 items and to provide a further comparison to assess the plausibility of the experimental items. The plausible sentences were active-voice sentences that expressed common activities (The mother fed the baby, The doctor treated the patient, etc.). Participants were instructed to read the sentences and to indicate, on a 1 (plausible) to 7 (impossible) scale, how likely the events expressed by the sentences were. The rating data were submitted to a one-way, repeated measures analysis of variance (condition: active-voice experimental items vs. passive-voice paraphrases vs. implausible vs. impossible vs. plausible). This analysis revealed an overall effect of condition [F(1, 15) = 39.1, MSE = 0.22, p < .0001]. Sentences like (4a) received a mean rating of 2.1 out of 7. Active-voice paraphrases of sentences like (4b) received a mean rating of 2.3. These two conditions did not differ (t < 1, n.s.). These two conditions [(4a) and (4b)] both differed from the implausible (mean rating = 5.7) and impossible (mean rating = 6.6; all ts > 15; all ps < .0001) conditions. The two experimental conditions were rated as being less plausible than the plausible norming sentences (mean rating = 1.3; all ts > 6.4, all ps < .0001). Thus, although the meanings of the experimental sentences [e.g., (4a) and (4b)] were not at ceiling in plausibility, they did not differ from each other.

Apparatus and procedure

The apparatus and procedure were identical to those of Experiment 1.

Analyses

We analyzed data from three scoring regions. The main-verb region consisted of the main verb of the sentence (e.g., tricked). The determiner consisted of the determiner following the main verb. The noun region consisted of the head of the postverbal noun phrase (e.g., cowboy). The data were analyzed using multilevel models. The first level of the model included the sentence type variable (active vs. passive). Group (deaf, English, bilingual) was entered at the second level of the model. Participant- and item-based analyses were conducted separately.

Results and discussion

Table 2 presents the mean reading times by scoring region and sentence type (active vs. passive voice).

Table 2 Mean self-paced reading times and standard errors (in parentheses) for the main-verb, determiner, and noun scoring regions by group (deaf vs. native English vs. bilingual) and sentence type (active vs. passive voice) for Experiment 2
Full size table

Reading time results

Verb

Multilevel models with sentence type (active vs. passive) at Level 1 and group (deaf vs. English vs. bilingual) at Level 2 indicated a main effect of sentence type in the deaf group [passive faster than active, t 1(130) = 3.02, SE = 19.3, p < .01; t 2(23) = 2.87, SE = 11.8, p < .01]. The hearing bilingual group had longer reading times in the passive than in the active condition [significant by participants t 1(130) = 2.14, SE = 14.6, p < .05; but not by items; t 2(23) < 1, n.s.]. In the English group, reading times at the verb were numerically shorter in the passive than the active condition. As a result, the magnitude of the sentence type effect in the English group did not differ from the sentence type effect in the deaf group (both ts < 1, n.s.). However, the numerical advantage in the English group for passive voice at the verb was not statistically significant (both ts < 1, n.s.).

Determiner

At the determiner, reading times for the passive were shorter than those for the active-voice sentences for the deaf group [t 1(130) = 8.37, SE = 5.48, p < .01; t 2(23) = 3.92, SE = 10.7, p < .01]. The hearing bilingual group also had shorter reading times in the passive-voice sentences [t 1(130) = 2.77, SE = 13.0, p < .01; t 2(23) = 5.04, SE = 14.5, p < .01]. The data from the English group did not produce an effect of sentence type (active vs. passive; both ts < 1, n.s.).

Noun

The data from the noun region produced null results for the sentence type effect in all three groups of participants (all ts < 1.4, n.s.). Reading times in the active-voice sentences were longer in the bilingual group than in the deaf group [t 1(130) = 2.22, SE = 38.8, p < .05; t 2(23) = 7.29, SE = 14.1, p < .01].

Individual-differences analysis for deaf readers

The multilevel models indicated that native signers read the noun faster than early or late signers [native vs. early, 413 vs. 494 ms, t(65) = 1.97, p = .05; native vs. late, 413 vs. 502 ms, t(65) = 2.04, p < .05]. We observed a nonsignificant trend in the same direction at the verb [native vs. early, 433 vs. 501 ms, t(65) = 1.68, p < .10; native vs. late, 433 vs. 498 ms, t(65) = 1.68, p < .10].

Figure 1 represents the reading time differences between the active- and passive-voice conditions for the three groups of deaf readers at the main verb. Positive values indicate that the participant read the passive condition faster than the active condition. All three groups of readers had shorter reading times in the passive- than in the active-voice conditions (differences ranging from 34 ms, in the native group, to 11 ms, in the late signer group). Although the biggest numerical effect was in the native signer group, the model failed to detect differences in the magnitudes of the sentence type effect between the three groups.

Fig. 1
figure 1

Reading times in the active-voice condition minus reading times in the passive-voice condition at the main verb in Experiment 2

Full size image

Some results in the deaf and bilingual groups could reflect “spillover” effects. For the deaf group, the verb and determiner regions were read faster when they followed function words. For the bilingual group, only the determiner followed that pattern.

For all of the readers (deaf, bilingual, and English), the multilevel models did not produce cross-level interactions in any of the scoring regions. These results indicate qualitatively similar responses across the three groups of participants to the sentence type manipulation. The fact that none of the groups produced slower reading times in the passive-voice sentences was surprising. Further experiments will be required to determine whether this pattern is reliable.

General discussion

In Experiment 1, we tested deaf bilingual, hearing bilingual, and native English readers’ response to sentences containing object- and subject-relative clauses. In some of the stimuli, the animacy of the critical nouns offered potentially useful cues to sentence structure and interpretation. Deaf readers exhibited both a general object-relative penalty and a reduction in that penalty when animacy cues pointed toward the correct interpretation while reading the relative clause and the main verb. Both groups of hearing readers showed similar effects of sentence type and animacy, but those effects emerged most strongly at the main verb. These results resemble results for hearing readers from previous studies (Traxler et al., 2002; Traxler et al., 2005). The absence of clear results in the relative clause for the hearing readers may reflect a more “risky” reading strategy on their part, or an ordinary Type II error.

In Experiment 2, we tested deaf bilingual, hearing bilingual, and native English readers’ response to sentences containing active-voice and passive-voice sentences. Tests for simple effects indicated that deaf readers, but not hearing readers, processed the main verb faster in passive-voice than in active-voice sentences. However, because the multilevel models did not produce cross-level interactions based on group membership (deaf vs. hearing bilingual vs. native English), we found no strong evidence supporting qualitative differences in processing between the three groups. Deaf and hearing bilingual readers exhibited evidence of lexically based spillover effects at the determiner, as reading times were shorter in the passive than in the active condition (i.e., determiner reading times were shorter after the word by than after a semantically weighty main verb). Relatively short reading times in the deaf group at the main verb could be similarly interpreted as reflecting a spillover effect.

These results provide an initial glimpse into real-time sentence interpretation in deaf readers. These are the first studies that we know of that have used self-paced reading methods to assess sentence processing in deaf readers while controlling first-language experience (cf. Kelly, 1995, 2003). Comprehension in all three groups of signers was lower than in native-English-speaking college-aged readers. However, comprehension was indistinguishable between deaf and hearing bilinguals who had both acquired a first language from birth, and then acquired English as a second language. Furthermore, whereas comprehension performance was lower in deaf readers as a group than in the other two groups, their performance was well above chance. These results demonstrate that deaf readers comprehend English sentences while undertaking the self-paced reading task. Hence, it appears that this technique is a feasible chronometric method for testing hypotheses relating to deaf readers’ online interpretive processing. Future studies should probe readers’ comprehension more systematically, as our experience suggests that adding additional comprehension questions will not unduly tax deaf readers’ endurance. Moreover, these results underscore the importance of considering deaf readers’ first-language experience in evaluating reading ability in the second language. Prior studies have generally disregarded first-language proficiency when evaluating reading outcomes.

Future studies should also explore the extent to which knowledge of ASL grammar is activated and used by deaf readers during the interpretation of English sentences. Previous studies of hearing bilingual readers have shown cross-language transfer from first-language grammars during second-language tasks (e.g., Hartsuiker, Pickering, & Veltkamp, 2004). Furthermore, evidence indicates that deaf signers activate ASL signs when making semantic decisions about pairs of English print words (Morford, Wilkinson, Villwock, Piñar, & Kroll, 2011). English sentence processing may be affected by transfer from ASL grammar, especially for highly proficient native signers.

Finding such cross-language transfer effects is not guaranteed, however. What makes the situation with signers unique is that morphosyntactic features in ASL are signaled by visuospatial cues. Although some of the underlying syntactic relationships between words in ASL and words in English could be captured in a similar fashion (by phrase-structure or dependency diagrams), some aspects of ASL grammar (e.g., spatial verb agreement) do not have analogues in English. A detailed program of research will be required in order to determine the extent to which deaf readers activate and use ASL syntactic representations while interpreting sentences in English. Our working hypothesis is that ASL syntactic representations are activated.

Continuing to keep in mind that deaf readers are operating in a second-language context, Experiment 1 provides the first demonstration that we are aware of that the object-relative penalty can be reduced by animacy cues in bilinguals operating in their second language (but see Jackson & Roberts, 2010). We know very little about relative-clause structures and processing in ASL, except that these are among the most difficult structures to process for native signers of ASL (Boudreault & Mayberry, 2006). However, no evidence to date has indicated whether or not animacy influences relative-clause interpretation strategies in ASL. Thus, it is not yet clear whether deaf second-language learners of English transfer first-language strategies for the processing of relative clauses to second-language contexts, or whether they discover the utility of this cue solely from exposure to the second language. This state of affairs points toward the strong need for syntactic parsing studies of ASL itself. It also points toward the need for studies of English reading experience as a potential influence on the parsing of ASL.

The clause type by animacy interaction from Experiment 1 has been obtained in monolingual English (Traxler et al., 2002; Traxler et al., 2005), Dutch (Mak et al., 2002), Spanish (Betancourt, Carreiras, & Sturt, 2009), and French (Baudiffier, Caplan, Gaonac’h, & Chesnet, 2011) speakers, all of whom speak Western European languages that have similar typological roots. ASL is typologically distinct from all of these languages. Thus, the interaction of clause type and animacy may reflect something fundamental about constituent embedding and the canonical ordering of thematic roles. This conclusion is based on consistent patterns in the way that hearing readers respond across these several languages. Both subject- and object-relative clauses involve an embedding within a larger structure, which could lead to increased processing load relative to nonembedded expressions, because the contents and interpretation of a main clause must be held in abeyance, whereas the contents and interpretation of an embedded clause are worked out. The relative ease of processing object-relative clauses when the sentence has an inanimate subject and the relative clause has an animate subject indicates either that embedding in and of itself is not problematic or that the animacy cues somehow allow a comprehender to bypass potentially costly syntactic computations.

Conclusions

These experiments provide an initial snapshot of deaf readers’ online parsing and sentence comprehension performance. Deaf readers’ comprehension, on the whole, was somewhat lower than hearing readers’, although native signers performed as well as the other group of bilingual readers. Within the deaf readers, early and late signers read somewhat more slowly than native signers, and comprehended the sentences at a lower degree of accuracy. Deaf readers responded in much the same way as hearing readers to semantic cues in interpreting subject- and object-relative clauses. The differences in comprehension performance that we found across native, early, and late deaf signers illustrate the potential value of an individual-differences approach to building a theory of reading performance in deaf ASL–English bilinguals.