ETD

• anaphora resolution
• developmental dyslexia
• L2 acquisition
• lexical retrieval
• rapid automatised naming

The present study investigates whether Rapid Automatised Naming (RAN) training can influence anaphora resolution during online second-language reading in university students with developmental dyslexia. The processing of anaphoric relations under different linguistic structures is examined in dyslexic and typical readers, contrasting pronominal and lexical anaphora and manipulating the grammatical category of the item that recalls the antecedent (nouns, adjectives, verbs). A further aim is to investigate whether RAN training affects online processing and comprehension accuracy, thereby offering evidence on whether training effects extend from word- to discourse-level processes.
Developmental dyslexia has been widely described as a condition characterised by persistent difficulties in reading accuracy and fluency, often accompanied by reduced processing automaticity and limitations in working memory and lexical retrieval efficiency. There is growing interest in understanding whether the difficulties observed in dyslexia at lower levels of processing can extend to higher-level discourse processes, and, crucially, under what conditions such effects emerge.
Anaphora resolution represents an especially informative mean of investigation into this issue, although the literature on this topic in dyslexic populations remains limited. Resolving anaphora requires the reader to retrieve a suitable antecedent from the discourse model, evaluate potential competitors, and integrate syntactic, semantic, and pragmatic cues to establish reference. This process is cognitively demanding because it involves keeping multiple representations active, and rapidly updating them as new information becomes available. This may be a particularly challenging process for individuals with dyslexia, due to their working-memory impairments.
The relevance of investigating anaphora resolution becomes even clearer when the focus shifts from first-language reading to L2 reading. Lexical access may be slower in L2 reading because of weaker entrenchment of lexical representations. In particular, automatised mapping between orthographic and phonological representations may be weaker, especially in Italian learners of English, given the orthographic discrepancy between the two systems. If dyslexic readers already operate with reduced processing efficiency, an L2 context may be particularly vulnerable, especially under time pressure.
In this context, the choice to focus on pronominal versus lexical anaphora is theoretically motivated. Pronominal anaphora requires the reader to rely on contextual integration rather than on lexical access. Pronouns are interpreted through discourse accessibility, syntactic constraints, and pragmatic expectations. In contrast, lexical anaphora is more semantically informative and therefore better supports reference resolution. This distinction is important because it allows the study to test whether dyslexic readers show a heightened sensitivity to referential underspecification, and whether this sensitivity is magnified in an L2 context where processing is generally slower and lexical representations may be weaker.
Moreover, the thesis explores the effects of the role of the grammatical category of the item that recalls the antecedent. Antecedents recalled by nouns often map directly onto stable referents, whereas verbs and adjectives may require integration of event structure, predication, and contextual inference. These differences may translate into different working-memory and retrieval demands, making them especially relevant for populations characterised by reduced processing automaticity.
The second major pillar of the thesis concerns Rapid Automatised Naming (RAN). RAN tasks involve sub-processes that are shared with reading ability: serial visual scanning, lexical retrieval, rapid identification of familiar stimuli, and articulation. RAN abilities have long been employed as diagnostic measures for reading disabilities. As the Double-Deficit Hypothesis claims, naming-speed deficits have a prominent role in reading difficulties, as they are independent of phonological awareness.
Within this framework, naming-speed limitations can have substantial consequences for reading fluency and for the efficiency with which lexical items are accessed during reading. This makes RAN particularly relevant for understanding dyslexia in transparent orthographies, where decoding accuracy may be relatively less impaired, but fluency remains a core difficulty.
RAN is also employed beyond diagnosis as a rehabilitative tool. RAN-training may enhance automatisation in visual-verbal mapping, thus reducing cognitive costs associated with these processes. As a result, cognitive resources may be freed and available to be used in higher-order comprehension processes. This is precisely where the current thesis contributes to the literature. While RAN training has been widely used for intervention at the word and sentence levels, much less is known about its effects on discourse-level processes. Testing anaphora resolution therefore provides a stringent and theoretically meaningful way to evaluate whether RAN training can influence comprehension mechanisms that go beyond isolated word recognition.
Self-paced reading represents a particularly appropriate methodology to investigate online processing, as it detects precisely where in the sentence processing slows down. This is relevant to our research, as anaphora resolution often presents processing costs localised in critical regions, despite maintaining comprehension accurate. This is particularly relevant in dyslexia research, where readers may compensate for weaknesses in automaticity by increasing effort and time, thus achieving accuracy at the expense of speed. By combining self-paced reading times with comprehension accuracy, the study can distinguish between successful resolution achieved efficiently and successful resolution achieved through increased processing costs.
The main aim of the present study is to investigate how anaphora is processed across different linguistic structures in readers with and without dyslexia, and how this processing is modulated by RAN-based lexical reinforcement training. The thesis is structured around four research questions and related hypotheses. The first research question investigates whether dyslexic students trained with RAN exhibit reading patterns comparable to typically developing peers when resolving lexical and pronominal anaphora following RAN training, with the expectation that dyslexic readers will show overall slower reading times and lower comprehension accuracy than controls, reflecting reduced processing efficiency. At the same time, it is predicted that exposure to RAN training will result in a reduction in the gap between groups, though not necessarily in a complete convergence, consistent with the idea that training effects may be beneficial but constrained by persistent naming-speed limitations.
The second research question concerns whether anaphora type differentially affects reading fluency and comprehension accuracy across groups. The hypothesis is that dyslexic readers will show a larger processing difference between pronominal and lexical anaphora than typical readers.
The third research question investigates whether the grammatical category of the item that recalls the antecedent (noun, verb, adjective) affects anaphora resolution. The expectation is that noun-based anaphora will be processed more easily than adjective- and verb-based anaphora due to differences in the stability of referential representations and in the integration demands they impose.
Finally, the fourth research question explores whether RAN training facilitates the resolution of items associated with one antecedent type more than others. Given the lexical-access focus of the intervention, it is hypothesised that noun-based anaphora may benefit most, although the study also considers the possibility of broader effects mediated by reduced cognitive load and increased efficiency.
The results show that RAN-based lexical reinforcement can support L2 discourse-level processing in dyslexic readers. In fact, despite the persistent disadvantage in both speed and accuracy for dyslexic compared to control readers, notable improvements were observed following RAN training, especially as engagement with training increased. Such improvements were not limited to trained lexical items, thus confirming the role of RAN as a support to broader processing sub-mechanisms. The imbalance between speed and accuracy performance in dyslexic readers demonstrates that RAN training may first consolidate lexical access, whereas improvements in speed may require longer exposure.
Pronominal anaphora consistently emerged as harder to process compared to lexical anaphora, with a potential effect of cross-linguistic influence, that may have increased the cognitive load during processing. RAN appeared to offer more support to pronominal anaphora accuracy in the dyslexic group, whereas lexical anaphora benefitted more from training in the control group. A processing hierarchy emerged for what concerns the part of speech of the item recalling the antecedent, with nouns being processed fastest, followed by adjectives and, finally, verbs.
RAN training reduced sensitivity to part-of-speech processing differences in controls, whereas these differences remained more stable in dyslexic readers, where working-memory-related constraints may hinder processing neutralisation.
Correlational analyses provided results concerning the subskills most strongly associated with anaphoric efficiency, namely semantic fluency and visual search; by contrast, text reading was only weakly correlated with anaphoric efficiency.
These results carry important contributions for both the theoretical literature and in practical, pedagogical terms.