False memory for orthographically versus semantically similar words in adolescents with dyslexia: a fuzzy-trace theory perspective

Over the years, studies on memory functioning in dyslexia have been conducted in various methodological and theoretical paradigms. The relationship between dyslexia and working memory (especially its phonological component) is the best studied aspect of memory functioning in dyslexia (e.g., Giofrè, Stoppa, Ferioli, Pezzuti, & Cornoldi, 2016; Kramer, Knee, & Delis, 2000; Nicolson, Fawcett, & Baddeley, 1992; Sheng, Byrd, McGregor, Zimmerman, & Bludau, 2015; Smith-Spark & Fisk, 2007). Among other memory theories that have been used in studies on dyslexia, the fuzzy-trace theory (FTT) appears to be a very promising approach to study and explain the symptoms of developmental dyslexia.

Created and actively developed by Brainerd and Reyna (i.e., 1990, 2004, 2005; Reyna, 2012) and their collaborators, the FTT has provided a comprehensive framework for research on well known effects from a novel perspective (e.g., studies on framing effect or false memories: Reyna, 2012). The FTT is based on the assumption that the process of encoding of information creates not one, but two parallel and independent memory traces representing the same stimuli. These two memory traces are the verbatim trace and the gist trace. The first is a symbolic representation of given information/stimuli, a memory trace of the surface form. It means that it encodes information like, for example, the font color, the exact character order in a word, or the exact number order in a telephone number. Verbatim trace is a precise and concrete representation of a stimulus based on its perceptual characteristics. This trace is especially important in tasks that require a high level of precision to succeed, like calculations or dialing a telephone number.

The second kind of memory representation, the gist trace, is a mental representation of information/stimuli meaning—that is, semantic properties. It means that the gist trace encodes, for example, the meaning of the words/sentences or whether a given number was rather a small or high one. Because of this, gist trace is less precise then verbatim trace but allows a person to understand cultural symbols, metaphors, etc. People use it in communication and in order to make the reasoning process more effective and less cognitively effortful. Gist trace is also more durable then verbatim trace, which means that with time, our memory of verbatim information becomes weaker than our memory of the gist trace, which leads to greater reliance on the gist trace. As already mentioned, the two memory traces are encoded in parallel, which means that the encoding of traces occurs at the same time and independently. This independence is the major difference between the FTT and the psycholinguistic understanding of the nature of verbatim and gist traces from which verbatim versus gist distinction comes from (see Reyna, 2012). Psycholinguistic theory assumes that gist develops from verbatim and, in this case, depends on it. However, studies conducted from the FTT perspective have shown that there is no need of encoding one of the traces in order to encode the other. Because of this, gist trace can be encoded without verbatim encoding (e.g., in the case of the priming effect—gist information affects our cognition without reception of any verbatim data) and vice versa, verbatim trace can be encoded with a lack of gist encoding (e.g., in the case of nonsense syllable learning).

In fuzzy-trace theory literature, we can also find assumptions about some processes involved in memory retrieval. Two of them—phantom recollection and recollection rejection—are of special interest in the context of false memory study (Brainerd, Reyna, Wright, & Mojardin, 2003; Brainerd, Wright, Reyna, & Mojardin, 2001; Stahl & Klauer, 2008, 2009). False recognition of stimuli takes place in the following cases: (1) when a retrieval cue evokes the retrieval of a gist trace but does not evoke verbatim trace retrieval, (2) when a subject is not sure about the stimulus status and makes a guess, and (3) in the case of phantom recollection, that is, due to a memory illusion based on a strong gist memory. Phantom recollection may occur when a retrieval cue is a new but very similar stimulus to the stimulus seen before; therefore, this related distracter shares the gist with a target. For example, someone may falsely accept the novel word ocean because it shares some meaning with the target word sea. However, the recollection rejection process may reduce the occurrence of false recognition and, if this is the case, the related distracter leads to target recollection and distracter rejection. For instance, if someone retrieves the verbatim trace of the target word sea she/he will reject ocean as a similar distracter.

The pattern of memory functioning in dyslexia, if we would like to describe it in terms of the FTT on the basis of previous studies, seems to be rather unclear. On the one hand, the results suggest that dyslexia is connected with an impairment of verbatim trace memory. Lots of research has shown that verbal short-term memory (STM) is impaired in people with dyslexia (e.g., Bauer, 1977; Cornwall, 1992; Kramer et al., 2000). Studies that have used procedures like letter-, word-, or digit-span (where participants have to recall a list of items in the order of their presentation) have shown that there is a significant impairment of verbal memory both in children and adults with dyslexia. Memory of the exact verbal form of the stimuli required for good performance in span tests seems to be based on verbatim trace retrieval.

It seems that various aspects of STM are impaired in people with dyslexia. In the study conducted by Smith-Spark and Fisk (2007), participants performed not only verbal but also visuospatial memory tests. There were three simple span tasks for verbal memory and one simple task for visuospatial WM, that is, the Corsi Block Span Task (where participants have to remember the location of the presented squares filled with “X” signs and recall them in the right order after each trial). In the study, there were also three complex span tasks measuring WM capacity: reading span, computation span, and spatial working memory span. Statistically significant differences in the memory performance of participants with dyslexia were observed for all the tasks, however, for the Corsi block span task, the significance was on marginal level (p = 0.053). These results show that the impairment of memory in dyslexia is not only connected with simple verbal tasks.

In a study which used the Hebb learning task, Bogaerts, Szmalec, Hachmann, Page, and Duyck (2015) showed how performance on the serial-order learning task is connected with dyslexia. Participants were asked to repeat—immediately after presentation—the blocks of nine consonant-vowel syllables (that were grouped into three pseudowords). One trial made use of many blocks, but one block was repeated many times throughout the whole trial. This repeated block is called the Hebb sequence. In the experiment, there were two parts of the learning phase with two separate Hebb lists. Researchers assessed how effective the learning of Hebb lists was immediately after the initial learning session and after the second and third sessions—conducted after 24 h and after a 1-month delay. The findings of this experiment showed that serial-order learning in people with dyslexia is impaired in comparison to the control group.

A study conducted by Miles, Thierry, Roberts, and Schiffeldrin (2006) examined directly how verbatim and gist trace recall works in dyslexia. In this study, participants were asked to repeat a set of sentences of varying complexity—from short and easy to complex. Researchers examined how many and what type of errors the participants had made. Their results corroborate the hypothesis that memory of verbatim trace is impaired in students with dyslexia. In comparison with typical learning students, people with dyslexia committed more errors connected with verbatim information but preserved the gist of the sentences.

On the other hand, some studies on memory functioning in dyslexia suggested that people with dyslexia differ in terms of gist trace memory in comparison with typically developing peers. However, the pattern of difference is not clear. There are definitions of dyslexia (e.g., Fletcher et al., 2002) that explicitly assume an impairment of gist memory functioning, which is manifested in the impairment of reading comprehension. The results of an experiment conducted by Brainerd, Forrest, Karibian, and Reyna (2006), in which the performance in the DRM task was compared between learning disabled and nondisabled children, correspond with this assumption. However, the presented studies of other investigators suggest the opposite pattern of connections between developmental dyslexia and gist memory. This research (e.g., Blau, 2013; Miles et al., 2006) showed better gist memory in the group of people with dyslexia in comparison with the control group. For example, in the study of Miles et al. (2006), participants with developmental dyslexia made errors and used some answering strategies connected with gist trace more often than their peers without specific learning disorders—for example, they used a synonymous word in the place of the exact word more often when they failed to recall the form of the sentence they had to repeat.

In another study, Blau (2013) used the DRM (Deese–Roediger–McDermott paradigm) task (Roediger & McDermott, 1995)—a well known praradigm, which allows researchers to evoke and examine false memories of critical words—these words are not presented on the study list during the learning phase of a memory experiment but are semantically connected with all the words from the study lists. This procedure creates a strong false memory of the critical word occurrence during the study (the probability of recall/recognition of the critical word as a word from the study list is nearly as high as the probability of a correct answer for targets). In terms of the FTT, accurate recognition is based on the retrieval of verbatim trace, while false recognition of critical words should be based on gist trace retrieval (cf., Brainerd & Reyna, 2002). The results of Blau’s (2013) study showed that participants with uncompensated language and learning disorders (LLD) rely more on gist trace retrieval then persons without LLD and those with compensated LLD—they assert that the critical word was present on the study list more often than the participants from the two other groups. There was no difference in the memory of targets, which suggests similar reliance on verbatim trace retrieval.

Nevertheless, most studies on memory functioning in dyslexia referring to the FTT (e.g., Blau, 2013; Brainerd et al., 2006; Miles et al., 2006; Voss, 2013) did not use any process-based modeling analyses. However, there are methods available that allow for process-based analysis and are adapted for the FTT—i.e., multinomial processing tree (MPT) models (e.g., Erdfelder et al., 2009). This method brings new methodological possibilities into studies of fuzzy-trace memory functioning in dyslexia. “MPT models address categorical data based on the assumption that the sample frequencies observed for a well-defined set of responses follow a multinomial distribution “(Erdfelder et al., 2009 p. 108; Riefer & Batchelder, 1988; Stahl & Klauer, 2008, 2009). This means that when there is a finite and fixed set of possible reactions on certain stimuli, the probability of every alternative reaction always sums up to 1.

A particular MPT model (adapted to a specific theory) depicts the hypothetical processes assumed in the theoretical framework, defining the set of parameters and equations—branches of the MPT model that present every possible sequence of processes leading to a particular reaction to a specified stimulus. For every kind of stimulus, there is a specific tree (the set of equations taken from the model’s branches) that describes the processes assumed by the applied theory, which can lead to every possible answer. MPT models allow researchers to study latent processes underlying overt reactions; therefore, this analysis is conducted on a more elementary level than the standard methods used in cognitive psychology research on memory in dyslexia.

The MPT model, based on the fuzzy-trace theory, was created by Brainerd, Reyna, & Mojardin, (1999) for the conjoint recognition paradigm. This paradigm refers to the distinction between the identity and similarity of a stimuli. The test list includes three types of stimuli: identical to targets, related (similar in meaning) to targets, and unrelated, novel items. In the first type of testing procedure, participants only have to accept the stimuli that are identical to the targets and reject similar and novel ones; in the second test condition, they only have to accept similar targets and reject all the other items; and in the third condition, participants have to accept both identical and similar stimuli and reject novel items. A potential difficulty of the conjoint recognition paradigm is the fact that this experimental procedure is not economical in terms of time and resources for the researcher wanting to use it in her/his studies. However, Stahl and Klauer (2008, 2009) have prepared a more convenient version of the model—the simplified conjoint recognition paradigm. In contrast to the original method, the experimental procedure includes only one test condition in which the participants receive three answer options: they recognize if the test item is the same as the target, whether it is new but related to the target, or if it is novel (i.e., new and not similar).

In the presented experiment, a simple false memory for the occurrence procedure was used (e.g., Brainerd, Gomes, & Moran, 2014; Brainerd et al., 2003; Nieznański & Tkaczyk, 2017, Stahl & Klauer, 2008) in which participants study a list of unrelated words and, during the recognition test, are presented with targets, unrelated distracters, as well as semantically (e.g., sofa when couch was the target) or orthographically (e.g., sofa when soda was the target) related distracters. This extension of the original procedure (Stahl & Klauer, 2008, 2009) allows the effect of perceptual (typo-graphical) similarity on memory functioning of dyslectic participants to be studied. This extension implies some differences in the assumptions concerning the parameters of the processing trees for the two types of distracters. The general model applied in the study contains one set of seven parameters for the semantic similarity condition and another set of seven parameters for the orthographic similarity condition.

Method

Participants

Seventy-one adolescents 15 to 19 years of age (M = 16.6 years) participated in the study; all of them were high school students. Thirty-three of the participants were students with dyslexia (14 male) and 38 were students without any learning disorders (22 male). The assignment to the first group was based on the formal diagnosis of dyslexia as documented by school files. Each participant from the control group declared that she or he has never suffered from any learning disorders.

Material and procedure

The material used in the experiment consisted of two sets of 48 triads (each including a target word, its synonymous and orthographically similar word, e.g., sofa - couch - soda or sword - blade - sworn) and two sets of 24 unrelated words. During the study, in each condition, 54 target words were presented, among which three words were added to the study list as a primacy buffer, and three words were added as a recency buffer. In the “semantic condition,” the test list consisted of 24 targets, 24 related synonyms, and 24 unrelated distracters. In the “orthographic condition,” the test list was made up of other 24 targets, 24 orthographically similar words, and 24 unrelated distracters. In each condition, two test list versions were prepared so that if a study word appeared as a target in the first version, it was replaced by its related word in the second version and vice versa. Each version was assigned to approximately an equal number of participants.

The participants were tested in small groups, on personal computers, in a computer classroom. The presentation of stimuli materials and response recording were controlled using the E-Prime program 2.0. Written instructions were presented on computer screens and the participants started the task once they became familiar with the instructions. The study words were presented in random order at a rate of 3 s on a 19-in. computer screen. Words were printed in capital letters, Times New Roman font, with a 40-point font size. When the study list was complete, the participants solved arithmetic problems for several minutes as a buffer task. At the test phase, the participants were instructed to recognize if the presented word was the same as a target presented at study, similar to a target word or completely new (unrelated to any target). The participants were given an example of a similar word (semantically or orthographically dependent on the experimental condition). During the presentation of each test item, the response options were presented at the bottom of the screen, reminding the participants which key refers to a particular response.

Experimental conditions (semantic versus orthographic similarity) were manipulated between lists but within participants. That is, each participant took part in both conditions, half of them started from the orthographic condition and the other half from the semantic condition. The item types (target vs. related vs. novel) were manipulated within participants and lists. The dependent variables constituted the response frequencies for every kind of test stimuli and the parameters of the multinomial model measuring verbatim and gist memory processes, and response biases.

Multinomial model

As the measurement model, we used the Stahl and Klauer (2008, 2009) multinomial model for the simplified conjoint recognition paradigm. However, as mentioned earlier, we extended the basic procedure by adding a new type of similar distracters. Apart from related distracters that are semantically similar to the targets (i.e., sharing the gist), we also included distracters that are orthographically similar. In the model, there are three processing trees—each for a different type of stimuli—and seven distinct parameters representing latent memory and decision processes. These processes are defined according to the FTT and represent the probability of retrieving the verbatim trace when a target is presented at test (Vt), as well as the probability of retrieving the verbatim trace when a related distracter is presented at test (i.e., recollection rejection) (Vr). Given no verbatim memory but available gist memory—the probability of retrieving the gist trace is represented by parameter Gt or parameter Gr when a target or related distracter is presented at test, respectively. Phantom recollection (Pr) occurs for related distracters in the absence of recollection rejection and leads to a false “target” response. The two relevant guessing processes are represented by parameters a and b. Given that no verbatim or gist memory is available for the test item, the observer may guess it is old with probability b. Then, parameter a is the probability of guessing that this undetected item is a target. Research conducted by the authors of this model (Stahl & Klauer, 2008, 2009) as well as other researchers (Nieznański & Tkaczyk, 2017) showed evidence for the validity of the model. The full version of the model contains seven parameters, which is too many in relation to 6 degrees of freedom in the data. Therefore, it is not mathematically identifiable and at least one restriction has to be imposed on the free parameters before the model can be applied. Figure 1 presents the simplified conjoint recognition model. On the left are the item types used at test, i.e., the targets, similar distracters, and unrelated distracters. On the right are the types of participants’ responses which are connected with the item types by the branches of the processing tree.

Open image in new window