Processing past tense in the face of conflicting cues

Thèse de Master , 2022 , 66 Pages , Note: 1,3

Philologie Anglaise - Linguistique

The present thesis investigates the past tense debate in the face of different cues. For this purpose, a lexical decision task has been designed upon a masked priming paradigm. The current research centers around the question of how simple past primes facilitate the visual recognition of a regular past tense target. The study will investigate how morphological related and form related primes influence the response times in a masked prime lexical decision task. The general purpose of this study is to create another account in the debate by using different parameters that have not been considered in previous research to a great extent. This will be achieved by using different prime-target pairs that correspond to several factors that affect visual word recognition of past tense forms.

Table of Contents

List of Figures and Tables

1. Introduction
1.3. Organization of the text

2.1 Key terms lexica
2.1.1. Mental lexicon
2.1.2 Visual word recognition: interactive models
2.2 Review of related research and literature
2.3 Research hypothesis

3. Methods
3.1. Priming
3.1.1. Masked priming
3.1.2. Lexical decision task
3.2. Methodological and procedural specifications
3.3. Participants
3.4. Experimental design
3.4.1. Procedure and apparatus

4. Discussion
4.1. Findings in the context of the masked priming paradigm
4.2. Single-route or dual-route mechanism?
4.3. Limitation of the present study

5. Conclusion

6. References

Appendix

List of Figures and Tables

Figure 1 The interactive-activation model of visual word recognition (McClelland & Rumelhart, 1981, 1982)

Figure 2 The bottom-up and top-down models (McClelland & Rumelhart, 1981,1982, p. 378, p.380)

Figure 3 DRC model (Coltheart et al., 2001)

Figure 4 PDP framework: triangle diagram from Seidenberg and McClelland (1989)

Figure 5 PDP model by Plaut et al. (1996)

Figure 6 Illustration of the word-and-rules theory (Pinker & Ullman, 2002)

Figure 7 Illustration of trail structure in the masked condition

Figure 8 Correctness of responses to word target

Figure 9 Mean RT in ms among primes

Table 1 Sample of prime-target pairs

Table 2 Frequency of occurrence: mean, standard deviation, and range

Table 3 Means of word length among prime-target pairs

Table 4 Correctness of responses, words, and distractors

Table 5 Mean, SD, and Count of RT

Table 6 One-way ANOVA of summary data (RT)

Table 7 Post-hoc tests

Table 8 Mean lexical decision times (in ms) and standard error and priming effects

1. Introduction

A central aspect of language is its processing and organization of the information in the language users’ minds. Human language consists of many unique features. As the primary communication system, language is a system for linking signals with meanings. The system is an underlying set of principles and rules that have been examined over a long period of time. Different linguistic areas of language have brought insights into the language system and the structural organization of words and sentences.

Especially the processing and the mental representation of words have been in the focus of psycholinguistic, morphological, and psychological research. Studies in the psycholinguistic field have stressed the importance of rules unfolding aspects of language. For example, one of the most discussed issues in cognitive science is the English past tense. This is due to the rule-like regular past tense forms and the somewhat idiosyncratic irregular past tense forms. Over decades now, researchers have been investigating how exactly the past tense is processed and stored in the mental lexicon. The debate started when Rumelhart and McClelland (1986) challenged the assumption that people can use the mental rules of grammar to form the regular past tense. Rumelhart and McClelland (1986) proposed an associative linked computational language processing model that works with a single route. Following up, the other debate camp argues for an abstract rule forming the regular past tense and whole-word representations for irregular verbs.

This marked the beginning of extensive research of the mental representation and processing of the past tense. Researchers used different techniques and methods to get to the core of this debate, but still, the body of literature concerning this issue is inconsistent. Previous studies have used priming paradigms to investigate the cognitive mechanism involved in processing the English past tense. These priming experiments can be embedded in visual word recognition, speech production, or neurolinguistic methods. The focus of the current thesis will be on the visual word recognition drawn upon priming paradigms. In the course of priming experiments, Forster and Davis (1984) developed a new technique to eliminate factors that may lead to misleading results. They proposed a masked priming paradigm to avoid strategic responses in lexical access.

The present thesis investigates the past tense debate in the face of different cues. For this purpose, a lexical decision task has been designed upon a masked priming paradigm. The current research centers around the question of how simple past primes facilitate the visual recognition of a regular past tense target. The study will investigate how morphological related and form related primes influence the response times in a masked prime lexical decision task. The general purpose of this study is to create another account in the debate by using different parameters that have not been considered in previous research to a great extent. This will be achieved by using different prime-target pairs that correspond to several factors that affect visual word recognition of past tense forms.

1.3. Organization of the text

After the introduction, Chapter 2 discusses the key term lexica on which the theoretical basis of this thesis is built. From models of the mental lexicon to inflectional morphology in terms of the English past tense, Chapter 2 provides a solid basis for the experiment of this thesis. At the end of Chapter 2, the research questions and hypothesis are displayed. Chapter 3 then outlines the experimental framework, including the research methods and experimental stimuli. The experiment results are discussed in Chapter 4. Chapter 5 presents an overview of the findings and their implications for the theoretical background and current theories.

2. Theory

In this section, key terms are discussed. These terms underlie the experimental part of this thesis. Theories about the mental lexicon and its structure, as well as theories about the interactive models of visual word recognition, are discussed in more detail. These theories are crucial to understanding the results of the lexical decision task, and they provide a basis for later discussion.

2.1 Key terms lexica

2.1.1. Mental lexicon

English inflectional morphology has been a critical topic for many researchers in the field of linguistics as well as the fields of psycholinguistics and neurology. The latter established experiments to discover dissociations in processing the past tense following brain damage (Tyler et al., 2002). In addition to the neuropsychological approach, psycholinguists have put effort into exploring the nature of the past tense’s representation in the mental lexicon.

Investigating the mental lexicon is a continuing concern within the linguistic field. It is a major area of interest within morphology but also beyond it. Psycholinguists have long studied the mental lexicon and the significant aspects that encircle it. After several attempts, they eventually reached a sufficient and consistent theory about how the mental lexicon works and how it is constructed. Insights into how language users store, retrieve, and organize words are examined based on the mental lexicon.

In early accounts, it was argued that the mental lexicon is structured similar to a dictionary. In one of his early publications, Aronoff (1976) asserted that the dictionary’s list of entries represents a language user’s systemic knowledge of words. In particular, the word-formation rules are the only rules that generate forms and that are therefore listed in the lexicon (Bauer, 2003). Furthermore, Aronoff based his considerations on a work by Morris Halle. In his prolegomena of the theory of word-formation, Halle states that if, for instance, a word has a syntactic exception and a semantic or syntactic anomaly, its idiosyncrasies are then included in the dictionary. Additionally, these features should be treated as distinct from actual words (Halle, 1973). An essential lesson that can be taken from Halle’s work is that there will always be a substantial number of words in a language that must be placed into a lexicon due to their abnormalities. This lexicon takes care of a significant part of this assignment because it seeks to list the class of possible words in a language. However, the list of words at a speaker’s disposal at any given time is not complete. It is worth noting that the abnormalities displayed by words in dictionaries are related to their persistence or the fact that they are mentioned. It seems reasonable to assume that such irregularities are not characteristics of new words coined by a speaker, as these words have not been used long enough to become fixated on an idiosyncrasy. Additionally, Halle considers the lexicon to be preceded by “a filter on the output of word-formation rules and makes sure that only existing words are entered in the lexicon and, thus, used in sentences” (Bauer, 2003, p. 23). In Dominiek Sandra’s (1994) analysis of the morphology of a reader’s mental lexicon, he proposes that the mental lexicon is based on Forster’s and Fodor’s proposals (1983, 1975, and 1985), which he describes as a modular conception of the language processor. This notion holds that particular types of information are stored and processed as distinctively designed components. Elaborating on Forster’s proposal, Sandra states that the lexical processing of a word occurs autonomously and resembles a “stimulus-driven bottom-up process” (1994, p.6). However, he also emphasizes that no linguistic knowledge or information assists this bottom-up process. Instead, lexical access is transmitted to other components in the language processor in the lexical representation.

In terms of the storing properties of the mental lexicon, the theory of full entry was discussed in early psycholinguistic literature. The whole entry theory argues that every existent word enters a list in the mental lexicon. This also means that every affix is listed in the mental lexicon, although they may be placed in a separate list.

Psycholinguistic research has also investigated what corresponds to the lexical element of the lemma or headword. For example, Pounder (2000) suggests three hypotheses concerning an entry or lemma status. One hypothesis states that every word and its inflectional forms, and word-formations have their own entry. The second states that not every word has its own entry. Instead, regular inflection forms and word-formation forms are part of the entry for the base word and are thus accessed via the base. The third states that entries are dedicated to stems and roots regardless of the regularity of other forms of these stems or roots.

Cognition linguistics states that language works as part of cognition and thus tends to involve the application of networks to language structure. On this basis, declarative language knowledge should exist in a speaker’s mind rather than knowledge of procedures that include an input and output.

(...) a past- tense verb has a suffix after its stem, and that the stem and suffix together comprise the whole of the verb; these are relationships among the concepts “past-tense verb,” “its suffix” and “its whole” (…). We do not know “how to form a past-tense verb” – such knowledge cannot be accommodated, as such in a network. (Simonsen et al., 2001, p.53).

The notion of the network is also described as a mental map. This is shown in Jean Aitchison’s (2012) introduction to the mental lexicon, in which he compares the mental lexicon to the London underground map. Also, Hudson states that a network is analogous to a static map. It is not a set of instructions for getting from point A to point B. On the other hand, however, the network can be used as a guide for activity because of its declarative nature, similar to a map. On this basis, one attempts to create a diagram of the connections in the mental lexicon, which correlates to London’s underground map. Langacker (in Dirven et al. 1995) also believes that language is organized in a network. He suggests that it is common for lexical units, particularly those that are frequently used, to be polysemous, exhibiting a variety of established but interrelated senses. Moreover, each established sense is described “encyclopedically” in connection to one or more cognitive domains (i.e., knowledge structures of varying complexity and sophistication) (Dirven et al. 1995).

2.1.2 Visual word recognition: interactive models

In visual word recognition, the stimulus is analyzed according to its features, which consist of multiple layers of orthographic representation. Rastle (2018) describes the architecture of visual word recognition as follows: features map onto a level of representation that codes abstract letter identity as well as letter position, and these representations then activate representations of known words that are structured morphologically and shaped through experience. (p.53)

On the foundation of this basic structure, it is essential to examine the mechanism inside the features’ transmission. One model of word recognition and lexical retrieval is the interactive-activation model proposed by McClelland and Rumelhart (1981, 1982). The focal element of this model expects that the handling of information during reading comprises a series of levels that involve visual features, letters, and words. This model is utilized to clarify the word superiority effect, in which individuals perceive letters more effectively when they are introduced inside words and contrasted to detached letters as opposed to when they are introduced inside nonword strings (Chase & Tallal, 1990). The word superiority effect postulates that identification and recognition are more accurate if letters are presented in existing words.

Similarly, pronounceable nonwords exhibit the same superiority effect when compared to unpronounceable nonwords (McClelland & Rumelhart, 1981). The interactive-activation model accounts for the word superiority effect based on the concept of cascading processing. Cascadic processing states that activated representation must not reach its response threshold before influencing the activation of other presentations (Cortese & Balota, 2012). The superiority effect and the interaction between the visual features of words are considered in the interactive-activation model, as depicted in Figure 1. The essential thought is simply the presentation of strings of letters on the second layer, activates. Because these activations become more robust, they activate detectors for words on the third layer that are coherent with the letters (when present). Furthermore, these activated detectors create feedback that strengthens the detectors’ activation for the letters in the word. As a result, letters in words are more distinguishable because they get more activation than portrayals of either single letters or letters in an irrelevant setting (McClelland & Rumelhart, 1981). This means that the interactive-activation model consists of several levels of processing, in which the different stages are concerned with forming a representation of the input at a different level of abstraction.

Furthermore, it is essential to note that information flows bi-directionally and fluently between the different levels of representations through the nodes. McClelland and Rumelhart visualized this in the bottom-up and top-down models. It is assumed that visual input takes place simultaneously on different levels and that this applies to more than one feature at a time. Hence, a bottom-up and top-down interaction is created by the interacting data generated by the bottom-up processes, resulting in what one eventually perceives. It is also noteworthy that the detector’s, or in other words, nodes’ interaction can be either “excitatory or inhibitory,” as stated by Rastle (2018). For example, the latter ensures that close but inconsistent levels are inhibited. For example, stimulus D would activate word nodes for door, do, and dark but inhibit

Figure 1 The interactive-activation model of visual word recognition (McClelland & Rumelhart, 1981, 1982)

Abbildung in dieser Leseprobe nicht enthalten

Note. Illustration extracted from Rastle (2018)

word nodes for tree, car, and make (Rastle, 2018). The arrowed lines in the figure represent the levels connected by facilitatory pathways and the knobbed lines of the inhibitory pathways. Cortese and Balota (2012) sum up the bottom-up and top-down processes as follows:

When processing a letter string, letter-level representations activate word-level representations via facilitatory connections. More interesting, letter-level representations are reinforced via top-down activation from the word level. Also, activated representations inhibit competing representations within and between levels so that, eventually, only the appropriate representation reaches its threshold. (p. 161)

The association between levels is expected to occur through a spreading activation process administered by these excitatory and inhibitory cycles. The excitatory interaction increases the degree of activation of their recipients (illustrated with arrows in Figure 2), and the inhibitory connections decrease the enactment of their inheritors (connections marked with circles in Figure 2). Specifically, inhibitory interactions caused by strongly activated nodes on one level constrain representations of weakly activated nodes because they compete for potential clarifications of the visual input (Perry et al., 2008). The inhibition occurs when letter nodes are strongly activated. Furthermore, the activation dynamic occurs either in or adjacent to crosswise levels, whereas connections between non-neighboring levels on the same levels cannot occur. The activation process depends on the excitatory or inhibitory activation of neighboring nodes. Inactive nodes are said to have an activation value of zero or below, while active nodes have a positive value. The nodes of letters associated with excitatory activation with the visual element level nodes are actuated and spread thusly activation to the word-level nodes they are steady with (McClelland & Rumelhart, 1981). If the visual component nodes associated with the letter T (Figure 2) in the first position of a word are initiated, the visual component nodes of the vertical and level lines feed excitatory associations with the letter nodes for T, amidst others. The degree of activation of the nodes with predictable visual features will be pushed over their resting level. The nodes that the visual features have actuated will be contending, whereas the nodes with the highest level of activation will surpass the activation level of the other nodes. The situation wherein the dominant letter node T, which is in the first letter position of a word, will then enact the reliable words with T in their first position

Abbildung in dieser Leseprobe nicht enthalten

Figure 2 The bottom-up and top-down models (McClelland & Rumelhart, 1981,1982, p. 378, p.380)

Although the IA (interactive model) is an essential milestone in word recognition processes, other models that extended the IA model or challenged it have also been proposed. Coltheart et al. (2001) proposed a computational model that extends the IA model of McClelland and Rumelhart. Critics of the IA model have stated that two routes are necessary to process words: a lexical route and a sublexical route (Cortese & Balota, 2012). The main difference between the two models is that the IA model only applies to four-letter words, whereas the dual-route cascaded model can apply to words made up of one to eight letters (Coltheart et al., 2001). The general structure of the DCR (dual-route cascaded) model consists of three routes: the semantic lexical route, the non-semantic sublexical route, and the grapheme-to-phoneme conversion route. All three routes contain several underlying layers. Coltheart et al. (2001) identified two ways in which the different layers interact. These interactions are similar to those in the IA model. The first is an inhibition interaction between the unit's activation (connections marked with arrows in Figure 3). The other method is through excitation (marked with circles in Figure 3), in which the activation of one unit contributes to the activation of other units.

The sublexical route uses GPC (grapheme-to-phoneme-conversion) rules to convert letter strings to phonological representations. The activation of a word’s letters activates the word’s letter units, thereby activating the word’s entry in the orthographic lexicon. The unit in the orthographic lexicon then activates the unit in a phonological lexicon. Figure 3 highlights that the communication between the phonological and orthographic lexicon units is only excitatory, whereas communication between features and layers is one-directional, as in the IA model. The frequency sensitivity of the DRC model is also similar to the IA model. Orthographic units in the lexicon correspond to high-frequency words more quickly.

The GPC, a set of grapheme-to-phoneme conversion rules, is employed in the non-semantic lexical route, which is limited to rules in which a set of letters map onto a single phoneme (Coltheart et al., 2001). In visual word recognition, the rules are searched to find a suitable rule to convert the letter into a phoneme. This process adds activation to that of the lexical route. The GPC route converts letter-by-letter graphemes to phonemes. Specifically, it is essential to note that graphemes such as ph are translated in one phoneme, in contrast to pr, which is converted into two phonemes. Cortese and Balota (2012) highlight the process of the GPC rules with an example for the irregular word “pint,” which violates the pronunciation rules of the model. This is due to the fact that the application of GPC rules to the word “print” would harvest a pronunciation rhyming with “mint.” To read this word correctly, the lexical route is required because converting letter-by-letter would be misleading. However, to recognize nonwords, the non-semantic lexical route with the employment of GPS rules is required because there are no entries in the mental lexicon that could be retrieved. Still, in processing words, both routes, lexical and sublexical, are activated, although the sublexical route has to run faster in case of irregular forms since it has to block regular representations. The speed of processing in the sublexical/nonlexical route depends on the frequency of the visual input. High-frequency words are easier to process since they have higher resting activations than low-frequency words.

Furthermore, conversion occurs until all letters in a word are encoded. The next aspect of the DCR model might be crucial to the experiment in the present thesis. Coltheart et al. (2001), based on Forster and Davis's (1991) masked form priming experiment in a naming task, reported that the nonlexical route only operates from left to right, meaning that facilitation of the target word preceding a prime can only occur if the letters in the onset share a phoneme or letter. Coltheart et al. (2001) state that the lexical-semantic route is a system of semantic representations containing phonological and orthographic representations for each word (right side of Figure 3). The lexical-semantic route is adopted from the IA model and extended with different parameter settings. It should be noted that the DCR model includes a dual-route dogma compared to the IA model. It emphasizes the need for two routes for lexical processing for either words or nonwords.

Seidenberg (2012), however, challenges this approach and model by arguing that it is not sufficient to explain results in behavioral data. He argues that the model has been developed upon formalisms and empirical data and therefore does not bear “robust theoretical generalization that explains target phenomenon” (Seidenberg, 2012, p.197).

Figure 3 DRC model (Coltheart et al., 2001)

Abbildung in dieser Leseprobe nicht enthalten

On the basis of the notion of a two-route activation model for visual word recognition, Seidenberg and McClelland (1989) introduced their connectionist parallel distributed processing (PDP) model. The main difference to the previously discussed models is that different units are processed simultaneously at a given level. The distributed part of the model means that each word is linked with a distinctive activation pattern across a set of units used for processing every word (Cortesa & Balota, 2012). Apart from distributed representations, the framework includes distributed knowledge about the relationship between entitles, encoded within a high number of connections weights, which then encode further mapping (Plaut et al., 1996).

[...]

• Université: University of Siegen
• Note: 1,3
• Auteur: Dialechti Koutsanta (Auteur)
• Année de publication: 2022
• Pages: 66
• N° de catalogue: V1246083
• ISBN (PDF): 9783346674753
• ISBN (Livre): 9783346674760
• Langue: anglais
• mots-clé: inflectional morphology, lexical decision task, masked priming, simple past, frequency, psycholinguistic, quantative linguistics