The Effects of Semantic Priming on the Detection of Words

Table of Contents

Abstract

List of tables

List of figures

1 Introduction
1.1 Theoretical background
1.2 Question
1.3 Hypothesis

2 Material and Methods
2.1 Experimental subjects
2.2 Stimulus material
2.3 Test procedure

3 Results
3.1 Descriptive Statistics
3.2 Inferential statistics

4 Discussion
4.1 Conclusion
4.2 Critique
4.3 Outlook

Bibliography

Abstract

In order to retrieve information more efficiently and quickly, our central nervous system makes use of implicit preactivation of associative neural networks (Collins & Loftus, 1975). In this study, 78 participants were instructed to identify a sequence of word pairs consisting of either German words, nonwords or pseudowords within a lexical decision task. The procedure was carried out under three different conditions: a) no associations within a word pair, b) connection through general second-level cooccurrences, and c) connection through lemmatized second-level cooccurrences. The analysis of variance revealed highly significant differences in reaction time and error rate between lemmatized second-level cooccurrence compared to general second-level cooccurrences. Both, error rate and reaction time, were lower for lemmatized second-level cooccurrences. Stimuli consisting of words with second-level association had a positive effect on the reaction time and error rate. This could be proven due to a stimulus onset asynchrony of 50ms, avoiding semantical competition that could cause inhibitory effects on the reaction time. Linear regression also revealed that lemmatized second-level cooccurrences had a greater influence on the reaction time up to the target and the error rate compared to general second-level cooccurrences. This information could be used to improve models that explain the process of word recognition by adding the influence of the lemmatized second-level cooccurrence.

Keywords: AROM, Spreading Activation Network Model, semantical priming, IAM, Interactive Activation Model, general associations, lemmatized, cooccurrences, error rate, reaction time, second-level cooccurrences, lexical decision task, SoA, stimulus onset asynchrony, Bim, Jo, JoBimText, semantical association, priming effect

List of tables

Table 1: Comparison of mean group differences for the reaction time

Table 2: Comparison of mean group differences for the error rate

Table 3: Model overview of the multiple linear regression of all predictors

Table 4: Model overview of the multiple linear regression of second-order cooccurrences

List of figures

Figure 1: Spreading activation model (after Collin & Loftus, 1975)

Figure 2: Interactive Activation Model (after McClelland & Rumelhart, 1981)

Figure 3: Associative Read-Out Model (Hofmann et al., 2011)

Figure 4: Comparison of the different types of cooccurrences

Figure 5: Test procedure and conditions of cooccurrences

Figure 6: Chronological sequence of one trial

Figure 7: Mean reaction times of the three conditions

Figure 8: Mean error rates of the three conditions

Figure 9: Linear regression showing variance of the predictors with the criterium reaction time (left) and error rate (right)

Figure 10: Linear regression showing variance of second-order cooccurrences with the criterium reaction time (left) and error rate (right)

1 Introduction

1.1 Theoretical background

The brain of all human beings is fundamental and responsible for how we perceive the environment we live in (Pinel, 2018). There are various features to the brain that allow processing, encoding, consolidation and retrieving of information to ensure the evolutionary purpose of the development and the survival of our species (Craick & Lockhart, 1972). One of these features of the brain is priming. Within psychology, priming is a commonly used tool to influence the response time on a subsequent stimulus (target) by prior exposure to another stimulus (prime). The participants’ response to the stimulus is not determined by their intention, but instead happens unconsciously (Myers, 2008). The first stimulus activates parts of the brain that are important for memory right before a certain reaction to the stimulus is carried out (Anderson, 1976).

There are several types of priming that can be applied, such as emotional priming, perceptual priming, conceptual priming and semantical priming. Emotional priming is characterized by priming stimuli that cause changes in emotion and on that way the response time on subsequent target stimuli (Fazio, 2001). Perceptual primes focus on the form of the items in order to use them as a prime, if they reflect similarity to some extend and conceptual priming is used in cases where the meaning of a prime and target is similar (Vaidya, Gabrieli, Monti, Tinklenberg & Yesavage, 1999).

The following will be confined to semantic priming in particular. Within semantic priming, the priming word and the target word are usually from the same semantic category (McNamara, 2005). If the meanings or associations of these words relate to one another, associative networks are activated that contain a mental representation of words. The spreading activation network model invented by Collin and Loftus in 1975 aims to illustrate the process by which a specific word is selected in the brain. The neural network consists of vertices, meaning the terms, and the associative connections, that are represented through lines. After one term is activated, the selection of other similar terms is activated by using the associative connections (Collins & Loftus, 1975).

Abbildung in dieser Leseprobe nicht enthalten

Figure 1. Spreading activation model (after Collin & Loftus, 1975).

After another model, the “Interactive Activation Model” (IAM), was invented in 1981, semantic priming became especially important (McClelland & Rumelhart, 1981) since the IAM gives an insight into the Blackbox that is referred to the time between the stimulus and the response and divides the process of letter and word recognition into different steps (Hofmann & Jacobs, 2014). The invention of the IAM was inspired by the work of Hubel and Wiesel that discovered neurons in the visual cortex which only respond to certain kinds of stimuli such as vertical lines or horizontal lines (Hubel & Wiesel, 1962). The process is then divided into feature units, letter units and word units that can be seen in figure 2 (McClelland & Rumelhart, 1981).

Abbildung in dieser Leseprobe nicht enthalten

Figure 2. Interactive Activation Model (after McClelland & Rumelhart, 1981).

Another model, the “Associative Read-Out Model” (AROM) replenishes the IAM by adding a semantic associative level (Hofmann, Kuchinke, Biemann, Tamm & Jacobs, 2011). The AROM shows word recognition in relation to semantically related word by cooccurrences (Figure 3). In this case, the AROM is based on the “Multiple Read-Out Model” (Grainger & Jacobs, 1996). The AROM intends to predict how easily a word is recognized in a specific verbal-semantic-associative context (Hofmann et al., 2011).

Abbildung in dieser Leseprobe nicht enthalten

Figure 3. Associative Read-Out Model (Hofmann et al., 2011).

Semantic priming effects can be shown within lexical decision tasks and investigated on that way (Meyer & Schvaneveldt, 1971). Therefore, we differentiate between three different types of cooccurrences in this experiment. These consist of associations between first-level cooccurrences, words that share semantic association by general second-level cooccurrences, as well as words that share semantic association by lemmatized second-level cooccurrences (Figure 4). First-level cooccurrences reflect the probability of two different words being found in the same sentence (Hofmann & Jacobs, 2014). Second-level cooccurrences reflect the number of commonly associated words, meaning that words do not have to exist within the same sentence, but can instead be explained by synonymous words (Hofmann & Jacobs, 2014). In contrast to that, lemmatized second-level cooccurrences describe words that are connected by commonly associated verbs.

This last condition is based on a component of the “JoBimText Visualizer” (Biemann & Riedl, 2013). By lemmatization, every word is reduced to their basic form, the lemma (Collins English Dictionary, 2015). In this case, the basic form is a semantically related verb. This form of semantic relation can be found within the interactive visualizing-component of the JoBimText (Biemann & Riedl, 2013). JoBimText is a semantical analyzing program and an open source platform for large-scale distributional semantics based on graph representations, trying to establish a system for semantical relations. By using this program, problems like parsing and substitutability can be addressed.

Abbildung in dieser Leseprobe nicht enthalten

Figure 4. Comparison of the different types of cooccurrences.

1.2 Question

This experiment was conducted to replicate the results found by Jonides & Mack in 1984. They postulated a reduction in the reaction time within one word-sequence due to semantic effects (Jonides & Mack, 1984). In the following, there will be a differentiation between the effects of lemmatized second-level cooccurrences, the effects of general second-level cooccurrences, as well as the effects of non-semantic associations on the word recognition process. The different types of cooccurrences will be compared in terms of their reaction times and error rates in order to draw conclusions about their level of influence on the detection of words in a semantic context and with regards to the improvement of the “Associative Read-Out Model”.

1.3 Hypothesis

1. The error rate and reaction time up to the target stimulus without semantic association differ from the error rate and reaction time up to the target stimuli of global second-level cooccurrences.
2. The error rate and reaction time up to the target stimulus differ from semantic associations by means of global second-level cooccurrences compared to the error rate and the reaction time up to the target stimulus using lemmatized second-level cooccurrences.

2 Material and Methods

2.1 Experimental subjects

The sample size was N = 78, consisting of 43 women and 35 men. The mean age was 31.3 years. The participants had to be right-handed, of legal age and native speakers of the German language as a requirement to participate in this experiment. In addition to that, data about the level of education or profession was also collected, showing a heterogeneous group of participants (appendix a). Exclusion criteria consisted of chronic sleep deprivation, acute consumption of medication or drugs that could have an effect on the reaction time and error rate, as well as psychosomatic disorders because of possible influences on the variables (Cox, Abramson, Devine, & Hollon, 2012). Retrograde exclusion criteria consisted of dyslexia, an increased error rate within the lexical decision task and a difference higher than three standard deviations between the individual reaction time and the mean reaction time. The exclusion criteria being taken into account, three men and nine women were excluded from the study due to an error rate higher than 30 %, leaving a total of 66 participants with a mean age of 30.9 years (SD = 13.5) that ranged from 19 to 63 years.

2.2 Stimulus material

The stimulus material consisted of 300 nouns of the German language. The target stimuli consisted of 150 nouns of the German language, 75 pseudowords and 75 nonwords. Pseudowords scarcely had phonetic or orthographic neighbors and were generated by switching two letters to create new combinations consisting of vowels and consonants. Nonwords consisted of random strings. The relation between the prime stimuli and the target stimuli was one of the three factor levels. Under the first condition the prime and the target did not have any semantic association (1stC = 0). Under the second condition prime and target were linked by second order cooccurrences (2ndC) and under the third condition prime and target were linked by lemmatized second order cooccurrences (2ndC-L). Each word used as a prime or target stimuli consisted of three to eight upper-case letters. The size of the stimuli was adjusted to the size of the fixation cross in the experimental procedure. The typeface of the stimuli was “Times New Roman” and the type size was 40 to ensure a normal reading distance of a 1/3 angle degree per letter (Radach, Huestegge & Reilly, 2008). The distance between the monitor and the participant was also calculated.

One of the word corpora that the stimulus material in this experiment was based on is BAWL-R (Võ, Conrad, Kuchinke, Hartfeld, Hofmann und Jacobs, 2009). The Leipziger corpora collection (Biemann, Heyer, Quasthoff & Richter, 2007) was also used, as well as the JoBimText generator, since it contains lemmatized second-order cooccurrences (Biemann & Riedl, 2013). There were no synonyms or repetitions of a noun. The collection of data took place in dark and quite surroundings to ensure a standardized procedure with minimal stimuli of distraction. The collection of data was done on personal electronical devices of the experimenter by using the software PsychoPy v3.1.

2.3 Test procedure

Before conducting the experiment, participants were asked if they wanted to participate in an experiment about “lexical decision tasks”. The purpose of the experiment was not mentioned at any time before the procedure. To avoid possible influences and misconceptions about the task, participants received a standardized and clear instruction on the monitor ahead. During the experiment, the experimenter stayed in another room to avoid effects due to their presence. At the beginning of the experimental procedure, the participants read the instructions at their own pace and later confirmed them to begin. The instruction contained the information that this lexical decision task had to be done in the quickest and the most accurate way possible. After confirming the instruction, participants had to go through one practice trial. During this practice trail, the experimenter stayed in the room to answer upcoming questions about the experimental procedure.

T he practice trial contained twenty stimuli that were linked through common associations of first and second order cooccurrences. After the practice trial followed three sections that were divided by two breaks. After each section of the experimental procedure, the rate of correct answers was shown to the participants on the monitor as a percentage. Every participant received the same sections with randomized sequences within them to eliminate variables of potential influence. Aside from the 50 words of the German language for each section, there were 25 pseudowords and 25 nonwords. These words were distributed equally to the sections depending on their length and type. Words of the German language, pseudowords and nonwords were not presented more than three times in a row. A roundup of the test procedure and the different conditions of cooccurrences can be seen in figure 5.

Abbildung in dieser Leseprobe nicht enthalten

Figure 5. Test procedure and conditions of cooccurrences.

[...]