What I acquired in reading “The Acquisition of Chinese Characters”
2008/03/28 at 7:54 pm | Posted in Cognition, Research | Leave a commentThe authors of this article, Xing, Shu, and Li, investigated the awareness of phonological structure of Chinese phonograms based on the corpus analysis and the self-organized model simulation. This issue is suggested to be relevant to the phonological awareness of Chinese readers. They classify the regularity of Chinese phonograms based on the reliability of phonetic representing the pronunciation of whole character: regularity, semi-regularity, and irregularity. The consistency of Chinese phonograms refers to whether all characters that share the same phonetic are regular. Xing et al. recommended the effects of regularity and consistency reflecting the acquisition of Chinese characters in the children learning.
Their motivation came from the empirical researches that examined the regularity and consistence effects in the learning process of elementary school students through grades. As for the regularity effect, many researchers have found that students younger than grade 3 perform well in writing and pronouncing the regular character. This phenomenon also appeared in these students who learn new characters with regular phonetic. The previous studies brought the authors’ interest to the consistency effect show that the students’ knowledge of consistency grows with their amount of vocabulary.
The authors’ goal of analyzing the corpus is to decide how the critical factors, character frequency, regularity, and consistency, comprise the students’ phonological knowledge with their increasing grades. In the simplified Chinese phonograms, a quarter of phonograms that children learned in the elementary school are the regular characters, and the amount of regular characters increase with grade. The knowledge of a phonetic representing the pronunciation of whole character should be established in the mind of students at very early grade. But their corpus analysis implies that the awareness of consistency might come to the mind of students in late grades. Till the 2nd grade, students learned relative rare cases of consistent characters relative to those of regular characters.
To speculate the acquisition process of pronouncing phonograms and the relative developmental stages of regularity and consistency effects, Xing et al. used the self-organization neural net in modeling this process. This network would assign a node on the knowledge map to the each input data pattern. This characteristic makes the network capture the similarities between inputs according to their statistical structures. Their simulations are consistent with the previous empirical researches on the interaction of character frequency and regularity/consistency. In addition, their simulations show that learning new characters has been influenced by the regularity at early grade, but influenced by the consistency at later grade.
This study provides an explicit demonstration of acquisition process in learning the orthography and phonology mapping. In my view, we could make this model more plausible on several ways. A child eventually performs as skilled as the adult in reading characters. A model with the advantage in accounting human performance should capture the development of this skill through novice to experienced individual. As Xing et al. suggested, more and more studies have considered the consistency as a continuous variable rather than a categorical variable of which Xing et al. thought. In addition to this, their definition of consistency just focuses on the phonetic with reliable pronunciation. This definition ignores the numerous Chinese characters with irregular or unpronounceable phonetics. Any computational model which could capture the processing of acquiring the knowledge of these words would extend our understanding the questions about the orthography and phonology mapping.
Cautions of using delayed naming latency
2008/03/19 at 9:23 pm | Posted in Psychologist, Research | Leave a commentNaming task is a frequently used method in the psycholinguistic discipline. The procedure is simple: the participants read the word aloud as quickly and accurately as they can, and the data collection is easy: the experimenters are able to record the time during stimuli displayed and the voice key triggered (the sound blaster card is the frequently available device presently). This elapsed time, which is termed naming latency, is the dependent variable the researchers have to analyze in the end. This task is popular in the discipline because it is sensitive to the lexical factors, such as word frequency, which construct the fundamental processing of human language.
As the other dependent variables in psychological research, the naming latency reflects not only the nature of lexical processing but also the demand for executing pronounce. To measure the naming latency actually been attributed to the lexical processing, the delayed naming was introduced to measure the part of latency purely spent on the execution level. If the time for execution is estimated, then the rest of naming latency will be the time for completing the lexical processing. This notion stands on a core assumption of delayed naming: a sufficient long delay always effectively isolate the execution level processes. However, this assumption has never been confirmed since the delayed naming was introduced to the psycholinguistic discipline. This is the Kawamoto and his colleagues’ mission in their study (Kawamoto et al., 2008).
Sternberg and his colleagues’ theory of motor control and execution specifies the constituents of delayed naming latency, and Kawamoto’s crew find one of them are not as fixed as the traditional core assumption. In Sternberg’s theory, a delayed naming latency contains the base time to detect the response signal (Tb), the retrieval time for the stimulus (S), and the unpacking time for the subprogram of the stimulus (U). Because our mouth needs some time to prepare for articulation, Kawamoto’s crew suggest an alternative term for representing the execution time for the stimulus (C). In the experimental circumstance, the execution time is the interval of articulator onset (start to move necessary muscle) to acoustic onset (start to generate the acoustic energy); I use the short term AAI in the following. The components, S, U, and C compromise the chain of execution that the processing time is anticipated to be measured by the delayed naming. There is one internal factor and one external factor Kawamoto team suggests causing the variation of AAI. The internal factor is the manner of articulation, plosive or non-plosive word-initiation. Past studies (in summary of Lehiste, 1970) have found the fact that the intra-oral pressure is necessary for building up plosives, so that their acoustic energy typically generate about 50-100 ms later than non-plosives. It has to be noted that a significant difference between
two critical conditions is generally as short as 25-40 ms. The external factor cared by Kawamoto team refers to the preparation for releasing the chain of execution. In the traditional view of delayed naming, this chain of execution initiates after the signal to response showed. On the Kawamoto’s side, this chain has began its operation following the pre-execution level process and before the signal to response. If Kawamoto team’s argument is right, to estimate the naming latency for the execution level processing is impossibly. However, worse than this is that the variation of AAI would cover the range from before to after the signal to response due to the content of experimental instruction.
Kawamoto team’s experiments indicated that the longer delay duration not only increases the magnitude of AAI but also decreases the difference of acoustic latency between plosives and non-plosives. These results support their argument that the manner of articulation indeed decides the length of AAI and the initial time of generating acoustic energy with the delay duration. Their final experiment showed that the chain of execution always starts in advance of the signal to response. In this experiment, the difference of acoustic latency between plosives and non-plosives appears in the standard naming and the delayed naming with partial preparation (e.g., put the tongue on the teeth while waiting signal). These findings mean the delayed naming just cutting the tail of the execution level processing, rather than all of the processing in the traditional view.
Prof. Kawamoto proposes his worry about the case like delayed naming generally affecting the tasks in the psycholinguistic discipline. I am still optimistic that the standard naming task is able to offer valid latency data in measuring the impact of lexical processing. Firstly the chain of execution is temporarily abandoned in the standard naming. There is no time for retrieval and unpacking in this circumstance. The variation of AAI is what we only have to concern, and the manner of articulation is the primary factor influencing its variation. With the well-designed experimental comparison and the appropriate analysis technique, we could restrict its affection on the naming latency.
Kawamoto, A. H., Liu, Q., Mura, K., & Sanchez, A. (2008). Articulatory preparation in the delayed naming task. Journal of Memory and Language, 58(2), 347-365. DOI: 10.1016/j.jml.2007.06.002
Philosophy of connectionism
2008/03/06 at 2:56 pm | Posted in Cognition, Thought | Leave a commentBefore starting learn the computational model in following project, I fueled myself by reading the classical articles of establishing, challenging and defending the connectionism. Before 20th century, several philosophers had provided the idea of associationism. There were various ways of assocating two ideas proposed by them. For example, John Locke (1637-1704) argued that some ideas associated by natural connections while others associated through custom. The associationists left the contemporary connectionists a philosophical notion of source of knowledge: each idea is combinable for each other through some appropriate association manner. Approaching the 20th century, the early psychologist William James (1842-1910) suggested a goal that searching the evidence of association in the operations of neural mechanisms.
The wave of connectionism raised during 1980, since the classical PDP models were proposed by James McClelland, David Rumelhart, and the other leading researchers. Throught these years, I always have a question why they prefer the distributed representation as their framework of models. From the article of McClelland, Rumelhart, and Hinton, the appeal of parallel distributed processing, I found this preference comes from their philosophical choice of knowledge nature. In their words, the learned knowledge is stored within the connections among neuron-like units rather than isolated representation nodes. During the processing, the learned knowledge is “recreated” through the activations among units. Based on this choice, simple connection strength modulation mechanisms are the approriate account for adjusting the strength of connections between units based on information locally available at the connection. Now, another question I possess is whether the preference of speculation, such as feedfoward connection or inclusion of hidden layer, depends on this philosopical choice?
==== End at the morning, March, 6, 2008
Blog at WordPress.com.
Entries and comments feeds.
暗微幽光 