论文部分内容阅读
AbstractTwo lines of research on eye movements in reading are summarized. One line of research examines how adult readers identify compound words during reading. The other line of research deals with how a specific reading goal influences the way long expository texts are read. Both lines of research are conducted using Finnish as the source language. With respect to the first research question, it is demonstrated that compound words are recognized either holistically or via their components, depending on the length of the compound word. Readers begin to process whatever information is readily available in the foveal vision(i.e., either the whole-word form or the initial component). The second line of research demonstrates that(1)a specific reading goal is capable of exerting an early effect on readers' eye fixation patterns,(2)time course analyses based on eye movement patterns can reveal interesting individual differences, and(3)working memory capacity is linked to the efficiency to strategically allocate attention as well as to encode information to and retrieve it from the long-term memory. It is concluded that the eye-tracking technique is an excellent research tool to tap into the workings of the human mind during the comprehension of written texts.
Key wordseye movements, word recognition, compound words, text comprehension, working memory capacity.
In this article I summarize two lines of eye movement research that we have conducted over the past several years in Finland. One line of research deals with how competent adult readers identify compound words during reading. The other line of research deals with how a specific reading goal(we call it the reading perspective)influences the way a long expository text is read and what is remembered of the text*. In both lines of research we have applied the eye movement technique by registering readers' eye movements when they read silently single sentences or longer texts for comprehension. As will become evident, eye-tracking has proved to be a highly valuable research tool to tap into the ongoing mental processes as they evolve over time when people read texts.
Identifying Compound Words During Reading
By definition, compound words are words that consist of 2 or more components. An English example could be headache, which consists of two components, head and ache. The word meaning is derived by putting together the meanings of the two components to yield something like 'pain in the head'. Finnish is an agglutinative language, in which word compounding is very productive. Novel compounds can readily be constructed out of existing words, particularly of nouns. Finnish compounds can include several components, as in lumipallosota, which means 'snowball fight' and consists of 3 components, lumi(snow), pallo(ball), and sota(fight). All Finnish compounds are concatenated in that spaces never appear between the components. This feature can also be found in other alphabetic languages, such as German, Dutch, or Swedish. English, on the other hand, is more irregular in the sense that compounds may appear in one of 3 forms: unspaced(e.g., snowball), spaced(e.g., traffic light), or hyphenated(e.g., eye-tracker). Chinese is also highly interesting in the sense that in Chinese most words are compounds in that they consist of more than one character. Finnish compounds are similar to Chinese ones: in neither language spaces are added to signal where the boundaries exist between the components.
One key question we have addressed is "To what extent are compounds identified holistically and to what extent via their components?" In order to answer this question we have either(1)manipulated the frequency of the whole word while matching for the frequency of the components(and other relevant features, such as length), or(2)manipulated the frequency of the components as separate words while matching for whole-word frequency. This general method was introduced by Taft to study the recognition of morphologically complex words[1]. The logic is as follows. If word frequency is found to have an influence on the identification process, it is assumed that the holistic route to compound word meaning gets activated during the course of processing. More specifically, the holistic route is implicated when a low-frequency compound word takes longer to process than a high-frequency compound word, given that the two types of words are equated for the frequency of the components. On the other hand, a componential route is implicated, for example, by a finding that a compound with a low-frequency first component takes longer to process than a compound with a high-frequency first component, given that they are equated for whole-word frequency. Finally, the presence of both constituent frequency and word frequency effects suggests a dual-route model, in which both access routes are operative.
In our studies[2~7]we have observed that the length of the compound word is a crucial factor in determining the extent to which a compound word is identified holistically or via its parts[2]. Until now we have only studied the processing of 2-component compounds, so everything that is said below concerns compounds of that sort(most of the compounds used in our studies have been noun-noun compounds, such as headache). There are some other eye-tracking studies on compound word processing conducted in English and in German[8~12].
Our results demonstrate that when a compound word has no more than 8 letters (as in sivuovi=side door), the word is processed holistically. This becomes apparent in that(1)gaze duration(i.e., the duration of all fixations landing on the word before exiting it to the right or left)is reliably influenced by word frequency, but(2)the frequency of the first component does not affect gaze duration. Word frequency exerts an early effect on the processing, as indexed by the effect being observed already in the first fixation duration. All this is taken to suggest that shorter compounds are identified as single wholes and that the access to the components is not an integral part of the process.
The story is quite different for longer compounds(i.e., compounds that contain 12 letters or more). For long compounds, an early effect of first constituent frequency is observed. The duration of first fixation on long compounds was found to be longer when the first constituent was infrequent than when it was frequent. It may be noted that the first fixation typically also landed on the first constituent. A highly reliable effect of first constituent frequency was also observed in gaze duration. When the frequency of the second constituent was manipulated, its effect was not observed in the first fixation duration, but only from the second fixation onwards(it is worth noting that typically the second fixation on the word landed on the second constituent). The duration of second fixation was longer when the second constituent was infrequent than when it was frequent. An effect of second constituent frequency was also observed in gaze duration. The pattern of these results is compatible with the view that long compounds are identified via their components by first accessing the first constituent followed by the access to the second constituent. The time course of effects nicely supports this view, as we obtain an early effect of first constituent frequency and a relatively later effect of second constituent frequency.
However, the story is not quite simple as that. For long compounds we have also obtained an effect of whole-word frequency that is statistically reliable from the second fixation onwards. There is even a marginal tendency for a word frequency effect in the first fixation duration although the effect size is only 4~5ms. A reliable effect of word frequency suggests that also the holistic route is in operation in the processing of long compounds. However, it does not come into play as early as the decomposition route. In other words, the decomposition route is initially prioritized over the holistic route in the processing of long compounds, but the two routes appear to run in parallel(i.e., the decomposition route does not finish before the holistic route becomes in operation).
Combined with the results obtained for shorter compounds, the picture that emerges is one where, after fixating the compound, the reader starts processing whatever high-acuity information is available from the word. If all the letters of a compound fall onto the fovea when it is fixated, as is typically the case with shorter compounds, the holistic route gets activated from the first moment onwards and the decomposition route does not become activate at all(or only very minimally). On the other hand, when some of the letters fall outside the foveal reach when the compound is initially fixated, as is the case with longer compounds, the decomposition route initially wins over the holistic route. Thus, the identification process is initiated by first attempting an access to the first constituent, followed by an access to the second constituent and to the word as a whole. As mentioned above, the holistic route is assumed to run in parallel with the decomposition route, although it completes its job later than the decomposition route(i.e., it needs to accumulate information also about the second component).
As is evident from above, the identification of long compounds appears to be a relatively serial process in the sense that the processing proceeds from the access of the first component to the access of the second component. As a further test of this seriality assumption, we have recently conducted an eye-movement contingent display change experiment[13]. This technique allowed us to manipulate what is initially available in the parafoveal vision of the second constituent when the reader is fixating on the first constituent. In the display change condition, we replaced the last letters of the second constituent with visually similar letters(the first two letters were kept intact), which made the second constituent to appear as a non-word (i.e., pioneeriryhmä=pioneer group initially appeared as pioneeriryknÖ). During the saccadic eye movement made from the first to the second constituent, the last letters were changed back to what they should have been in the first place. Due to saccadic suppression, the change itself is left unnoticed, namely those trials where the system was not able to make the display change during the saccade were excluded from the analyses. The display change condition was compared to the no-change condition, where the second constituent was available throughout the target word viewing.
The results showed a very noticeable parafoveal preview effect. Gaze duration on the compound was 101ms longer when the second constituent initially appeared as a non-word than when it was kept intact. This preview effect is significantly bigger than what is typically observed for parafoveal previews that are separated from the foveated word by a space[3,13]. Despite this big preview effect, it is quite remarkable that the effect did not show up in the fixation time spent on the first constituent, but the effect only surfaced after the second constituent was fixated. We take this as strong evidence in support of the seriality view, according to which the initial stages of processing long compound words are devoted to the attempt of accessing the first constituent, and it is not until the second constituent is fixated that its lexical access is attempted. Note however that readers do process some features of the second constituent while fixating on the first constituent; otherwise we would not have obtained the big preview effect. But importantly, whatever information is picked up of the second constituent parafoveally it does not affect the fixation time on the first constituent. It is exactly this aspect of the data that supports the seriality view. We currently think that the information picked up parafoveally of the second constituent is orthographic information(i.e., information about the letter identities)and not lexical information[9,10].
To sum up the first section, I want to point out that our research on compound word identification has not only taught us a good deal about how morphologically complex words are recognized, but it has also proved to be a good testing ground to study how the eyes and attention is guided in text during reading. It has also demonstrated the fruitfulness of the eye-tracking technique, which has allowed us to tap into the exact time course of compound word processing. Had we used a single processing measure, such as lexical decision time, we could not have been able to observe many of the effects discussed above(the differentiation between early versus late effects).
As far as I know, all the eye-tracking studies conducted on compound word processing have so far dealt with alphabetic languages, such as Finnish, German, and English. It would be highly interesting to extend the results to non-alphabetic scripts like Chinese. Chinese is also interesting in the present context in that most Chinese words are compound words in the sense that they consist of two or more characters. In order to get a better understanding of the basic reading processes in Chinese, I think it would be important to find out about how the meaning of Chinese words is derived out of its component characters and their meanings.
Effects of a specific reading goal on reading long expository texts
In the second part of my article I review our eye movement research on text comprehension processes. I focus particularly on the effects of a specific reading goal on on-line text comprehension as it is reflected in the readers' eye movement records. The starting point to our research was the classic study of Anderson and Pichert who investigated what consequences a perspective adopted by a reader may have on what he or she remembers of the text[14]. We have also tested the memory for text after reading, but our primary interest has been on the effects of a reading perspective on the ongoing comprehension processes.
In our studies[15~17] we have given the readers a specific reading goal, a reading perspective, from which to approach the text. For example, before reading a text on familiar diseases(flu, diarrhea, AIDS, and chicken pox)the reader was given the following reading perspective (or goal): "Imagine that you are going to give a health education class for elementary school pupils. You are supposed to tell them about flu: for example, what causes it, how you can treat it, and how to prevent from getting the disease. Now read the text in order to be able to do the job." This kind of reading perspective renders some text information relevant and other text information irrelevant. Specifically, information given in the text about flu is perspective-relevant, while information about the other diseases is perspective-irrelevant. We selected 10 sentences from the text that provided information about the flu that was considered, according to a pre-test, to be typically associated with this disease. These sentences were matched in length and in average word frequency to 10 irrelevant sentences that conveyed information about diarrhea.(The reading perspective was counterbalanced across participants so that for half of them diarrhea was the target disease, in which case the flu sentences comprised the perspective-irrelevant sentence set.)
We were also interested in what role working memory capacity may play in on-line text comprehension. Thus, we administered the reading span text[18]to our participants. On the basis of the test, we formed a group of high-capacity, a group of medium-capacity, and a group of low-capacity readers. To simplify the story a bit, below I only summarize the data for hig-and low-capacity readers. According to one influential theory of working memory[19], working memory capacity is related to the efficiency with which attentional resources are devoted to task-relevant information and away from task-irrelevant information. If this is indeed the case, high-capacity readers should be more efficient than low-capacity readers in allocating their visual attention to perspective-relevant information. In other words, we reasoned that high-capacity readers should demonstrate a greater effect of reading perspective than low-capacity readers.
Three types of sentence-level eye fixation measures were computed for the target sentences (10 perspective-relevant, 10 perspective-irrelevant sentences):(1)progressive first-pass fixation time=the summed duration of fixations landing on an unread part of a sentence before exiting it;(2)first-pass rereading time=the summed duration of fixations returning to already read parts of a sentence before exiting the sentence; and(3)look-back fixation time=the summed duration of fixations launched from subsequent text back to the target sentence. By adding up the first two measures(progressive first-pass fixations and first-pass rereading fixations)we obtain the first-pass fixation time, which is a measure of the overall time readers spent looking at the target sentence before moving to the next. The three measures mentioned above were designed to tap into the time course of processing, the progressive first-pass fixation time indexing the most immediate effect and look-back fixation time a delayed effect[20].
In our first eye movement study[16]on this topic we observed an effect of reading perspective both in the first-pass fixation time and in look-back fixation time. Readers spent significantly longer time reading the perspective-relevant sentences than the perspective-irrelevant sentences.(It should be mentioned that in all our experiments we have also found a better memory for perspective-relevant than perspective-irrelevant information.)Moreover, we observed differences in the timing of the perspective effect among our reader groups. They all demonstrated a reliable perspective effect in processing, but in different points in time. The high-capacity readers demonstrated the effect already in the first-pass fixation time, whereas the low-capacity readers showed it only in the look-back fixation time(i.e., with some delay). These data are consistent with the strategic allocation of attention view of working memory[19], according to which high-capacity readers are better able to efficiently allocate attention to task-relevant information and away from task-irrelevant information.
In the follow-up study[17], we wanted to compare the effect of reading perspective for a text on familiar contents to that observed for a text on unfamiliar contents(in the 2002 study, we used a text on unfamiliar contents; the text described four remote countries). The text with familiar contents described four diseases(see above)the readers had ample prior knowledge about, while the text with unfamiliar contents described four largely unknown diseases(trigeminusneuralgy, typhus, cystic fibrosis, and scleroderma). Theoretically, the comparison between texts of familiar versus unfamiliar contents is interesting as it makes contact with the notion of long-term working memory put forth by Ericsson and Kintsch[21]. According to this conception, working memory capacity is related to the ability to efficiently use prior knowledge during encoding. If so, high-capacity readers should be able to make better use of their prior knowledge than low-capacity readers in encoding to memory perspective-relevant information of the familiar disease text.
This was in fact what we found. In the familiar diseases text, the high-capacity readers took no longer time to read the perspective relevant than perspective irrelevant sentences. Yet, after reading they showed a clearly better memory for relevant than irrelevant information. We argue that this is because they have fast access to prior knowledge that can be readily brought to bear to help encode the relevant information to memory during reading. That is why they did not need to fixate any longer on relevant than irrelevant target sentences. The low-capacity readers, on the other hand, did need to do so. Their first-pass and look-back fixation times were significantly longer for perspective relevant than perspective irrelevant sentences. All in all, the pattern of our results is consistent with the notion of long-term working memory proposed by Ericsson and Kintsch[21].
The unfamiliar diseases text replicated our previous results[16] in showing that the high-capacity readers showed the perspective effect in an earlier phase of processing than the low-capacity readers. In order to get a more detailed picture of the allocation of visual attention to perspective relevant versus irrelevant text segments, we ran a control experiment, where no specific reading perspective was given. The participants were simply told to read the text in order to be able to write a summary of the text contents after reading. Comparisons to the control group made it possible to tease apart the nature of the perspective effect: to what extent the perspective effect is due to slowing down on relevant information and to what extent it is due to speeding up on irrelevant sentences. These comparisons were made separately for the high-capacity and low-capacity readers. In order to be able to do so, we tested in the control experiment a group of high-and low-capacity readers who had not participated in the experiment proper.
The results showed that, in comparison to their controls, the high-capacity readers invested extra processing time to relevant information already during the initial encounter of that information. In other words, their progressive fixation times were significantly longer on the relevant sentences than those of the control group who were not given a specific reading perspective. On the other hand, they did not initially slow down their reading of the irrelevant sentences(again compared to the progressive fixation times of their controls). The pattern of results looked quite different for the low-capacity readers. Their immediate response, as indexed by the progressive fixation time measure, was to speed up on the irrelevant sentences. On the other hand, the low-capacity readers did not slow down their reading of the relevant sentences during the first-pass reading, but they invested extra visual attention to relevant sentences only later, as indexed by more look-backs made to the relevant sentences. These data are compatible with the strategic allocation of attention view of working memory[19]in that the high-capacity readers swiftly allocated extra attention to relevant information immediately when it was encountered, while low-capacity reader did so only with some time delay.
In conclusion, our studies of the perspective effects on on-line text comprehension have revealed that(1)a reading goal is capable of exerting an early effect on readers' eye fixation patterns,(2)eye movement recordings tap into the time course of processing,(3)time course analyses can reveal interesting individual differences, and(4)working memory capacity is linked to the efficiency to strategically allocate attention as well as to encode information to and retrieve it from the long-term memory.
Final Remarks
In this article, I hope to have been able to show that the eye-tracking technique is an excellent research tool to tap into the workings of the human mind during the comprehension of written texts. In the first section I showed that it provides the researcher with detailed analyses of the cognitive processes related to word identification. In the second section I have showed that it can also be used to study the mental processes needed to successfully comprehend longer texts as well as to study differences in individual reading strategies[22].
I feel our results are sufficiently promising to encourage further research on these topics. Studies on the compound word processing in Chinese would be particularly welcome for(at least)two reasons. First, to understand the basic reading and word recognition processes in Chinese we need to know how word meaning is accessed(or extracted)when the word consists of two or more characters. Second, compound word studies in Chinese render possible comparisons between alphabetic and non-alphabetic scripts. Such comparisons help us to estimate what processing features are universal and what are language-specific.
The use of the eye-tracking method to study on-line comprehension of long texts has so far been relatively limited. There are probably several reasons for this. One is that the research has not yet been able to clearly establish what the most fruitful eye movement measures would be to study the processing of extended discourse. Although it is probably true that no standards have been set yet, several potentially useful measures have recently been suggested[20]. Another possible reason for the scarcity of studies is that the analyses tend to be time-consuming and work-intensive. Although there is no denying this, I strongly feel that eye movement analyses are worth the effort, as they provide valuable data on the detailed time course of processing other on-line measures, such as sentence reading time measures, cannot provide. Moreover, the method provides useful measures to study differences in individual reading strategies. Adult readers appear to differ significantly in how much they reread a sentence before moving on to the next one and particularly how much they look back to previous sentences when reading long texts[22]. This kind of reading behavior is readily captured by the eye movement method.
References
1 Taft M. Recognition of affixed words and the word frequency effect. Memory & Cognition, 1979, 7: 263~272
2 Bertram R, HyÖnä J. The length of a complex word modifies the role of morphological structure: Evidence from eye movements when reading short and long Finnish compounds. Journal of Memory and Language, 2003, 48: 615~634
3 HyÖnä J, Bertram R, Pollatsek A. (in press). Are long compound words identified serially via their constituents? Evidence from an eye-movement contingent display change study. Memory & Cognition
4 HyÖnä J, Pollatsek A. Morphological processing of Finnish compound words in reading. In A. Kennedy, R. Radach, D. Heller & J. Pynte(Eds.), Reading as a perceptual process. Oxford: Elsevier, 2000. 65~87
5 HyÖnä J, Pollatsek A. The role of component morphemes on eye fixations when reading Finnish compound words. Journal of Experimental Psychology: Human Perception and Performance, 1998, 24: 1612~1627
6 Pollatsek A, HyÖnä J. (in press). The role of semantic transparency in the processing of Finnish compound words. Language and Cognitive Processes
7 Pollatsek A, HyÖnä J, Bertram R. The role of morphological constituents in reading Finnish compound words. Journal of Experimental Psychology: Human Perception and Performance, 2000, 26: 820~833
8 Andrews S, Miller B, Rayner K. Eye movements and morphological segmentation of compound words: There is a mouse in the mousetrap. European Journal of Cognitive Psychology. 2004, 16: 285~311
9 Inhoff A W, Radach R, Heller D. Complex compounds in German: Interword spaces facilitate segmentation but hinder assignment of meaning. Journal of Memory and Language, 2000, 42: 23~50
10 Inhoff A W, Starr M, Shindler K L. Is the processing of words during eye fixation in reading strictly serial? Perception & Psychophysics, 2000, 62: 1474~1484
11 Juhasz B J, Inhoff A W, Rayner K. (in press). The role of interword spaces in the processing of English compound words. Language and Cognitive Processes
12 Juhasz B J, Starr M, Inhoff A W, et al. The effects of morphology on the processing of compound words: Evidence from naming, lexical decisions, and eye fixations. British Journal of Psychology, 2003, 94: 223~244
13 Rayner K. Eye movements in reading and information processes: 20 years of research. Psychological Bulletin, 1998, 124: 372~422
14 Anderson R C, Pichert J W. Recall of previously unrecallable information following a shift in perspective. Journal of Verbal Learning and Verbal Behavior, 1978, 17: 1~12
15 Kaakinen J K, HyÖnä J, Keenan J M. Individual differences in perspective effects on text memory. Current Psychological Letters: Behaviour, Brain & Cognition, 2001, 5: 21~32
16 Kaakinen J K, HyÖnä J, Keenan J M. Perspective effects on on-line text processing. Discourse Processes, 2002, 33: 159~173
17 Kaakinen J K, HyÖnä J, Keenan J M. How prior knowledge, working memory capacity, and relevance of information affect eye-fixations in expository text. Journal of Experimental Psychology: Learning, Memory, and Cognition, 2003, 29: 447~457
18 Daneman M, Carpenter P A. Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 1980, 19: 450~466
19 Engle R W, Kane M J, Tuholski S W. Individual differences in WMC and what they tell us about controlled attention, general fluid intelligence, and functions of the prefrontal cortex. In A. Miyake and P. Shah(Eds.), Models of working memory: Mechanisms of active maintenance and executive control. Cambridge, UK: Cambridge University Press. 1999. 102~134
20 HyÖnä J, Lorch R F, Rinck M. Eye movement measures to study global text processing. In J. HyÖnä, R. Radach & H. Deubel(Eds.), The mind's eye: Cognitive and applied aspects of eye movement research. Amsterdam: Elsevier Science. 2003. 313~334
21 Ericsson K A,Kintsch W. Long-term working memory. Psychological Review, 1995, 102: 211~245
22 HyÖnä J, Lorch R F, Kaakinen J K. Individual differences in reading to summarize expository text: Evidence from eye fixation patterns. Journal of Educational Psychology, 2002, 94: 44~55
Key wordseye movements, word recognition, compound words, text comprehension, working memory capacity.
In this article I summarize two lines of eye movement research that we have conducted over the past several years in Finland. One line of research deals with how competent adult readers identify compound words during reading. The other line of research deals with how a specific reading goal(we call it the reading perspective)influences the way a long expository text is read and what is remembered of the text*. In both lines of research we have applied the eye movement technique by registering readers' eye movements when they read silently single sentences or longer texts for comprehension. As will become evident, eye-tracking has proved to be a highly valuable research tool to tap into the ongoing mental processes as they evolve over time when people read texts.
Identifying Compound Words During Reading
By definition, compound words are words that consist of 2 or more components. An English example could be headache, which consists of two components, head and ache. The word meaning is derived by putting together the meanings of the two components to yield something like 'pain in the head'. Finnish is an agglutinative language, in which word compounding is very productive. Novel compounds can readily be constructed out of existing words, particularly of nouns. Finnish compounds can include several components, as in lumipallosota, which means 'snowball fight' and consists of 3 components, lumi(snow), pallo(ball), and sota(fight). All Finnish compounds are concatenated in that spaces never appear between the components. This feature can also be found in other alphabetic languages, such as German, Dutch, or Swedish. English, on the other hand, is more irregular in the sense that compounds may appear in one of 3 forms: unspaced(e.g., snowball), spaced(e.g., traffic light), or hyphenated(e.g., eye-tracker). Chinese is also highly interesting in the sense that in Chinese most words are compounds in that they consist of more than one character. Finnish compounds are similar to Chinese ones: in neither language spaces are added to signal where the boundaries exist between the components.
One key question we have addressed is "To what extent are compounds identified holistically and to what extent via their components?" In order to answer this question we have either(1)manipulated the frequency of the whole word while matching for the frequency of the components(and other relevant features, such as length), or(2)manipulated the frequency of the components as separate words while matching for whole-word frequency. This general method was introduced by Taft to study the recognition of morphologically complex words[1]. The logic is as follows. If word frequency is found to have an influence on the identification process, it is assumed that the holistic route to compound word meaning gets activated during the course of processing. More specifically, the holistic route is implicated when a low-frequency compound word takes longer to process than a high-frequency compound word, given that the two types of words are equated for the frequency of the components. On the other hand, a componential route is implicated, for example, by a finding that a compound with a low-frequency first component takes longer to process than a compound with a high-frequency first component, given that they are equated for whole-word frequency. Finally, the presence of both constituent frequency and word frequency effects suggests a dual-route model, in which both access routes are operative.
In our studies[2~7]we have observed that the length of the compound word is a crucial factor in determining the extent to which a compound word is identified holistically or via its parts[2]. Until now we have only studied the processing of 2-component compounds, so everything that is said below concerns compounds of that sort(most of the compounds used in our studies have been noun-noun compounds, such as headache). There are some other eye-tracking studies on compound word processing conducted in English and in German[8~12].
Our results demonstrate that when a compound word has no more than 8 letters (as in sivuovi=side door), the word is processed holistically. This becomes apparent in that(1)gaze duration(i.e., the duration of all fixations landing on the word before exiting it to the right or left)is reliably influenced by word frequency, but(2)the frequency of the first component does not affect gaze duration. Word frequency exerts an early effect on the processing, as indexed by the effect being observed already in the first fixation duration. All this is taken to suggest that shorter compounds are identified as single wholes and that the access to the components is not an integral part of the process.
The story is quite different for longer compounds(i.e., compounds that contain 12 letters or more). For long compounds, an early effect of first constituent frequency is observed. The duration of first fixation on long compounds was found to be longer when the first constituent was infrequent than when it was frequent. It may be noted that the first fixation typically also landed on the first constituent. A highly reliable effect of first constituent frequency was also observed in gaze duration. When the frequency of the second constituent was manipulated, its effect was not observed in the first fixation duration, but only from the second fixation onwards(it is worth noting that typically the second fixation on the word landed on the second constituent). The duration of second fixation was longer when the second constituent was infrequent than when it was frequent. An effect of second constituent frequency was also observed in gaze duration. The pattern of these results is compatible with the view that long compounds are identified via their components by first accessing the first constituent followed by the access to the second constituent. The time course of effects nicely supports this view, as we obtain an early effect of first constituent frequency and a relatively later effect of second constituent frequency.
However, the story is not quite simple as that. For long compounds we have also obtained an effect of whole-word frequency that is statistically reliable from the second fixation onwards. There is even a marginal tendency for a word frequency effect in the first fixation duration although the effect size is only 4~5ms. A reliable effect of word frequency suggests that also the holistic route is in operation in the processing of long compounds. However, it does not come into play as early as the decomposition route. In other words, the decomposition route is initially prioritized over the holistic route in the processing of long compounds, but the two routes appear to run in parallel(i.e., the decomposition route does not finish before the holistic route becomes in operation).
Combined with the results obtained for shorter compounds, the picture that emerges is one where, after fixating the compound, the reader starts processing whatever high-acuity information is available from the word. If all the letters of a compound fall onto the fovea when it is fixated, as is typically the case with shorter compounds, the holistic route gets activated from the first moment onwards and the decomposition route does not become activate at all(or only very minimally). On the other hand, when some of the letters fall outside the foveal reach when the compound is initially fixated, as is the case with longer compounds, the decomposition route initially wins over the holistic route. Thus, the identification process is initiated by first attempting an access to the first constituent, followed by an access to the second constituent and to the word as a whole. As mentioned above, the holistic route is assumed to run in parallel with the decomposition route, although it completes its job later than the decomposition route(i.e., it needs to accumulate information also about the second component).
As is evident from above, the identification of long compounds appears to be a relatively serial process in the sense that the processing proceeds from the access of the first component to the access of the second component. As a further test of this seriality assumption, we have recently conducted an eye-movement contingent display change experiment[13]. This technique allowed us to manipulate what is initially available in the parafoveal vision of the second constituent when the reader is fixating on the first constituent. In the display change condition, we replaced the last letters of the second constituent with visually similar letters(the first two letters were kept intact), which made the second constituent to appear as a non-word (i.e., pioneeriryhmä=pioneer group initially appeared as pioneeriryknÖ). During the saccadic eye movement made from the first to the second constituent, the last letters were changed back to what they should have been in the first place. Due to saccadic suppression, the change itself is left unnoticed, namely those trials where the system was not able to make the display change during the saccade were excluded from the analyses. The display change condition was compared to the no-change condition, where the second constituent was available throughout the target word viewing.
The results showed a very noticeable parafoveal preview effect. Gaze duration on the compound was 101ms longer when the second constituent initially appeared as a non-word than when it was kept intact. This preview effect is significantly bigger than what is typically observed for parafoveal previews that are separated from the foveated word by a space[3,13]. Despite this big preview effect, it is quite remarkable that the effect did not show up in the fixation time spent on the first constituent, but the effect only surfaced after the second constituent was fixated. We take this as strong evidence in support of the seriality view, according to which the initial stages of processing long compound words are devoted to the attempt of accessing the first constituent, and it is not until the second constituent is fixated that its lexical access is attempted. Note however that readers do process some features of the second constituent while fixating on the first constituent; otherwise we would not have obtained the big preview effect. But importantly, whatever information is picked up of the second constituent parafoveally it does not affect the fixation time on the first constituent. It is exactly this aspect of the data that supports the seriality view. We currently think that the information picked up parafoveally of the second constituent is orthographic information(i.e., information about the letter identities)and not lexical information[9,10].
To sum up the first section, I want to point out that our research on compound word identification has not only taught us a good deal about how morphologically complex words are recognized, but it has also proved to be a good testing ground to study how the eyes and attention is guided in text during reading. It has also demonstrated the fruitfulness of the eye-tracking technique, which has allowed us to tap into the exact time course of compound word processing. Had we used a single processing measure, such as lexical decision time, we could not have been able to observe many of the effects discussed above(the differentiation between early versus late effects).
As far as I know, all the eye-tracking studies conducted on compound word processing have so far dealt with alphabetic languages, such as Finnish, German, and English. It would be highly interesting to extend the results to non-alphabetic scripts like Chinese. Chinese is also interesting in the present context in that most Chinese words are compound words in the sense that they consist of two or more characters. In order to get a better understanding of the basic reading processes in Chinese, I think it would be important to find out about how the meaning of Chinese words is derived out of its component characters and their meanings.
Effects of a specific reading goal on reading long expository texts
In the second part of my article I review our eye movement research on text comprehension processes. I focus particularly on the effects of a specific reading goal on on-line text comprehension as it is reflected in the readers' eye movement records. The starting point to our research was the classic study of Anderson and Pichert who investigated what consequences a perspective adopted by a reader may have on what he or she remembers of the text[14]. We have also tested the memory for text after reading, but our primary interest has been on the effects of a reading perspective on the ongoing comprehension processes.
In our studies[15~17] we have given the readers a specific reading goal, a reading perspective, from which to approach the text. For example, before reading a text on familiar diseases(flu, diarrhea, AIDS, and chicken pox)the reader was given the following reading perspective (or goal): "Imagine that you are going to give a health education class for elementary school pupils. You are supposed to tell them about flu: for example, what causes it, how you can treat it, and how to prevent from getting the disease. Now read the text in order to be able to do the job." This kind of reading perspective renders some text information relevant and other text information irrelevant. Specifically, information given in the text about flu is perspective-relevant, while information about the other diseases is perspective-irrelevant. We selected 10 sentences from the text that provided information about the flu that was considered, according to a pre-test, to be typically associated with this disease. These sentences were matched in length and in average word frequency to 10 irrelevant sentences that conveyed information about diarrhea.(The reading perspective was counterbalanced across participants so that for half of them diarrhea was the target disease, in which case the flu sentences comprised the perspective-irrelevant sentence set.)
We were also interested in what role working memory capacity may play in on-line text comprehension. Thus, we administered the reading span text[18]to our participants. On the basis of the test, we formed a group of high-capacity, a group of medium-capacity, and a group of low-capacity readers. To simplify the story a bit, below I only summarize the data for hig-and low-capacity readers. According to one influential theory of working memory[19], working memory capacity is related to the efficiency with which attentional resources are devoted to task-relevant information and away from task-irrelevant information. If this is indeed the case, high-capacity readers should be more efficient than low-capacity readers in allocating their visual attention to perspective-relevant information. In other words, we reasoned that high-capacity readers should demonstrate a greater effect of reading perspective than low-capacity readers.
Three types of sentence-level eye fixation measures were computed for the target sentences (10 perspective-relevant, 10 perspective-irrelevant sentences):(1)progressive first-pass fixation time=the summed duration of fixations landing on an unread part of a sentence before exiting it;(2)first-pass rereading time=the summed duration of fixations returning to already read parts of a sentence before exiting the sentence; and(3)look-back fixation time=the summed duration of fixations launched from subsequent text back to the target sentence. By adding up the first two measures(progressive first-pass fixations and first-pass rereading fixations)we obtain the first-pass fixation time, which is a measure of the overall time readers spent looking at the target sentence before moving to the next. The three measures mentioned above were designed to tap into the time course of processing, the progressive first-pass fixation time indexing the most immediate effect and look-back fixation time a delayed effect[20].
In our first eye movement study[16]on this topic we observed an effect of reading perspective both in the first-pass fixation time and in look-back fixation time. Readers spent significantly longer time reading the perspective-relevant sentences than the perspective-irrelevant sentences.(It should be mentioned that in all our experiments we have also found a better memory for perspective-relevant than perspective-irrelevant information.)Moreover, we observed differences in the timing of the perspective effect among our reader groups. They all demonstrated a reliable perspective effect in processing, but in different points in time. The high-capacity readers demonstrated the effect already in the first-pass fixation time, whereas the low-capacity readers showed it only in the look-back fixation time(i.e., with some delay). These data are consistent with the strategic allocation of attention view of working memory[19], according to which high-capacity readers are better able to efficiently allocate attention to task-relevant information and away from task-irrelevant information.
In the follow-up study[17], we wanted to compare the effect of reading perspective for a text on familiar contents to that observed for a text on unfamiliar contents(in the 2002 study, we used a text on unfamiliar contents; the text described four remote countries). The text with familiar contents described four diseases(see above)the readers had ample prior knowledge about, while the text with unfamiliar contents described four largely unknown diseases(trigeminusneuralgy, typhus, cystic fibrosis, and scleroderma). Theoretically, the comparison between texts of familiar versus unfamiliar contents is interesting as it makes contact with the notion of long-term working memory put forth by Ericsson and Kintsch[21]. According to this conception, working memory capacity is related to the ability to efficiently use prior knowledge during encoding. If so, high-capacity readers should be able to make better use of their prior knowledge than low-capacity readers in encoding to memory perspective-relevant information of the familiar disease text.
This was in fact what we found. In the familiar diseases text, the high-capacity readers took no longer time to read the perspective relevant than perspective irrelevant sentences. Yet, after reading they showed a clearly better memory for relevant than irrelevant information. We argue that this is because they have fast access to prior knowledge that can be readily brought to bear to help encode the relevant information to memory during reading. That is why they did not need to fixate any longer on relevant than irrelevant target sentences. The low-capacity readers, on the other hand, did need to do so. Their first-pass and look-back fixation times were significantly longer for perspective relevant than perspective irrelevant sentences. All in all, the pattern of our results is consistent with the notion of long-term working memory proposed by Ericsson and Kintsch[21].
The unfamiliar diseases text replicated our previous results[16] in showing that the high-capacity readers showed the perspective effect in an earlier phase of processing than the low-capacity readers. In order to get a more detailed picture of the allocation of visual attention to perspective relevant versus irrelevant text segments, we ran a control experiment, where no specific reading perspective was given. The participants were simply told to read the text in order to be able to write a summary of the text contents after reading. Comparisons to the control group made it possible to tease apart the nature of the perspective effect: to what extent the perspective effect is due to slowing down on relevant information and to what extent it is due to speeding up on irrelevant sentences. These comparisons were made separately for the high-capacity and low-capacity readers. In order to be able to do so, we tested in the control experiment a group of high-and low-capacity readers who had not participated in the experiment proper.
The results showed that, in comparison to their controls, the high-capacity readers invested extra processing time to relevant information already during the initial encounter of that information. In other words, their progressive fixation times were significantly longer on the relevant sentences than those of the control group who were not given a specific reading perspective. On the other hand, they did not initially slow down their reading of the irrelevant sentences(again compared to the progressive fixation times of their controls). The pattern of results looked quite different for the low-capacity readers. Their immediate response, as indexed by the progressive fixation time measure, was to speed up on the irrelevant sentences. On the other hand, the low-capacity readers did not slow down their reading of the relevant sentences during the first-pass reading, but they invested extra visual attention to relevant sentences only later, as indexed by more look-backs made to the relevant sentences. These data are compatible with the strategic allocation of attention view of working memory[19]in that the high-capacity readers swiftly allocated extra attention to relevant information immediately when it was encountered, while low-capacity reader did so only with some time delay.
In conclusion, our studies of the perspective effects on on-line text comprehension have revealed that(1)a reading goal is capable of exerting an early effect on readers' eye fixation patterns,(2)eye movement recordings tap into the time course of processing,(3)time course analyses can reveal interesting individual differences, and(4)working memory capacity is linked to the efficiency to strategically allocate attention as well as to encode information to and retrieve it from the long-term memory.
Final Remarks
In this article, I hope to have been able to show that the eye-tracking technique is an excellent research tool to tap into the workings of the human mind during the comprehension of written texts. In the first section I showed that it provides the researcher with detailed analyses of the cognitive processes related to word identification. In the second section I have showed that it can also be used to study the mental processes needed to successfully comprehend longer texts as well as to study differences in individual reading strategies[22].
I feel our results are sufficiently promising to encourage further research on these topics. Studies on the compound word processing in Chinese would be particularly welcome for(at least)two reasons. First, to understand the basic reading and word recognition processes in Chinese we need to know how word meaning is accessed(or extracted)when the word consists of two or more characters. Second, compound word studies in Chinese render possible comparisons between alphabetic and non-alphabetic scripts. Such comparisons help us to estimate what processing features are universal and what are language-specific.
The use of the eye-tracking method to study on-line comprehension of long texts has so far been relatively limited. There are probably several reasons for this. One is that the research has not yet been able to clearly establish what the most fruitful eye movement measures would be to study the processing of extended discourse. Although it is probably true that no standards have been set yet, several potentially useful measures have recently been suggested[20]. Another possible reason for the scarcity of studies is that the analyses tend to be time-consuming and work-intensive. Although there is no denying this, I strongly feel that eye movement analyses are worth the effort, as they provide valuable data on the detailed time course of processing other on-line measures, such as sentence reading time measures, cannot provide. Moreover, the method provides useful measures to study differences in individual reading strategies. Adult readers appear to differ significantly in how much they reread a sentence before moving on to the next one and particularly how much they look back to previous sentences when reading long texts[22]. This kind of reading behavior is readily captured by the eye movement method.
References
1 Taft M. Recognition of affixed words and the word frequency effect. Memory & Cognition, 1979, 7: 263~272
2 Bertram R, HyÖnä J. The length of a complex word modifies the role of morphological structure: Evidence from eye movements when reading short and long Finnish compounds. Journal of Memory and Language, 2003, 48: 615~634
3 HyÖnä J, Bertram R, Pollatsek A. (in press). Are long compound words identified serially via their constituents? Evidence from an eye-movement contingent display change study. Memory & Cognition
4 HyÖnä J, Pollatsek A. Morphological processing of Finnish compound words in reading. In A. Kennedy, R. Radach, D. Heller & J. Pynte(Eds.), Reading as a perceptual process. Oxford: Elsevier, 2000. 65~87
5 HyÖnä J, Pollatsek A. The role of component morphemes on eye fixations when reading Finnish compound words. Journal of Experimental Psychology: Human Perception and Performance, 1998, 24: 1612~1627
6 Pollatsek A, HyÖnä J. (in press). The role of semantic transparency in the processing of Finnish compound words. Language and Cognitive Processes
7 Pollatsek A, HyÖnä J, Bertram R. The role of morphological constituents in reading Finnish compound words. Journal of Experimental Psychology: Human Perception and Performance, 2000, 26: 820~833
8 Andrews S, Miller B, Rayner K. Eye movements and morphological segmentation of compound words: There is a mouse in the mousetrap. European Journal of Cognitive Psychology. 2004, 16: 285~311
9 Inhoff A W, Radach R, Heller D. Complex compounds in German: Interword spaces facilitate segmentation but hinder assignment of meaning. Journal of Memory and Language, 2000, 42: 23~50
10 Inhoff A W, Starr M, Shindler K L. Is the processing of words during eye fixation in reading strictly serial? Perception & Psychophysics, 2000, 62: 1474~1484
11 Juhasz B J, Inhoff A W, Rayner K. (in press). The role of interword spaces in the processing of English compound words. Language and Cognitive Processes
12 Juhasz B J, Starr M, Inhoff A W, et al. The effects of morphology on the processing of compound words: Evidence from naming, lexical decisions, and eye fixations. British Journal of Psychology, 2003, 94: 223~244
13 Rayner K. Eye movements in reading and information processes: 20 years of research. Psychological Bulletin, 1998, 124: 372~422
14 Anderson R C, Pichert J W. Recall of previously unrecallable information following a shift in perspective. Journal of Verbal Learning and Verbal Behavior, 1978, 17: 1~12
15 Kaakinen J K, HyÖnä J, Keenan J M. Individual differences in perspective effects on text memory. Current Psychological Letters: Behaviour, Brain & Cognition, 2001, 5: 21~32
16 Kaakinen J K, HyÖnä J, Keenan J M. Perspective effects on on-line text processing. Discourse Processes, 2002, 33: 159~173
17 Kaakinen J K, HyÖnä J, Keenan J M. How prior knowledge, working memory capacity, and relevance of information affect eye-fixations in expository text. Journal of Experimental Psychology: Learning, Memory, and Cognition, 2003, 29: 447~457
18 Daneman M, Carpenter P A. Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 1980, 19: 450~466
19 Engle R W, Kane M J, Tuholski S W. Individual differences in WMC and what they tell us about controlled attention, general fluid intelligence, and functions of the prefrontal cortex. In A. Miyake and P. Shah(Eds.), Models of working memory: Mechanisms of active maintenance and executive control. Cambridge, UK: Cambridge University Press. 1999. 102~134
20 HyÖnä J, Lorch R F, Rinck M. Eye movement measures to study global text processing. In J. HyÖnä, R. Radach & H. Deubel(Eds.), The mind's eye: Cognitive and applied aspects of eye movement research. Amsterdam: Elsevier Science. 2003. 313~334
21 Ericsson K A,Kintsch W. Long-term working memory. Psychological Review, 1995, 102: 211~245
22 HyÖnä J, Lorch R F, Kaakinen J K. Individual differences in reading to summarize expository text: Evidence from eye fixation patterns. Journal of Educational Psychology, 2002, 94: 44~55