TOEFL 2000 Reading
Mary K. Enright
MS - 17
TOEFL 2000 Reading Framework:
A Working Paper
Mary K. Enright
Educational Testing Service
Princeton, New Jersey
Educational Testing Service is an Equal Opportunity/Affirmative Action Employer.
Copyright 2000 by Educational Testing Service. All rights reserved.
No part of this report may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopy, recording, or any information storage
and retrieval system, without permission in writing from the publisher. Violators will
be prosecuted in accordance with both U.S. and international copyright laws.
EDUCATIONAL TESTING SERVICE, ETS, the ETS logo, GRE, NTE, TOEFL, the TOEFL
logo, and TSE are registered trademarks of Educational Testing Service. The modernized
ETS logo is a trademark of Educational Testing Service.
Grolier is a registered trademark of Grolier Interactive, Inc.
To obtain more information about TOEFL programs and services, use one of the following:
Web site: http://www.toefl.org
The TOEFL Monograph Series features commissioned papers and reports for TOEFL 2000
and other TOEFL test development efforts. As part of the foundation for the TOEFL 2000
project, a number of papers and reports were commissioned from experts within the fields of
measurement and language teaching and testing. The resulting critical reviews and expert opinions
have helped to inform TOEFL program development efforts with respect to test construct, test
user needs, and test delivery. Opinions expressed in these papers are those of the authors and do
not necessarily reflect the views or intentions of the TOEFL program.
These monographs are also of general scholarly interest, and the TOEFL program is pleased to
make them available to colleagues in the fields of language teaching and testing and international
student admissions in higher education.
The TOEFL 2000 project is a broad effort under which language testing at Educational Testing
) will evolve into the 21st century. As a first step the TOEFL program recently
revised the Test of Spoken English (TSE
) and introduced a computer-based version of the
TOEFL test. The revised TSE test, introduced in July 1995, is based on an underlying construct of
communicative language ability and represents a process approach to test validation. The
computer-based TOEFL test, introduced in 1998, takes advantage of the new forms of
assessments and improved services made possible by computer-based testing while also moving
the program toward its longer-range goals, which include
• the development of a conceptual framework that takes into account models of
• a research agenda that informs and supports this emerging framework
• a better understanding of the kinds of information test users need and want from the
• a better understanding of the technological capabilities for delivery of TOEFL tests into
the next century
Monographs 16 through 20 are the working papers that lay out the TOEFL 2000 conceptual
frameworks with their accompanying research agendas. The initial framework document,
Monograph 16, describes the process by which the project will move from identifying the test
domain to building an empirically based interpretation of test scores. The subsequent framework
documents, Monographs 17-20, extend the conceptual frameworks to the domains of reading,
writing, listening, and speaking (both as independent and interdependent domains). As such, the
current frameworks do not yet represent a final test model. The final test design will be refined
through an iterative process of prototyping and research as the TOEFL 2000 project proceeds.
As TOEFL 2000 projects are completed, monographs and research reports will continue to be
released and public review of project work invited.
TOEFL Program Office
Educational Testing Service
The TOEFL 2000 framework monograph (Jamieson, Jones, Kirsch, Mosenthal, & Taylor,
1999) identifies a test domain and lays out a process for the design of a new TOEFL test based on
communicative language abilities. This monograph on the assessment of reading comprehension
addresses the proposed TOEFL 2000 framework described in Jamieson et al. (1999) and defines
how it can be realized and implemented in a test of reading comprehension. The reading
framework described in this document was developed by the authors, consisting of internal ETS
staff and external reading experts, who have worked together over the past two years.
This monograph documents how three broad perspectives were considered in defining the
construct of reading comprehension for assessment purposes: a processing perspective, a task
perspective, and a reader purpose perspective. The reader purpose perspective is recommended
to guide the new test design for a number of reasons. One perceived advantage of this approach is
that it is readily interpretable. It will be easier for test-score users, teachers, and examinees to
understand how the construct is being defined. At the same time, the reader purpose perspective
is seen to be compatible with both the processing perspective and the task perspective.
Four purposes for reading in the academic context are identified: reading to find information,
reading for basic comprehension, reading to learn, and reading to integrate information across
multiple texts. These four reading purposes are seen to form a natural hierarchy that can serve as
a basis for describing a continuum of reading proficiency. The first two purposes are addressed in
the current TOEFL reading test format. The third and fourth purposes, reading to learn and
reading to integrate information across multiple texts, would expand the construct being measured.
Some tasks that might be used to assess reading for different purposes are described.
Finally, technological issues specific to the delivery of the reading test are described and a
detailed research agenda related to the reading construct described in this document is provided.
Key phrases: TOEFL 2000 reading, academic reading purposes, new test design, reading to
learn, reading multiple texts
The authors gratefully acknowledge the support and suggestions of the TOEFL Committee of
Examiners, the members of the TOEFL 2000 Committees working on the speaking, listening, and
writing frameworks, and the ETS staff for their helpful comments and suggestions.
Table of Contents
1. Introduction .....................................................................................................................................................1
2. Conceptualizing Reading Proficiency.....................................................................................................2
The Processing Perspective...................................................................................2
The Task Perspective ...........................................................................................3
The Reader Purpose Perspective...........................................................................4
Using Reader Purpose to Guide Test Design ..........................................................4
Reading to Find Information.............................................................................5
Reading for Basic Comprehension ...................................................................5
Reading to Learn............................................................................................6
Reading to Integrate Information......................................................................6
A Difficulty Continuum ...................................................................................6
Differences in L1 and L2 Reading .........................................................................7
Transfer of L1 Reading Skills and Strategies.....................................................8
Probable Facilitation Stemming from L1-L2 Structural Similarity ........................9
Cross-linguistic Interactions during L2 Reading...............................................10
Processing Constraints Resulting from Limited Linguistic Knowledge...............11
Implications for Reading Assessment among L2 Learners...............................12
3. Reading Framework for the TOEFL 2000 Test...............................................................................14
Identifying the Test Domain.................................................................................14
Organizing the Test Domain ................................................................................14
Identifying Task Characteristics...........................................................................14
Grammatical/Discourse Features .............................................................18
Test Rubric ..................................................................................................28
Tasks to Assess the Four Types of Reading .............................................31
Types of Response Formats ....................................................................37
Linguistic Variables and Task Difficulty ...................................................38
4. Technological Considerations..................................................................................................................40
The Role of Technology in Reading Comprehension..............................................40
The Reading Comprehension Interface.................................................................41
5. Research Agenda .....................................................................................................43
Construct Identification .......................................................................................45
Pilot Testing .......................................................................................................47
6. A Better Reading Test..............................................................................................49
Appendix A Linguistic and Processing Variables ...................................................63
Appendix B Text and Task Variables ...................................................................67
Appendix C Other Factors that May Contribute to Test Variation...........................71
Appendix D L1 and L2 Differences......................................................................72
Appendix E Register Variations ...........................................................................74
Table 1 Types of Texts and Tasks for the TOEFL 2000 Reading Test..........................30
This document presents the framework for the TOEFL 2000 test of reading comprehension.
Part 2 discusses the construct of reading and explains how a “reader purpose” perspective will be
used to guide the design of the TOEFL 2000 reading test. Four purposes for reading are
highlighted: reading to find information, reading for basic comprehension, reading to learn, and
reading to integrate information across multiple texts.
Part 3 reviews the domain for the test proposed in the TOEFL 2000 framework document
(Jamieson et al., 1999), describes the TOEFL 2000 organizational scheme, and discusses the
proposed task characteristics in terms of the reading test. This section also describes tasks that
might be used to assess reading for different purposes.
Part 4 of this document considers technological issues involved in assessing reading
comprehension and makes some recommendations for interface design. Part 5 presents a detailed
research agenda needed to support the development of the test, and the final section, Part 6,
discusses the ways in which the new TOEFL 2000 reading test will improve on the earlier versions
of the test.
2. Conceptualizing Reading Proficiency
One of the major challenges for individual differences research remains the
discovery of a principled set of processing explanations for individual
differences, as opposed to a list of all processes that occur during reading.
Reading comprehension has been viewed from a number of perspectives. In thinking about
the construct, or constructs, of reading comprehension for the purposes of the TOEFL 2000
project, we considered three perspectives:
1. The processing perspective,
2. The task perspective, and
3. The reader purpose perspective.
Each of these perspectives is briefly discussed below. Although we settled on the reader
purpose perspective as the guiding principle for test design, we believe that the processing
perspective and the task perspective can also both be understood from, and inform, this broad
The Processing Perspective
In a recent review of reading comprehension research, Perfetti (1997) argued that research on
individual differences among readers is the key to understanding the nature of reading abilities.
That is, to understand reading we need to know what factors contribute consistently to differences
between better and weaker readers. Perfetti himself suggests that major sources of individual
proficiency differences include differences in processing efficiencies such as speed and
automaticity of word recognition, thoroughness of word representation knowledge, processing
efficiencies in working memory, fluency in syntactic parsing and proposition integration as part of
building text comprehension, and the development of an accurate and reasonably complete text
model of comprehension (see also Carpenter, Mikaye, & Just, 1994; Perfetti, 1994).
Recent efforts to explore the “simple view of reading” (Gough, Hoover, & Peterson, 1996;
Chen & Vellutino, 1997) have also argued that a good part of reading abilities can be related to a
combination of word recognition abilities and comprehension abilities. This view would
implicate word recognition fluency, accurate word representations, processing efficiency, text
model building, inferencing, and strategic processing and monitoring. Similar implications can be
drawn from the recent overview by Carver (1997), who brings together a set of research studies to
develop a model of components of reading. This model also argues that reading abilities are
fundamentally composed of fluency, word recognition accuracy, and rate of processing (which
might combine processing efficiency and reading rate).
In second language contexts, both Koda (1996, 1997) and Geva, Wade-Wooley, & Shany
(1997) make strong arguments for the importance of fluent word recognition, processing
efficiency, and reading rate in second-language (L2) reading.
In language testing contexts, it would seem fairly apparent that vocabulary is a key co-variate
with reading, as is, to a lesser extent, some measure of grammar knowledge. The high correlation
of listening test scores with reading test scores has been attributed to the importance of general
comprehension abilities associated with (a) generating a text model of comprehension,
(b) forming an appropriate situation model relating reader knowledge with text information,
(c) inferencing of certain types, and (d) monitoring comprehension strategically (see, for example,
Sticht & James, 1984; and Appendix A).
Thus, from a processing perspective, a small set of linguistic and processing variables can be
said to drive the construct of reading (see Appendix A for a review of research related to
linguistic and processing variables).
The Task Perspective
On the assumption that reading can be “defined” in terms of the tasks that readers are able to
accomplish, or to accomplish well if they are good readers, it is possible to develop a task-based
explanation of skills that readers possess, anchored to reading ability. While it may be possible to
develop such explanations using authentic tasks that are carried out in the real world of reading, it
is more useful for testing reading comprehension to develop a set of text and task variables that
can account for the variance in performance that occurs on reading test questions. Text and task
variables such as the frequency and usage of particular words involved in the task, the complexity
of syntax, the amount of text that must be processed, and the amount of time allowed for
completing a task may account for much of the variance in difficulty and task performance on test
questions. Research has already indicated the importance of task variables such as the occurrence
or absence of distracting information in the text, the degree to which the correct answer matches
the wording of the information in the text itself, and the concreteness of the information requested
(Kirsch & Mosenthal, 1990). To the extent that the identified task variables account for
performance differences on test questions, they may be said to provide a task-based framework
for interpreting the construct of reading as it is instantiated in the test design.
At one level, a degree of research interpretability is lost by using such an approach since
much of the research exploring the construct of reading is processing-based. Thus, one cannot
directly relate these types of task variables to processing notions such as efficiency of processing,
vocabulary knowledge, word recognition, text model formation, etc. At the same time, it is
possible to make these connections with a certain amount of reasonable inferencing so that task
variables can be argued to “account for” the processing and linguistic variables used in the
reading research literature.
At another level, the use of text- and task-based variables to “define” the construct of reading
provides a strong advantage: These variables can be used to create an interpretable description of
factors which cause difficulty for readers. Such descriptions can also be used to interpret more
directly some of the things that readers need to do to become better readers and do so in a way
that is not dependent on cognitive processing theories (see Appendix B for a review of text and
The Reader Purpose Perspective
A third way to conceptualize the construct of reading is to examine the different purposes for
which people engage in the process(es) of reading. One advantage of this approach is that the
defining notions are fully interpretable as concepts associated with reading comprehension. Thus,
“reading for the basic idea” or “reading to learn” are concepts that certainly can be said to belong
in a test of academic language abilities. We believe that it is possible to use this reader-purpose
perspective as a framework for the TOEFL 2000 reading test and that both the reading-process
and task-based views of reading are compatible with it.
Carver (1997) argues that a theory of reading typically needs to account for at least two types
of reading: what he calls “rauding” (or basic comprehension) and “reading to learn.” It seems to
us that both of these types of reading reflect important academic reading purposes that should be
included in our TOEFL 2000 test framework. The purpose for the “rauding” type of reading is
general comprehension or comprehension of the major points in a text. With the “reading to learn”
type of reading, the purpose is to construct an organized representation of the text that includes
major points and supporting details. A third type of reading, “search reading,” is also considered
minimally by Carver, in part as either skimming or scanning, and in much greater detail in studies by
Guthrie (1988; Guthrie & Kirsch, 1987; Guthrie & Mosenthal, 1987). The purpose for this type of
reading is to find discrete pieces of information by skimming and scanning a text or a non-prose
document such as a table. We believe this reading purpose should also be considered in planning
for the TOEFL 2000 test, since many types of reading practiced by students in academic contexts
involve search reading processes, or what we will call “reading to find information.”
One other type of reading that is not accounted for above is the reading of multiple texts.
“Reading of multiple texts” represents recent work by Perfetti (1997) and Goldman (1997) on
student efforts to generate “intertext models of comprehension.” This type of reading, sometimes
called a “documents model” of reading comprehension, has as its purpose the integration of
information across multiple texts. Perfetti (1997) has noted that the integration of information
across texts depends on the same comprehension abilities involved in the comprehension of single
texts, although the purpose is more complex. Interactive reading skills appear to be expected in
university classes across a range of disciplines and may be an important aspect of advanced
academic reading. We propose, then, that “reading to integrate information across multiple texts”
be included in the TOEFL 2000 construct definition as a fourth purpose for reading.
Using Reader Purpose to Guide Test Design
For the present purpose – that is, providing a foundation for a test of academic reading
abilities in English as a second language – we recommend that the guiding principle for test
design be a single broad construct that includes the four academic reading purposes discussed
1. Reading to find information (or “search reading”),
2. Reading for basic comprehension,
3. Reading to learn, and
4. Reading to integrate information across multiple texts.
Each type of reading, or “purpose for reading,” can be seen as representing a variation on one
basic reading construct called “reading comprehension.” Each requires a combination of word
recognition/processing efficiency and comprehension abilities, and can therefore also be related to
the skills-processing perspective and the task perspective.
We believe that this purpose-driven framework will make it possible to explain the principles
driving test design and test development to the general public. At the same time, it is possible to
overlay frameworks that are driven by processing views of reading and task-based views of
Below we briefly explain how a purpose-driven framework might work for each of the four
purposes for reading. The first two types of reader purposes, reading to find information and
reading for basic comprehension, are covered well in the current TOEFL reading test. The second
two, reading to learn and reading to integrate information across multiple texts, are not currently
tested and would expand the construct for the TOEFL 2000 test.
Reading to Find Information. One of the most basic purposes for reading is to locate and
comprehend discrete pieces of information. People search text for information in order to find
answers to questions that have been posed, to verify and repair any miscomprehension, and to find
the most relevant parts of a text for information purposes.
From a skills processing perspective, being proficient at finding discrete information typically
involves rapid, automatic identification of words, working memory efficiencies, and fluent
reading rates. Scanning text is done at a relatively rapid rate. From a task perspective, item types
in this category could include searching for and matching discrete pieces of information or
searching a longer text for specific sections.
Reading for Basic Comprehension. We are assuming that in general the person who has the
ability to read for basic comprehension also has the ability to find information in a text (Urquhart
& Weir, 1998). This comprehension purpose additionally requires a reader to understand the
main ideas or the main points of the text, or to form some understanding of the main theme of the
text, but does not necessarily require an integrated understanding of how the supporting ideas and
factual details of the text form a coherent whole. Reading for basic comprehension involves
understanding a subset of individual ideas, primarily those tied to the thematic content.
Comprehension of information that is not central to the main idea is unlikely to be required for
this reading purpose. In particular, detailed information in the text does not need to be integrated
conceptually at this level of understanding beyond the comprehension of main ideas.
From a skills-processing perspective, basic comprehension requires some ability to construct a
text model representation of what is read and also the ability to form a relevant situation model.
The ability to comprehend the major ideas from a text is likely to require cycling through and
integrating a range of information from various points in the text, which in turn requires a
reasonably efficient reading rate. Some tasks based on simpler texts, however, may only require
readers to identify a main idea statement (these are more like “locating” tasks). Types of task that
might test basic comprehension include distinguishing main ideas from minor ideas or inferring
the main topic. If longer passages are included in the test, it might be more appropriate to ask
examinees to identify a number of important ideas. Questions testing basic comprehension of
individual, discrete, and supporting ideas in the passage would also be included here.
Reading to Learn. We assume that reading to learn incorporates the ability to find
information and to develop a basic comprehension of the text. However, reading to learn also
requires the reader to integrate and connect the detailed information provided by the author into a
coherent whole. This sort of integration requires an understanding of cause-and-effect
relationships, comparisons and contrasts, classification relationships, and persuasive intent but
this type of reading can be done at a slower rate.
From a skills-processing perspective, reading to learn requires that a reader form linkages
between a more elaborated model of text construction and frames (such as cause/effect,
compare/contrast) to organize conceptual information and to understand the author’s rhetorical
intent. Conceptual knowledge that helps the reader integrate information in a text might include
information derived from the text and/or from background knowledge. As such, it might
represent an efficient alignment of the text model and the situation model. Tasks that require
reading to learn might require the reader to cycle through a range of information, integrate that
information, and at points, perhaps, evolve an appropriate rhetorical framework for interpretation
(e.g., comparison/contrast, cause/effect).
Reading to Integrate Information. The ability to integrate information from multiple sources
implicates all the reading purposes discussed above. This reading purpose requires a reader to
integrate information from more than one source. Such tasks require a reader to work across two
or more texts and generate an organizing frame that is not explicitly stated. Texts may include
diagrams, charts, graphs, illustrations, and prose.
From a skills-processing perspective, an intertext model of comprehension is necessary to
account for this type of advanced academic reading. Theories of learning, concept representation,
and long-term memory need to be considered. This type of task can never consist of simply
locating information. This reading purpose would require multiple cycles of integrating
information and would require examinees to generate a conceptual frame.
A Difficulty Continuum. To some extent the four reader purposes form a kind of difficulty
continuum. To be sure, easy tasks could be designed for reading to learn or reading to integrate
information, and difficult tasks asking examinees to find discrete information or read for general
comprehension could be designed by manipulating task and linguistic/syntactic variables. Still,
we are more likely to find reading to learn and reading to integrate information across texts
associated with more challenging academic tasks and to require more sophisticated processing
abilities than reading to find discrete information or reading for general comprehension. The
implication is that, as the reading purpose changes from reading to find information to reading for
basic comprehension to reading to learn and reading to integrate information across texts, more
reading is required and more efficient strategies are necessary. We believe, therefore, that reader
purpose itself may be one of the variables that can contribute to task difficulty when combined
with appropriate texts and tasks. It will be important to explore this as part of the research
agenda. As with any test situation, a number of other factors that may or may not be aspects of
the construct may be involved (see Appendix C). Below we discuss differences in L1 and L2
Differences in L1 and L2 Reading
It is important to keep in mind that differences between first- and second-language (L1 and
L2) readers may influence interpretation of the reading construct. This section discusses
fundamental differences in L1 and L2 reading (see Appendix D for a list of additional
differences). In this context, “L2 readers” refers specifically to individuals learning to read a
second language(s) after achieving reading competence in their L1, since they represent the great
majority of TOEFL examinees. With this particular subgroup of L2 readers, three fundamental
differences distinguish L1 and L2 reading: (a) L2 readers build on prior L1 reading experience,
(b) their reading processes are cross-linguistic, involving two or more languages, and (c) their
reading instruction usually commences before adequate oral proficiency in the target language has
developed. Not surprisingly, these differences tend to engender qualitatively different
comprehension procedures. The uniqueness of L2 reading stems from at least four additional
factors beyond those accounting for performance variability within L1 reading: (a) transfer of L1
reading skills and strategies, (b) facilitation resulting from L1–L2 structural similarity, (c) cross-
linguistic interactions during L2 reading, and (d) processing constraints imposed by limited
linguistic knowledge. Interestingly, these same factors also provide a basis for formulating
frameworks through which L2 reading behaviors can be described and explained – as well as
clarifying some of the ways in which L2 reading theory recasts accepted L1 constructs.
Such frameworks, moreover, permit several explicit predictions regarding performance
diversity, emanating from the unique experiences among L2 readers. One, given that language
processing skills are shaped to accommodate the structural and functional peculiarities of a
particular language, and that these skills transfer across languages, it is highly conceivable that
prior L1 processing experience is directly associated with procedural divergence at certain points
in the development of L2 proficiency, invoking qualitative differences in L2 processing
behaviors. Two, given that L2 readers have substantial L1 processing experience, it seems
reasonable to expect that L2 reading can be facilitated by prior L1 experience, at least to the
extent that the L1 and L2 linguistic systems share similar structural properties. This, in turn,
suggests that L1-L2 linguistic distance may be responsible, at least in part, for quantitative
differences in L2 reading performance between those with related and unrelated L1 backgrounds.
Three, since processing competence develops through experiential exposure to a particular
language, we can further predict that L1 and L2 processing experiences jointly impact L2
reading skills. And, finally, although cognitively and metacognitively mature, L2 readers are
linguistically limited, and thus likely to develop coping tactics, which are, at certain points in the
development of L2 proficiency, qualitatively different from those used by L1 readers of the target
language, within their restricted linguistic resources. The sections that follow elaborate upon
these predictions based on empirical findings from recent L2 reading research.
Transfer of L1 Reading Skills and Strategies. L1 comprehension studies suggest that
cognitive and metacognitive skills, once acquired, are transferable to other situations posing
similar cognitive requirements (e.g., Palinscar & Brown, 1984; Raphael & Pearson, 1982;
Guthrie, 1988). A number of L2 acquisition studies also demonstrate that various linguistic and
metalinguistic elements are transferred from L1 in both oral and written forms of L2 production:
e.g., morphosyntactic systems (e.g., Hakuta, 1976; Zehler, 1982; Gundel & Tarone, 1983; Yanco,
1985; Rutherford, 1983), Communicative strategies (e.g., Cohen, Olshtain, & Rosenstein, 1986;
Olshtain, 1983; Scarcella, 1983), and pragmatics (e.g., Irujo, 1986). These findings suggest that
some reading skills – acquired in one language – can be applied to another language.
A large number of studies have, in fact, investigated reading skills transfer across languages.
Two major perspectives have dominated this research: One based on the presupposition that
reading procedures are universal across languages (e.g., Goodman, 1973), and the other, on the
conviction that reading involves language-specific processes. Earlier transfer studies occurred
within the universal framework, focusing on two major issues: (a) the interrelationship between
L1 and L2 reading competence (e.g., Skutnabb-Kangass & Toukomaa, 1976; Cummins et al.,
1981; Legarreta, 1979; Troike, 1978; Cummins, 1979, 1991), and (b) the conditions inhibiting, or
facilitating, reading skills transfer from L1 to L2 (e.g., Clarke, 1979; Devine, 1987, l988). These
studies, however, give little attention to the precise nature of the skills to be transferred from one
language to another.
In recent times, controversy has developed among linguists, psychologists, and educators
about the universality of language acquisition and processing. Experimental psychologists, for
example, are challenging theories of word recognition stemming from data obtained exclusively
with English-speaking subjects. Subsequent cross-linguistic investigations have been carried out,
comparing skilled L1 readers with varying orthographic backgrounds. The findings generally
reinforce the likelihood that different information-processing procedures are used with particular
orthographies (e.g., Turvey, Feldman, & Lukatela, 1984; Navon & Shimron, 1984; Hasuike,
Tzeng, & Hung, 1986; Sasanuma, 1984; Vaid, 1995). Similarly, child language studies
demonstrate that children cannot deal systematically with linguistic forms which violate their
perceptions of the prototypical sentence structure of their native languages (e.g., Berman, 1986;
Slobin & Bever, 1982; Hakuta, 1982), thereby suggesting that learners are sensitized to the
specific linguistic features very early in their language development. Such linguistic conditioning
not only serves to shape cognitive strategies appropriate to individual languages, but also plays a
central role in regulating the perception and interpretation of linguistic input (e.g., Slobin, 1985;
Bates & MacWhinney, 1989).
The language-specific perspective of transfer has emerged from these newer
conceptualizations which contradict assumptions underlying the universal perspective. Current
transfer studies indeed demonstrate that L2 readers with typologically diverse L1 backgrounds
utilize qualitatively different procedures at certain points in development (e.g., Akamatsu, in
press; Brown & Haynes, 1985; Green & Meara, 1987), and more critically, that such procedural
diversity among L2 readers is identifiable with structural variations in their respective L1s (e.g.,
Koda, 1989, 1990, 1993; Ryan & Meara, 1991). A recent study (Koda, 1999) demonstrates,
moreover, more subtle, yet potentially significant, L1 influence on L2 lexical development among
ESL readers with non-Roman alphabetic (Korean Hangul) and non-alphabetic (Chinese
logographic) L1 backgrounds. Given that logographic and alphabetic readers engage in intraword
structural analysis to a differential degree during decoding, it was hypothesized that differential
amounts of L1 intraword analysis experience would be causally related to the formation of L2
intraword sensitivity and subsequent decoding competence. The results compound an already
complex picture. While the two groups differed neither in intraword sensitivity, nor in decoding,
a clear contrast existed in the extent to which intraword sensitivity and decoding skills related to
reading comprehension: i.e., while the three variables were closely interconnected in the Korean
data, no such direct relationships were found in the Chinese data. These results clearly indicate
that differential L1 orthographic experience does not, of necessity, result in any quantitative
differences, but such L1-based variations may induce a strong preference for particular processing
Viewed collectively, reading transfer studies make it plain that L1 processing experience has
a long-lasting impact on the development of L2 reading skills. Further research is desirable to
clarify the extent to which transferred L1 reading skills are incorporated in L2 reading, as well as
the ways in which the use of L1 skills alters the basic reading process among L2 learners.
Probable Facilitation Stemming from L1–L2 Structural Similarity. Given that reading skills
transfer occurs during L2 processing even when L1 and L2 are typologically unrelated, it can be
expected that the development of L2 reading skills is facilitated by L1 processing experience at
least to the extent that the two languages share similar structural properties. Should this be the
case, it can be further predicted that L1-L2 linguistic distance is in part responsible, at certain
points in development, for efficiency differences in L2 reading performance. L2 lexical
processing research, involving ESL learners with diverse L1 backgrounds, repeatedly
demonstrates superior performance (faster and more accurate) among those with congruent, rather
than incongruent, L1 processing experience (e.g., Green & Meara, 1987; Muljani, Koda, &
Moates, 1998; Koda, in press). Logically then, the critical question in this research is precisely
how L1 and L2 structural similarities facilitate L2 processing performance.
A recent study sheds substantial light on the issue by directly testing the L1-L2 distance
effects on L2 processing efficiency (Muljani, Koda, & Moates, 1998). By comparing lexical
decision performance among ESL learners with related (Indonesian; Roman-alphabetic) and
unrelated (Chinese; logographic) L1 orthographic backgrounds, the study showed first that
Indonesian subjects outperformed the Chinese across conditions; and second, that intraword
structural congruence (i.e., spelling-pattern consistency) between Indonesian and English
benefited the Indonesian, but not Chinese, subjects. The findings clearly suggest that the
performance superiority among Indonesian ESL students is attributable to the accelerated
efficiency in the precise aspects where their L1 and L2 pose identical processing requirements.
L1-L2 linguistic distance, thus, not only explains overall performance differences among learners
from related and unrelated L1 backgrounds, but also underscores the ways in which L1
experience facilitates L2 lexical processing.
One way of expanding linguistic distance research would be to systematically compare
longitudinal changes in processing competencies among L2 learners with related and unrelated L1
backgrounds. There are at least three logical possibilities. One, once L2 readers gain specific
levels of processing competence, linguistic distance should have little effect on performance
efficiency or processing strategies. Linguistic distance effects, in short, will be apparent until L2
readers attain processing competence at the threshold level, but not thereafter. Two, the initial
efficiency gap will never close. Should this be the case, we can expect that L2 learners from
unrelated L1 backgrounds will always lag behind those with related backgrounds. Three, L2
readers from unrelated L1 backgrounds may gain processing efficiency by using qualitatively
different processing mechanisms, which will result in diverse processing procedures among
different L1 groups.
To date, only a handful of studies on record have directly addressed the L1-L2 distance
effects in L2 reading. It would be considerably advantageous if more research could be
specifically directed towards the clarification of the long-term impacts of qualitatively different
L1 processing experience on L2 reading skills development.
Cross-linguistic Interactions during L2 Reading. Despite the fact that L2 reading involves
at least two languages, limited attention has been given to the cognitive interplay between the two
languages, as well as its resulting impacts on L2 reading behaviors. A simple comparison of the
surface features of the two languages, as in the case of the Contrastive Analysis Hypothesis, only
provides a limited view of the potential interaction intricacies. A simplistic analysis may not only
yield an inadequate description of the interactions (Zobl, 1983, 1984; Zehler, 1982), but also may
engender inaccurate predictions regarding the consequences of the interactions (e.g., Whitman &
Through careful comparisons of intraword awareness among L2 readers, systematic
investigations of cross-linguistic interactions during L2 processing are currently underway.
Intraword awareness refers to readers’ understanding of words’ internal structure and their ability
to use the structural insights during lexical processing. The role of such awareness in early
literacy development has attracted considerable attention among L1 reading researchers for
almost two decades. The emerging consensus is that learning to read is fundamentally
metalinguistic, involving the recognition of the basic units of spoken language and the units of the
writing system, and the mapping between the two (e.g., Nagy & Anderson, 1997; Fowler &
Liberman, 1995; Goswami & Bryant, 1992). Recent research, moreover, consistently
demonstrates that intraword awareness among young L1 readers develops primarily through
cumulative print processing experience in their language (e.g., Perfetti, Beck, Bell, & Hughes,
1987; Bowey & Francis, 1991; Vellutino & Scanlon, 1987). This, in turn, implies that processing
experience in the target language is likely to be a major force in shaping intraword awareness
among L2 readers, regardless of their L1 backgrounds. Given that L2 readers rely upon L1
processing skills, however, it is conceivable that L2 input is filtered through the structural
sensitivity developed in L1 reading (L1 intraword awareness). It can be expected, therefore, that
the resulting L2 awareness is an amalgamated form of cross-linguistic interactions of L1 and L2
Recent studies involving ESL readers (Koda, 1999, in press; Koda, Takahashi, & Fender, in
press) repeatedly show that (a) L1-L2 processing congruity is directly related to efficiency
differences among L2 readers with varying L1 backgrounds; (b) ESL learners with equivalent
proficiency are sensitized, to a similar degree, to the internal structure of English words,
regardless of their L1 backgrounds; and (c) variations in L1 processing experience predicted
performance differences in some, but not all, aspects of L2 intraword awareness. These results
indicate that the aspects of L2 processing competence, central to analyzing and manipulating L2-
specific linguistic features, develop primarily through repeated experience with the lexical
peculiarities in the target language and, therefore, are largely unaffected by differences in L1
processing experience. Nonetheless, the groups’ response patterns systematically varied in other
aspects as a consequence of their L1 processing experience. Viewed collectively, the findings
suggest that L1 and L2 knowledge are both operative, in their own unique ways, in the
development of L2 processing competence. Obviously, further research is necessary – and
essential – before we can uncover the complexities of cross-linguistic interactions transpiring in
Processing Constraints Resulting from Limited Linguistic Knowledge. A fourth dimension
distinguishing L1 and L2 reading is the degree of linguistic proficiency that learners have
acquired prior to reading instruction. In L1 reading, children have already mastered the basic
language structure before instruction begins. They are, moreover, continuously exposed to
written symbols in their cultural environment (e.g., food packages, commercial logos, trademarks,
billboards, etc.), enabling them to formulate visual images of words and establish strong
associations between the oral and written forms of their language (e.g., Sulzby, 1986; Ferreiro,
1986; Mason, 1980). This is rarely the case with adult L2 learners. Not only are they required to
read before attaining either adequate oral proficiency or “environmental literacy,” but, often, they
must also deal with highly decontextualized materials from the outset.
One way of assessing the consequences of insufficient linguistic knowledge is to determine
how such limitations constrain L2 reading processes. It is highly probable, for example, that
limited linguistic knowledge restricts L2 readers’ ability to identify the L2 morphosyntactic
features which provide essential information for comprehension. Empirical data, in fact,
demonstrate that information sampling patterns among L2 readers are strikingly different from
those used by native readers of the target language (e.g., Bernhardt, 1987; Hatch, Polin, & Part,
1974; Saito, 1989). Given that varying aspects of morphosyntactic knowledge contribute
differently to sentence processing (e.g., MacWhinney & Bates, 1989; Koda, 1990), the
differential visual processing patterns among L1 and L2 readers indicate that L2 readers pay less
attention to significant information — and more attention to less significant elements — than their
An alternative approach to assessing the effects of limited linguistic knowledge on L2
reading is to compare reading behaviors across proficiency levels. A notable efficiency difference
exists in lower-level verbal processing. It has been reported, for example, that with increased L2
proficiency, processing speed improves (Favreau & Segalowitz, 1982; Haynes & Carr, 1990), and
error rate decreases (Bernhardt, 1991). Similarly, eye-movement studies consistently demonstrate
that while the number of eye fixations does not differ widely across
proficiency levels, the fixation duration among lower-proficiency learners is considerably longer
than that among higher-proficiency learners (e.g., Oller, 1972; Oller & Tullius, 1973; Bernhardt,
1987; Saito, 1989).
Recent L2 studies repeatedly show that (a) oral language proficiency is not a strong predictor
of either lexical processing or reading comprehension (Durgunoglu, Nagy, & Hancin, 1993; Geva
& Siegle, 1999; Gholamain & Geva, 1999) and (b) inefficient word recognition reduces L2
reading performance among otherwise fluent bilinguals (Segalowitz, 1986; Segalowitz, Poulsen,
& Komoda, 1991). These findings clearly suggest that linguistic knowledge alone provides a
necessary — but insufficient — condition for developing L2 reading competence. Further, it has
been suggested that two specific dimensions of linguistic knowledge, orthographic and
phonological, independently influence word recognition (e.g., Stanovich & West, 1989;
Stanovich, 1991a, 1991b; Adams, 1990; Barker, Torgesen, & Wagner, 1992). Importantly, the
two may, or may not, develop concomitantly with other aspects of linguistic knowledge.
Presumably, then, we must assume that first, L2 readers do not always possess these specific
aspects of linguistic knowledge, and second, even if they do, they may not yet have developed the
skills to use the knowledge during reading.
Importantly, inefficient lower-level verbal processing skills have two major consequences for
comprehension performance. First, inasmuch as reading comprehension necessitates the
construction of meaning based on textual information, the meaning construction process is
seriously impaired when sufficient information is not extracted, or when the extraction is
inaccurate. Second, underdeveloped processing skills strain limited capacity working memory,
restricting higher-level conceptual processing. It seems likely, therefore, that L2 readers engage
in text-reader information integration to a far less extent than L1 readers at least until sufficient
lower-level processing skills are acquired. It has been reported, in fact, that lower-level
processing predominates in the reading process among beginning L2 readers (Clarke, 1979;
Horiba, 1993). Hence, the interactive mode of reading, commonly presumed in contemporary L1
reading models, may not be reflective of processing behaviors among L2 readers.
Implications for Reading Assessment among L2 Learners. In an attempt to explicate the
singular characteristics of L2 reading, four conditions have been described wherein performance
diversity among L2 readers is directly associated with their unique processing experiences. Given
that L2 readers rely on L1 processing skills even when L1 and L2 differ typologically, there is
good cause to assume that processing procedures used by L2 learners with diverse L1
backgrounds are qualitatively different at specific points in development. This, in turn, suggests
that all L2 learners may not develop the exact same skills repertory, and therefore, the cognitive
tasks required in L2 reading may not be accomplished in the same way. These unique
characteristics of L2 readers yield at least two important implications for L2 reading assessment.
First, L2 reading research demonstrates that the degree of L1 and L2 processing congruity is
closely related to performance efficiency — particularly processing speed — among L2 readers.
It seems likely, therefore, that L1-L2 linguistic distance may have a definitive impact on
assessment outcomes. Although superior processing performance is observable among those with
similar L1 backgrounds, benefits of this sort of facilitation occur only when L1 and L2 pose
identical processing demands, and generally are not transferable to the other sub-component
processes. Higher efficiency in one aspect resulting from congruent processing procedures, in
short, can spawn exaggerated, misleading indicators in overall reading competence among those
with typologically related L1 backgrounds. The restrictive, uni-dimensional nature of L1-linked
facilitation, thus, prompts the need for caution in using timed tasks in assessing a limited range of
L2 reading skills.
Second, current L1 reading models commonly assume that reading comprehension stems from
reader-text interaction. L2 research, nonetheless, repeatedly shows that lower-level processing
predominates in the L2 reading, thus indicating that L2 readers engage in far less interactive
processing than generally assumed in L1 reading. This discrepancy clearly warrants special
consideration: Namely, performance data obtained through assessment procedures based solely on
L1 principles may not accurately reflect presumed reading capability, especially among beginning
3. Reading Framework for the TOEFL 2000 Test
Identifying the Test Domain
The framework document for the TOEFL 2000 project identified the TOEFL 2000 test
domain (Jamieson et al., 1999, p. 10). Since the test is intended to “measure examinees’ English-
language proficiency in situations and tasks reflective of university life in North America,” the
reading component of the test will reflect the types of reading that occur in university-level
academic settings. The test will be designed to improve discrimination at the upper levels of
English as a second language/English as a foreign language (ESL/EFL) reading proficiency.
Organizing the Test Domain
It was decided that the most practical way to organize language tasks for the TOEFL 2000
framework is by modality (Jamieson et al., 1999). This decision was influenced by the desires of
admissions officers and graduate deans. Thus, “the test will include measures of speaking,
writing, listening, and reading. Within these four areas, the new test will include a variety of
language features, including not only grammar and vocabulary but also discourse, pragmatics, and
sociolinguistics as well as setting and task” (p. 12). It is important to note that the four modalities
will be tested both independently and integratively.
Identifying Task Characteristics
Assuming the TOEFL 2000 test domain and organizational scheme described above, we will
now consider reading task characteristics in terms of the three broad areas identified in the
TOEFL 2000 framework document (Jamieson et al., 1999). The following sections discuss
situation, text material, and test rubric respectively.
Situation. As outlined in TOEFL 2000 Framework: A Working Paper (Jamieson et al.,
1999) situation task characteristics are defined, based on Crystal (1991, 1992) as “extralinguistic
elements associated with language tasks.” Task characteristics to be considered include
participants, setting, content, purpose, and register.
Most academic reading occurs without reliance on shared physical or social contexts. These
aspects of situation seem to be more relevant to language modes such as speaking and listening,
which require more interpersonal skills than reading does. As Cummins (1991) points out,
written language is relatively decontextualized in the sense that the type of paralinguistic cues that
exist in face-to-face oral communication are missing. The immediate feedback that can help
correct breakdowns in communication is also missing. Durgunoglu (1997) argues that “in
contrast to the current emphasis on contextualization and communicative competence, formal
vocabulary knowledge is more strongly related to L2 reading proficiency than oral proficiency is.”
To some extent, therefore, we address issues related to participants and setting from a
necessarily different perspective than that likely to be used in the other skill modalities. Where
reading tasks are integrated with tasks in other language skill areas, however, other factors will
naturally influence these aspects of the reading situation. For example, in a combined reading and
listening task, the participants and setting may influence the overall situational context.
TOEFL 2000 Framework: A Working Paper (Jamieson et al., 1999) discusses participants
in terms of the people involved in the language act and the relationships among them, and suggests
that this variable can be operationalized in terms of gender, ethnicity, age, and role. The
identification of participants in the reading context, however, requires a more abstract view of the
participant than it does, say, for the listening, speaking, or even the writing context.
One participant is clearly the reader, who constructs a text model based on the information in
the text and a situation model to interpret the text that is based on his or her background
knowledge, goals for reading, motivation, attitudes, and evaluation of the information in the text
The authors of the reading passages or documents on the test may also be viewed as
participants. In some circumstances, e.g., in an argument in which authors do not agree,
characteristics of the authors as participants are likely to be relevant to understanding the text. In
other cases, the text’s author is fairly anonymous and is treated as unopinionated. This is the
pattern in the current TOEFL test passages. The role of the author as participant is likely to be
less important in some types of texts, such as expository texts, than in other types of texts, such as
argumentative ones. In cases where argumentative or persuasive texts are included, we would not
expect it to matter who the participants are in terms of gender, ethnicity, or age, but we believe
the role of the participants in the sense of their argumentative stance, their motives, and interests
in the argument, and so forth, may influence task difficulty.
Biographical/autobiographical narrative raises the author issue with respect to the relationship
between the author and the person being written about. If historical biographies/autobiographies
are used in the TOEFL 2000 test, this issue will need to be addressed.
In TOEFL 2000 Framework: A Working Paper (Jamieson et al., 1999, p. 15), setting is
defined as “the place where the language act occurs.” The framework document proposes that
three types of setting be represented: instructional milieus, which include “all places where formal
instruction takes place, such as lecture halls, labs, seminar rooms, and classrooms;” academic
milieus, which include “typical places outside of the classroom where aspects of academic life are
dealt with, such as a study room in a dormitory, the library, an instructor’s office, the bookstore, a
writing center, or a computer center;” and non-academic milieus, which include places that are
not usually associated with academic content but where social and business transactions take
place, such as “the business office, international students’ office, and the health center, as well as
dormitory room and dining areas.”
While authentic reading activities can certainly be carried out in each of these environments,
it does not seem likely that these different settings would require different types of reading texts,
or would influence the difficulty of the reading tasks associated with them. Therefore, we do not
recommend that these types of variations in settings be included in the design of the reading test
per se. The setting where a text appears could, of course, be addressed by indicating the source
from which it was taken (textbook, journal article, etc.) and providing a title or some type of
Physical settings for reading activities could most authentically be created in combination
with the measurement of other language skills. In the case of the reading construct, setting might
be relevant to some extent in situations involving reading to integrate across texts. For example,
in a lecture hall setting, where an examinee listens to a more extended lecture, some things might
be written on blackboards or presented as slides or handouts for reading. Or in a laboratory class,
after some oral instruction, a lab manual might present some written text information that needs to
be read before answering questions by writing or speaking. In order to situate the reading tasks in
a specific location, however, an oral component seems necessary. It is assumed, therefore, that
considerations of physical setting for reading tasks will generally be considered in integrative
contexts where the situational features of the oral environment might provide greater authenticity.
Any subject area that is typical of academic study could provide appropriate material for the
reading test. The current TOEFL test covers a range of very general academic topics broadly
classified as topics related to the Arts, Humanities, Social Sciences, Physical Sciences, or Life
Sciences. It seems appropriate to continue to include as much topic variety as possible in the new
test. As with the current test, however, care should be taken to ensure that specialized knowledge
of a particular field is not necessary to understand the information presented in the passages.
Appropriate pragmatic and rhetorical features for TOEFL 2000 reading texts are discussed
under the section on text material.
In TOEFL 2000 Framework: A Working Paper (Jamieson et al., 1999, p. 15), purpose is
defined as “the reason why we engage in tasks.” This definition is then linked to Halliday’s
(1973) list of seven categories, six of which are identified as relevant to TOEFL 2000 purposes
for which international students would use English in a North American university, namely for
heuristic, instrumental, regulatory, personal representational, and interactional purposes. While
some or all of these purposes are very relevant to other skill modalities, the heuristic purpose is
most relevant to reading in an academic context. Assuming that most of the reading that college
students do is heuristic, we believe that the reader’s purpose and goals, more narrowly defined,
are critical to reading performance. Van Dijk (1985) has argued that strategies are selected in
terms of the reader’s purpose and determined need. Goldman (1997) also notes that “. . . readers’
expectations about their task determine the knowledge and strategies that are brought to bear
during the comprehension process” (p. 366). Such a statement suggests a purpose-driven
framework linked more to processing demand than communicative goal. The purpose-driven
framework we have proposed for the reading test reflects this emphasis: (a) Reading to Find
Discrete Information, (b) Reading for Basic Comprehension, (c) Reading to Learn, and
(d) Reading to Integrate Information across Texts.
The term register, as a cover for the uses of texts and the author’s intentions, is reflected in
patterns of linguistic features that tend to co-occur as well as arrangements of discourse
information and conventional genre forms. For some researchers, register simply refers to
occupationally defined texts and specific textual sub-types such as legal register, business
register, sports-announcer register, etc. For other researchers, register is equated with genre, and
specific functional text types are seen as registers. Examples of these would include sermons,
lectures, letters of apology, etc. For yet other researchers, register is a superordinate category for
the author’s intent, e.g., to establish relations; provide information; narrate stories, procedures, or
ideas; indicate level of support; indicate closeness of relationship; or imply factualness. Biber
(1995) gives a good review of these positions.
Viewing register in terms of the varying functional uses that texts serve seems to be most
relevant, and most useful, for the TOEFL 2000 project. Identifying occupationally defined
language usually does not specify texts that occur all across the space of occupational
interactions, nor does it define in sufficient detail the critical marking aspects. A focus only on
specific genres is probably not very useful since most testing contexts will allow only a fairly
limited range of general texts as input for testing purposes.
A number of authors have proposed a set of underlying parameters that reflect the many uses
texts serve that are signaled by, and through, texts. The key to understanding and using these
various efforts is to recognize that the parameters are all active and play some role in every text.
So, unlike discussion of structure or speech acts or some discussions of style, the many
parameters of a useful register theory are all operating at the same time (see Appendix E for a
more detailed review of several sets of register variations).
In our view, the best way to proceed would be to set certain register parameters for the
TOEFL 2000 reading test and stay within them for the most part in order to develop test forms
that are parallel in terms of register. This would mean establishing some system for analyzing the
register parameters of reading passages, then incorporating this information into test
specifications in order to ensure that test forms of comparable overall register variation are
administered to different examinee populations. Later research and experimentation can establish
whether or not additional register dimensions should be added. The most common assumption
among test score users, and among L2 test takers, is that expository prose types with minimal
emotional impact and vagueness constitute the target reading text model (though we will explore
argument and historical/autobiographical narrative). We should have strong reasons for changing
these assumptions. There are plenty of resources available to continue development of typical
TOEFL reading passages of varying length and complexity, and there are many ways to create
new item types from these passages without adding completely new dimensions to the test
development matrix from the perspective of register variation.
Text Material. We recommend that tasks in the TOEFL 2000 reading test be based on a
variety of text materials. Some could be based on a single text, ranging in length from a short
paragraph to a lengthy selection; others could draw on multiple texts, also ranging in length. In
some cases, the text might be presented without any visual material, while in other cases a text
selection (or selections) could be accompanied by graphics, such as line drawings, schematics,
photographs, or maps, or by charts, tables, or graphs presenting data. If technology permits, some
of the tasks might incorporate video material.
As defined in TOEFL 2000 Framework: A Working Paper (Jamieson et al., 1999) text
material for the TOEFL 2000 test will consist of three types of features: (a) grammatical features,
which relate to the syntax of the sentences and the vocabulary used in the text;
(b) discourse features, which relate to specific discourse features that signal relationships among
parts of the text; and (c) pragmatic features, which relate to the intent of the text’s creator and, the
rhetorical features or organizational patterns of the text. Thus grammatical, discourse, and
pragmatic features have been regrouped here into two categories: grammatical/discourse features,
and pragmatic/rhetorical features.
Syntax. The role of syntax is likely to be very important with respect to understanding the
construct of reading but much less important as a set of knowledge and related abilities to include
in planning task difficulty and developing item types for the TOEFL 2000 test. The contribution
of syntactic knowledge to processing efficiencies in reading comprehension was briefly discussed
earlier. In particular, syntactic information in reading supports the extraction of appropriate
propositional information and also comprises a general set of instructions to readers for building
text-model structures that map onto existing structures or generate new sub-structures. Exactly
what aspects of syntactic information appear to contribute most to the processes of mapping
structures and generating new structures (“shifting” in Gernsbacher's terminology) is not clear,
but the current research of Gernsbacher and others is building a better picture (Gernsbacher 1990,
1996, 1997; Givón, 1995; Perfetti, 1997). For example, some research has explored the
contributions of articles, tense and aspect marking, locatives, and other systems and structures. It
should also be noted at the outset that syntactic features and discourse features may not be as
separable as many previous analyses assume; the combining of syntactic markers and discourse
features is common in the work cited above, as well as in the work of Freedle and Kostin (1993),
Nissan, De Vincenzi, and Tang (1996), and others.
The line of research on syntactic contributions to discourse processing does not typically
propose that there are specific syntactic structures that contribute to reading comprehension
difficulty in any way that would suggest the assessment of specific isolated structures. Rather, the
notion of syntactic support for reading comprehension rests more with the combined sets of
signals that structural information provides: it contributes to efficient processing of information in
working memory, it establishes and supports semantic relations between arguments and
predicates for proposition formation, and it adds contextual information to help disambiguate
lexical meanings. It may also provide important information for other purposes.
With respect to syntactic information as a source of specific reading difficulties, few research
studies have isolated significant structural characteristics that deserve to be included as task
development variables for testing purposes. Freedle and Kostin (1993) and Freedle (1997) have
argued that a small subset of syntactic variables accounts for difficulty in main idea reading
Another source of syntactic/structural differences is noted in the work of Just and Carpenter
(1992; Carpenter et al., 1994). They argue that syntactic complexity, as a singular concept, leads
to lower reading efficiencies. They also note that competing noun-predicate (NP) arguments in
the immediate environment reduce processing efficiencies, as does referential distance; the longer
the distance between antecedent and co-referent, the lower the efficiency.
Vocabulary. The role of vocabulary in determining task and item difficulty is likely to be a
large one. In addition to needing some way to provide frequency data on vocabulary use,
vocabulary indices for which definitions are used for a word (knowing a word’s different
meanings), and for other possible variations (see below) need to be explored. Formality measures
(a register feature) related to vocabulary can be assessed by determining the percentage of
Latinate words in a text, or word length in a text. This particular measure has generated mixed
results, but it should be examined again for the TOEFL 2000 test. Type-token measures or some
other measure of the number of new words in a text should be explored. Additionally, a measure
of the percentage of uncommon words in a text – an uncommon-type/token ratio – might be more
revealing for text readability for second-language (L2) readers. It might also be useful to
investigate the difficulty of vocabulary by text type. Other factors relating to vocabulary might
include any or all of the following (see Richards, 1976; Nation, 1990):
• Collocability (in fairly tight phrasings)
• Functional limits according to use and situation (register and function constraints)
• Syntactic behavior (parts of speech, sub-categorization, case roles, transitivity)
• Basic forms and derivational possibilities and typical occurrences
• Associational patterns with other words in domains of knowledge and use
• Idiosyncratic features of specific words
• Learning difficulties of certain words (e.g., similar-looking words with different meanings)
• Degree of abstractness-concreteness
Discourse features. Still other consistent contributors to text processing difficulties may be
found at the level of discourse organization. The specific role of transition markers is commonly
examined, though there does not appear to be a clear synthesis of this research available. Part of
the problem is that this topic is addressed through a number of discipline areas and research
methods; further, it is not easy to delimit the scope of the notion of “transitional marker.” Most
researchers simply note the subset they want to address with a brief nod to some prior rationale,
but this methodological step does not help us establish the potential contribution of transitional
discourse markers to differences in reading comprehension abilities.
A number of other features of discourse marking and information structuring should be
considered at some point by TOEFL 2000 research studies. These include the roles of theme-
rheme structuring, given and new information, definiteness and indefiniteness, noun-predicate
density in texts, the positions of main idea and topic sentences in texts, as well as other issues that
can be developed from the research literature. A full study of such potential contributing features
might be a useful resource document for future specific research studies and for the development
of comprehension item types.
Written discourse has been classified by researchers in a variety of schemes (Brewer, 1980;
Britton & Black, 1985; D’Angelo, 1980; Moffett, 1983; Mosenthal, 1985; Vacca & Vacca, 1996;
van Dijk & Kintsch, 1983). We recommend that reading passages for the TOEFL 2000 test be
classified according to their dominant pragmatic and rhetorical features. The pragmatic features
convey the primary intent of the author, while the rhetorical features indicate the higher level
organization of the text. The following sections describe these features.
Pragmatic features. According to Brewer (1980), written discourse may serve to inform,
entertain, or persuade, or may have a literary-aesthetic intent. The TOEFL 2000 test will focus on
academic reading tasks across a range of topics and fields; appropriate types of text materials in
these subject areas are texts classified as “expository” or “argumentative persuasive,” the primary
intent being to inform or persuade the reader. An additional type of text in these areas may be
classified as “historical/biographical narrative,” which would include passages about the
contributions of individuals to the disciplines. Fuller descriptions of each of these types of text
Exposition – This type of text primarily serves to inform the reader. Often this is the type of
reading most prevalent in college classrooms. Lengthy expository text passages may include
descriptions, comparisons, contrasts, explanations, and elaborations that provide details about
concepts, objects, persons, places, events, and other phenomena. The following passage is
characteristic of expository text in that it informs the reader, in this case about the different types
of brain wave electrical activity measured by an electroencephalogram.
When you are fully awake and alert, your EEGs contain many beta waves,
relatively high-frequency (14 to 30 Hz), low-voltage activity. As you enter a
quiet, resting state – for example, just after getting into bed and turning out the
light – beta waves are replaced by alpha waves, EEG activity that is somewhat
lower in frequency (8 to 13 Hz) but slightly higher in voltage. As you begin to
fall asleep, alpha waves are replaced by even slower, higher-voltage delta waves.
The appearance of delta waves seems to reflect the fact that increasingly large
numbers of neurons are firing together in a synchronized manner.
(Baron, 1992, p. 140)
Typically, the rhetorical features of the text (discussed later in this section) guide the reader in
a basic understanding of the passage. Readers who are new to the discipline are often unfamiliar
with the vocabulary that is introduced in expository passages; accordingly they must utilize
accompanying figures for clarification of the new material, and must have well-developed reading
and study strategies for choosing the salient aspects of the passage.
Argumentation/Persuasion/Evaluation – Argumentative/persuasive texts present a point of
view about a topic and provide supporting evidence in favor of a position in the analysis of the
topic. Good persuasive texts will include carefully crafted positions with reasons and evidence
along with an analysis of the opponent’s errors in reasoning.
Argumentative/persuasive texts are characterized by diction that may be personal in tone, by
vocabulary that points to an attitude or perspective toward the topic, and by a style that departs
from a measured, unbiased stance. For instance, the following example of an argumentative text
shows the writer’s opinion about the custom of sending Christmas letters.
I would like to hold a contest for the most fatuous Christmas letter, but I’m afraid
I’d be deluged with entries. It is hard to attribute the Christmas letter to a
particular type of person or a particular station in life, because almost everyone
who has ever had an eventful year, taken an exciting trip, or accomplished a great
deal has felt the urge to compose one. I have received them from internationally
famous professors who were attempting to describe their world travels, from
graduate students describing their Ph.D. research in the field, and from relatives
recounting the latest family gossip. Perhaps mimeographed Christmas letters
should be used as a vanity indicator, since they expose those among us who
yielded to, rather than resisted, the pervasive temptation to blow one’s own horn.
(Johnson, 1971, pp. 44-5)
Argumentative texts are typically found in editorials, essays, political satire, and other types
of texts where the intention is to present a point of view. The work of the reader is to analyze the
writer’s perspective in relation to the topic and judge the worth of the writer’s presentation, line
of reasoning, and evidence.
Historical Biographical/Autobiographical Narrative –- Narrative text tells a story.
Consequently, essential elements in this type of text are the setting (which includes the characters
and story context) and episodes to reach a goal or solve a problem, which include the initiating
event, internal response, attempt, the consequence, and the reaction (Mandler & Johnson, 1977;
Stein & Glenn, 1979; Stein & Policastro, 1984). These descriptors of story elements constitute a
story grammar and are considered critical components for comprehension of narrative text.
The most common types of college academic reading that contain narrative discourse are
historical, biographical/autobiographical, or literary (fiction) text. The TOEFL 2000 test will
include passages from a variety of fields, but the use of literary text is not recommended because
of fairness issues associated with the cultural background knowledge inherent in such texts.
Types of text selected from historical biographical/autobiographical narratives feature the lives of
prominent individuals throughout history. The following examples illustrate this type of passage.
For a few weeks every autumn, when the fields around Chicago were ripe for
harvesting, children of Mexican migrant workers joined our classes at the Eugene
Field School in Park Ridge. They never stayed long, because their families were
always moving on to the next harvest. One year, a boy who was older than I and
big for his age took to pushing me and my friends around on the playground.
Whatever motivated him, his bullying quickly aroused my fear and my dislike.
My reactions to this one boy might well have spilled over to my feelings about
the rest of the migrant kids if my mother had not encouraged me to volunteer,
along with other girls in my church youth group, to baby-sit for the migrants’
younger children on Saturdays so that their older brothers and sisters – my
classmates – could join their parents working in the fields.
Just seeing the camp where the families lived made me think for the first time
about how my classmates spent their time when they were not in school. I had
never before known people who lived in trailers. When we went inside, the
mothers seemed nervous about leaving their babies and toddlers in the care of
twelve-year-olds who spoke no Spanish. I began to realize the lack of familiarity
cut both ways.
The day passed uneventfully, and when the mothers returned, they expressed
their pleasure at seeing their children well cared for. It was the return of the
fathers, though, that made the greatest impression on me. When the buses
dropped them off at the base of the long road to the trailer camp, the children ran
as fast as they could to greet them. They were filled with excitement, the same
excitement I felt when my own father came home from work at the end of the
day. Suddenly those migrant children didn’t seem so different from me. This
brief encounter helped me begin to appreciate the importance of making
judgments about individuals instead of stereotyping whole groups. It also gave
me a lot of satisfaction at an early age to be serving families who worked so hard
for so little.
(Clinton, 1996, pp. 182-3)
Jung began attending seances and table turnings which were held at the home of
relatives every Saturday night. His interest in the occult never diminished, and
for his doctoral dissertation he investigated the behavior of a medium, a fifteen
year-old girl who performed at the seances of his relatives.
These mysterious phenomena were instrumental in turning Jung’s interest to
psychology and psychopathology. That fall when he returned to the university he
read a textbook on psychiatry by Krafft-Ebing in preparation for his final
examinations. The first chapter struck him like a bolt of lightning; he knew
immediately that psychiatry was his destined field. In his twenty-fourth year,
then, Jung had finally found the field that was compatible with his interests,
speculations, and ambitions. Everything fell into place.
His professors were dismayed by his decision. They were astonished that he
would sacrifice a promising medical career for such an absurd field as psychiatry.
The medical profession generally was contemptuous of psychiatry; they thought it
a lot of nonsense and considered the psychiatrist as peculiar as the patients he
treated. Jung characteristically held a firm position on his choice.
(Hall & Nordby, 1973, pp. 21-2)
Although historical biographical/autobiographical narratives vary in style, the primary
distinction between this type of text and the other classifications (expository or
argumentative/persuasive) is that the writer focuses on a historical sequence of events in the
person’s life. Further, the author may choose salient events that point towards the reason this
particular person is remembered, or, if autobiographical, the authors may choose events that were
critical to who he or she has become. Unlike the events related in literary texts, such descriptions
of historical and biographical/autobiographical events are recounted as factual episodes. Thus,
the intent of this type of text is to inform the reader through a narrative of true events about
significant outcomes in the history of the discipline, whether it is psychology, sociology, art,
history, biology, or botany, or about a significant impact the events have had on the lives of
prominent individuals. Therefore, the task of the reader is to analyze these significant events and
to infer the historical and conceptual link to the discipline or individual if not explicitly stated.
Rhetorical features. For the TOEFL 2000 reading test, we recommend that the top-level
rhetorical patterns of texts be taken into account in addition to their pragmatic intent. Common
rhetorical classifications are taught in many college freshman composition classes (Axelrod &
Cooper, 1996; D’Angelo, 1980; Hale et al., 1996) and in content area reading classes (Vacca &
Vacca, 1996). Although writers often use a variety of rhetorical types throughout extensive
pieces of text, they also employ specific rhetorical patterns, particularly in expository passages, to
convey an overall theme or main point in shorter passages. Coupled with the use of cohesive ties
to provide transitions between supporting points, writers selectively use rhetorical types to shape
ideas and present a cogent point to the reader.
Rhetorical patterns from TOEFL 2000 Framework: A Working Paper (Jamieson et al., 1999)
are listed in Appendix D of that document. The following is an elaboration of these patterns for
the reading section of the TOEFL 2000 test using authentic academic reading texts as examples.
Definition – Writers use definitions to explicate the meanings of terms. When using technical
terminology, writers may employ a simple definition, which may be a phrase or sentence; readers
may utilize context clues to determine such definitions. However, as a rhetorical pattern, writers
use extended definitions to provide full descriptions of concepts. The intent of the entire passage
is to describe unfamiliar terminology, elaborate on terms specific to the discipline, and clarify
specific uses of the terminology. The following text is an example of an extended definition for
the concept of “pollutant” in a college biology lab manual.
A pollutant that reduces a certain population at any step of this food chain will
have a similar effect on all other levels of the food chain because all steps of the
chain are linked. Thus, water polluted with chemicals such as herbicides that kill
algae will decrease populations of zooplankton that eat the algae; this ultimately
affects populations of fish and humans that are part of the same food chain. A
pollutant is any physical or chemical agent that decreases the aesthetic value,
economic productivity, or health of the biosphere. There are many kinds of
pollutants, such as noise, chemicals, radiation, and heat.
(Vodopich & Moore, 1996, p. 161)
Illustration – To explicate a concept, a writer may choose to provide examples, a short
anecdote, or a familiar description so that vague or abstract concepts are fully described in
concrete terms. Cohesive ties that mark this type of rhetorical pattern include “for example” and
“to illustrate.” The following example about jet lag illustrates the concept of circadian rhythms.
Under normal conditions, circadian rhythms pose no special problems. Most
people try to schedule their activities to coincide with their personal highs and
lows. Unfortunately, though, there are circumstances under which circadian
rhythms may get badly out of phase with our daily activities.
The first of these occurs as a result of modern travel – especially by jet plane.
When individuals cross time zones, they have experienced considerable difficulty
n adjusting to their new location. The reason for this is clear: Their internal
biological clocks are calling for one level or type of activity, while the external
world is calling for another one. For example, consider a traveler who flies from
New York to London. She departs at 8:00 p.m. and, after a six-hour flight,
arrives in London at what her body perceives to be 2:00 a.m. She is very tired
and would like to sleep, but in London it is now 7:00 a.m. The day is just shifting
into high gear, people are eating their breakfast, and it is broad daylight. The
result: Our traveler feels awful. Yes, she can cope – and a few cups of coffee or
tea help her get revved up. Yet, if she is like most people, she will soon feel a
return of fatigue and will experience many unpleasant sensations as the day
progresses. Then, to make matters worse, when she gets into bed at 10:00 p.m.
London time, her body has not yet reset its biological clock; to her body it is
something like 5:00 p.m., a time when she usually feels especially active. The
result: She has trouble falling asleep and experiences even more discomfort.
(Baron, 1992, pp. 132-3)
Classification – This rhetorical pattern groups several items together according to similar
features or principles. Writers employ a classification scheme in order to indicate how discrete
items belong to a larger group. Readers, in turn, need to note such groupings. Cohesive markers
which indicate that the writer is generating such a pattern include transitions marking
membership in a class. Textbooks may include classification charts, hierarchical arrays, or tree
diagrams that accompany the text passage to further clarify which items belong to a specific class
and which terms provide the general category name. The following passage illustrates this
rhetorical text structure.
Compound Pinnate Leaf
Leaves typically consist of a blade and a petiole. The petiole attaches the leaf
blade to the stem. Simple leaves have one blade connected to the petiole,
whereas compound leaves have several leaflets sharing one petiole. Palmate
leaflets of a compound leaf arise from a central area, as your fingers arise from
your palm. Pinnate leaflets arise in rows along a central midline.
Leaves are also classified according to their venation (i.e., arrangement of
veins). Parallel veins extend the entire length of the leaf with little or no
crosslinking. Pinnately veined leaves have one major vein (i.e., a midrib) from
which other veins branch. Palmately veined leaves have several main veins
each having branches. Veins of vascular tissue in leaves are continuous with
vascular bundles in stems.
(Vodopich & Moore, 1996, pp. 270-1)
Comparison/Contrast – Writers employ this rhetorical pattern to designate distinctions
among concepts, particularly regarding their similarity or dissimilarity. Cohesive markers such as
“similar,” “compared to,” “in contrast to,” or “different from” designate whether concepts are
being compared or contrasted. The following passage illustrates this rhetorical pattern.
This automatic processing involves the performance of activities with relatively
little conscious awareness. Such processing seems to make little demand on our
attentional capacity. Thus, several activities, each under automatic control, can
occur at the same time. . . . You engage in automatic processing when you drive
your car and listen to the radio at the same time. Automatic processing with
respect to a given activity tends to develop with practice, as the components of
the activity become well learned and associated with specific stimulus conditions.
In contrast, controlled processing involves more effortful and conscious
control of behavior. While it is occurring, you direct careful attention to the task
at hand and concentrate on it. Obviously, this type of processing does consume
significant attentional capacity. As a result, only one task requiring controlled
processing can usually be performed at a time.
Research on the nature of automatic and controlled processing suggests that
they differ in several aspects. First, as you might guess, behaviors that have come
under the control of automatic processing are performed more quickly and with
less effort than ones that require controlled processing. . . .
(Baron, 1992, pp. 134-5)
Cause/Effect – When writers wish to explain why something happened, or the effects of
something, they use a cause/effect pattern. Cohesive markers such as “as a result,” “the effect
of,” “because,” and “consequently” are often indicators of this pattern. When reading cause/effect
texts, readers need to analyze the causes and effects in relation to the overall point.
Another, especially disturbing type of sleep disorder is apnea. Persons suffering
from apnea actually stop breathing when they fall asleep. Needless to say, this
often causes them to wake up. Since this process can be repeated literally
hundreds of times during the night, apnea can seriously affect the health
of persons suffering from it.
(Baron, 1992, p. 143)
Problem/Solution – Similar to the pattern of cause/effect, writers utilizing a problem/solution
text pattern describe a specific problem or series of problems, then propose a solution, which is a
plausible, salutary effect on a course of action.
In eastern forests of Canada and the United States bacterial insecticides may be
one important answer to the problems of such forest insects as the budworm and
the gypsy moth. In 1960 both countries began field tests with a commercial
preparation of Bacillus thuringiensis. Some of the early results have been
encouraging. In Vermont, for example, the end results of bacterial control were
as good as those obtained with DDT. The main technical problem now is to find a
carrying solution that will stick the bacterial spores to the needles of the
evergreens. On crops this is not a problem – even a dust can be used. Bacterial
insecticides have already been tried on a wide variety of vegetables, especially in
(Carson, 1962, p. 255)
Analysis – Although many of the previously described text patterns employ an analysis of
concepts, this text pattern provides a specific critical review of the facets of a situation, event,
idea, or case. Writers utilizing a predominant text pattern of analysis provide an in-depth
coverage of specific aspects of a general topic. When writers also provide this analysis in light
of a framework of criteria, then the writer employs evidence for justifying a point of view.
In our own American experience, families used to live closer together, making it
easier for relatives to pitch in during pregnancy and the first months of a
newborn’s life. Women worked primarily in the home and were more available
to lend a hand to new mothers and to help them get accustomed to motherhood.
Families were larger, and older children were expected to aid in caring for
younger siblings, a role that prepared them for their future parenting roles.
(Clinton, 1996, p. 70)
So state and nation may often fail to coincide. Yet in ordinary English we use
them almost synonymously (along with “country” which probably should really
refer only to physical terrain and landscape). And we regard instances where
they do not coincide, like those described above, as anomalous. How did a legal
concept and a social concept, which certainly do not automatically coincide,
come to be so closely associated in our minds?
The reason is that even if a government is generally accepted as appropriate
to rule a state, it still faces the difficult task of ensuring people's acquiescence in
its laws. It is terribly expensive and counterproductive to enforce all laws, all the
time by overt force. Dwight Eisenhower could not simultaneously send troops to
all schools that were slow to desegregate! Rather, governments must persuade,
cajole, and only to a certain extent force people to obey the laws. Beyond even
this, governments generally want the people of a state to be more than just
passively obedient to the laws. They want them to contribute positively to the
state, as voters, soldiers, volunteer workers in cooperative enterprises. As the
modern state began to emerge in the eighteenth and nineteenth centuries, the
leaders of states discovered that they could develop a more enthusiastic citizenry
if they linked national identity to the state's boundaries. This meant that when the
government asked citizens to do something on behalf of the state (obey public
health laws, pay taxes, join the army), it was asking not just on its own behalf but
on behalf of the nation of which all felt themselves a part.
(Shively, 1997, pp. 120-1)
Evidence about the role that top-level organizing structures play in expository prose is
inconsistent. While most results in this line of research show some text-structure pattern as being
more difficult for readers, the patterns tend to vary from study to study (e.g., Carrell, 1992;
Freedle & Kostin, 1993; Horowitz, 1987; Meyer, 1987). Most studies do agree, however, that a
simple description of facts seems to be easiest to comprehend. Any synthesis of this research area
would need to account for L1/L2 differences in results.
We have recommended certain pragmatic and rhetorical features, given the test purpose and
the organizational framework we propose for the reading test. At this time we do not believe the
alternative types of text that we propose are likely to contribute to differences in difficulty of the
items associated with them. However, we think it is important for the sake of content
representativeness and content comparability to investigate ways in which pragmatic and
rhetorical features of parallel tests can be accounted for in test specifications.
Test Rubric. The matrix in Table 1 illustrates the relationships among types of text, reader
tasks, and reader purpose. Part A identifies the types of text we recommend for inclusion in the
TOEFL 2000 reading test defined in terms of the various pragmatic intentions and rhetorical
patterns. Part B identifies various tasks that a reader may perform in engaging an author’s writing.
The readers’ tasks are related to the four reading purposes that we have chosen to organize
the domain. Thus, the task of summarizing may be most compatible with the purpose of reading to
learn, the task of comparing/contrasting might most usefully be related to the purpose of reading to
integrate information across multiple texts, and so on. The reason for using these combinations of
reader tasks/types of texts is that they cover the major domain of reading as it is practiced in
formal college and university settings.
Given the various possible combinations between A and B in Table 1, we recommend that
these combinations be used to define the domain of reading for the TOEFL 2000 test. For
example, for reading to find information and reading for basic comprehension, readers would be
required to identify information and interpret texts (and, in some instances, non-prose text
information), characterized by various rhetorical patterns: define/describe/elaborate,
compare/contrast, present a problem and a solution, explain/justify, persuade, and narrate. For
reading to learn, readers might be required to summarize or define/describe/elaborate information
based on an author’s attempt to define, describe, etc. However, we also recognize that purpose
and reader task will require some flexibility; thus, reading purposes (in relation to task) are
separated by dashed lines.
We believe that these combinations of types of texts and tasks mesh nicely with the
framework we proposed in the previous section, i.e., the four reader purposes applicable in the
academic context: (a) reading to find information; (b) reading for basic comprehension;
(c) reading to learn; and (d) reading to integrate information across multiple texts.
Types of Texts and Tasks for the TOEFL 2000 Reading Test
A. Types of texts defined by pragmatic and rhetorical features
Reader Purpose: Reading to Find Information and Reading for Basic Comprehension
Reader Purpose: Reading to Learn
Reader Purpose: Reading to Integrate
Tasks to Assess the Four Types of Reading
The following sections describe the types of tasks that could be used in the TOEFL 2000 test
to assess the four purposes for reading: reading to find information, reading for basic
comprehension, reading to learn, and reading to integrate information across multiple texts.
Reading to find information. At its simplest level, reading to find information involves the
process of locating the information that is noted in a question and matching it to identical or
closely paraphrased corresponding information in a text. For example, a question might ask an
examinee to simply circle a particular word in a text or to circle every occurrence of a particular
word or phrase. It is unlikely that the TOEFL 2000 test would measure this type of reading
purpose in isolation, however, because the task and level of difficulty would be inappropriate for
the test purpose and the task would be unlikely to discriminate among TOEFL examinees unless
time is constrained so that automatic processes have to be engaged.
At present, this reading purpose is assessed in combination with reading for basic
comprehension. For example, in the question “Why did the Census Bureau revise the definition
of ‘urban’ in 1950?” the information from the text that needs to be located in order to answer the
question is: “. . . in 1950 the Census Bureau radically changed its definition of ‘urban’ . . .”
Locating this information would require the reading to find information purpose, and answering
the question would require reading for basic comprehension, which is discussed in more detail
below. Some of the current TOEFL reading tasks weigh more heavily on search processes while
some involve other comprehension processes to a greater extent.
Another type of reading to find information task might be the question type that asks
examinees to locate where in the text a particular type of information can be found. “Where-
stated” tasks might be classified as reading to find information tasks or reading for basic
comprehension tasks. Ultimately research will determine how particular tasks can best be
combined with reading purposes so as to produce the most useful information.
In the TOEFL 2000 test, reading to find information might also require skimming and scanning
a prose text or a non-prose document to find a discrete piece of information that can be easily
matched to the information requested through a literal match or a close paraphrase. An example
of this type of reading to find information might be: “Highlight the information in the bar graph that
names the era in which dinosaurs lived,” or a question requiring examinees to locate the particular
section of a longer text where a specific type of information would be found or to scan multiple
texts for a particular type of information.
Reading for basic comprehension. While reading to find information is an important
underlying skill of reading, and might sometimes be tested directly as an independent reading
skill, it will certainly be measured indirectly in connection with the second type of reading purpose
on the TOEFL 2000 reading test, i.e., reading for basic comprehension. As defined for the
purposes of this test, this type of reading requires readers not only to locate information in the
text by matching information from the question to the text but also to identify additional new
information in the text that answers the question. For example, in the illustrative question earlier,
after examinees have located the place in the text where the Census Bureau is mentioned in
conjunction with the year 1950, they still need to understand why the Census Bureau revised the
definition of “urban.” Questions testing this type of reading involve both the reader’s knowledge
of vocabulary and cohesion/coherence devices and the ability to identify and interpret facts.
The current TOEFL reading test has been extensively analyzed in terms of difficulty variables
(Freedle & Kostin, 1993; Sheehan, Ginther, and Schedl, in press). A preliminary analysis of the
reading comprehension tasks based on three variables identified by Kirsch and Mosenthal (1990),
— type of information, type of match, and plausibility of distractors
— indicates that many of the
variables that have been useful in predicting difficulty on other reading comprehension tasks will
also be useful for describing TOEFL assessment tasks.
In designing new reading to find information tasks and basic comprehension tasks for the
TOEFL 2000 test, we recommend that the current paper-and-pencil TOEFL reading texts and
tasks be expanded to include not only prose texts but also non-prose texts such as pictures,
diagrams, process schematics, procedural schematics, matrix documents, locative documents
(including general reference and topographic maps), and quantitative documents (e.g., pie charts,
bar charts, line graphs, and timelines). Moreover, in addition to multiple-choice items, new forms
of response modes could be used. These might include: (a) open-ended responding with words,
phrases, or sentences; (b) point-and-click on the text, and (c) point-click-and-drag information
from one part of the screen to another. We would also recommend that a broader range of
rhetorical patterns be included. In addition to texts that define, describe, elaborate, illustrate,
explain, and justify (as current TOEFL texts do), we recommend including texts that compare and
contrast, persuade, and narrate. These recommendations would be in line with making the
TOEFL 2000 reading test more similar to the conditions under which college and graduate
students read in formal academic settings.
Reading to learn. Reading to learn requires readers to integrate and connect detailed
information from the text in a manner that is consistent with the rhetorical pattern of the text.
Here we are distinguishing the comprehension of individual main ideas or main points presented in
a text, which would be included under the “basic comprehension” purpose and the
comprehension of the whole text in the reading to learn purpose, which requires the integration of
these ideas into a coherent framework.
From a task perspective, item types that assess reading to learn would require examinees to
understand the rhetorical pattern of the text as well as to integrate the content information. For
this reading purpose, the issue of text material or text selection becomes especially important
because the type of text is more intimately connected to the information presented. It seems likely
that the types of texts that are appropriate for this purpose and the types of tasks associated with
them will increase the importance of pragmatic and rhetorical features and thereby increase the
likelihood that text readability will be associated with item difficulty.
See references in TOEFL 2000 Framework: A Working Paper (Jamieson et al., 1999) for a complete list of
We see two constructed-response alternatives, or two possible ways, for operationalizing
reading to learn tasks. One option is to set a time limit for reading a text, then take the text away
and ask examinees to summarize it in order to measure how well they have actually “learned.”
This would allow us to include reading rate as a variable related to difficulty. We would want to
prototype and research the advantages and disadvantages of examinees being allowed to take
notes during the reading period that they could then use in writing the summary. The other
possibility would not involve taking the text away but would still require examinees to integrate
information across the entire text and evolve a framework for interpretation.
We are not ruling out the possibility that reading to learn could be tested in short-answer,
semi-productive, or multiple-choice formats. Research should investigate a number of possible
item types. These alternatives need to be explored and research investigating different ways of
presenting the prompt needs to be carried out before decisions can be made about the new item
types. The research should also look at different language groups to determine whether both
alternatives provide equally good information.
Research on types of texts chosen for reading to learn tasks is also needed. We would like to
assess whether examinees recognize an author’s rhetorical pattern and can integrate the text
information in terms of it. It would be useful to have research that looks at texts varying in terms
of how strongly the organizational pattern or classification categories are signaled, i.e., ranging
from texts where all of the organizational categories are provided by the author to texts where the
organizational pattern is weakly signaled. The same texts could be used and the amount of
signaling varied to determine how much this changes the difficulty or the feasibility of the task.
We would also like to see research with texts which vary in terms of the types of texts outlined in
Finally, we are assuming that passages used for reading to learn tasks that require productive
spoken or written responses would also be used for other types of multiple-choice or short-answer
tasks so that the test time is used most efficiently. While the reading to learn purpose could be
tested in combination with other modalities, it may also be possible to test it alone. We
recommend that reading to learn tasks be associated with single passage texts and that multiple
texts be reserved for reading to integrate across multiple texts tasks.
Although the following example represents a task that would be scored as an integrated-skills
task, we do not mean to suggest that reading to learn tasks have to be connected to other skills. It
would be appropriate to test reading to learn as a measure of reading comprehension whether or
not it is integrated with other language skills. Research should investigate the feasibility and
construct relevance of a number of alternatives.
That said, one way to measure reading to learn might be to require readers to summarize
and/or recall text information. An example of a reading to learn task at this level involving a
comparison/contrast might be the following text, taken from a history book:
Compare the ideas of Locke and Hobbes. In what ways were Locke’s beliefs
influenced by Hobbes’ thinking? In what ways did their thinking differ?
The ideas of two English philosophers, Thomas Hobbes and John Locke, changed
the way Europeans viewed the individual's role in society. During the 1640s,
Hobbes witnessed the upheaval of a civil war in England. As a result, he became
convinced that if people were left alone without government they would
constantly fight among themselves. In 1651, he published his ideas in Leviathan.
Hobbes described life in a state of nature in which people had no government.
Such a life, he claimed, would be “nasty, brutish, and short.”
According to Hobbes, to escape the chaos of their natural state, people
entered into a contract in which they agreed to give up their freedom to a ruler
who guaranteed peace and order. The best government, Hobbes said, was one in
which the ruler had absolute power. He insisted that once people entered into
such a contract, they could not rebel, even if they thought the ruler was a tyrant.
Hobbes' ideas, therefore, supported the rule of absolute monarchs.
In 1690, John Locke published Two Treatises on Government in which he
agreed with Hobbes that the purpose of government was to create order in
society. He also saw government as a contract between the ruler and the ruled.
However, Locke's other ideas about government differed greatly from those of
Locke had a more optimistic view of human nature than Hobbes did. He
thought people were basically reasonable and would cooperate with one another.
Moreover, Locke argued that rulers could stay in power only as long as they had
the consent of those they governed. If a ruler were a tyrant, he or she had broken
the contract and the people then had the right to rebel.
Locke also presented other ideas that were important in the development of
democracy. He believed people had natural rights, including the right to life,
liberty, and property. Government was responsible for protecting these rights, he
said, but its power should be limited.
Examinees might be given such a task after having already answered several selected
response questions. They might be asked to read the question and take a minute to reread the text
before answering the question. Alternately, they might be asked to respond when the text is no
longer available. In this case they might be provided with several key points to address in their
response to help them remember the text content. For example, in the case of this particular text,
they might be asked to compare Hobbes’ and Locke’s respective views on human nature and
human rights and their views on the role of government and the role of rulers.
It is important to note here that in suggesting one possible task and sample text we do not
mean to preclude other possibilities. We recommend that extensive research be carried out,
particularly in the areas of reading to learn and reading to integrate multiple texts, before
decisions are made about operational test design. Reading to learn tasks could be designed as a
measure of reading comprehension alone, and reading to learn could also be associated with
spoken responses. Research should investigate all possibilities.
Research also needs to be carried out on scoring criteria and equatability of tasks across the
various types of texts listed in Table 1.
Reading to integrate information across multiple texts. Reading to integrate requires
readers to generate appropriate organizational frameworks which can then be used to relate
information from two different passages (or information sources). This integration may require
readers to compare/contrast, identify a problem and a solution, explain/justify, persuade, or narrate.
The selection of text material will be most critical in establishing the reading to integrate purpose.
Careful consideration will need to be given to the pragmatic and rhetorical features of texts used in
developing tasks for this purpose. In addition, multiple texts can be combined in a variety of ways:
a text passage may be followed by a diagram, a diagram might be followed by a text passage, a
text passage might be followed by a map, an outline might be followed by a figure, etc.
Reading to integrate tasks may also make use of language input from other modalities, so that
texts may be either spoken or written. From a task perspective, items might ask examinees to read
a longer text and summarize it in writing or speaking, or to read a text and complete diagrammatic
information (or create a diagram), or listen to one text and read a second text and then compare
and contrast them. Multiple texts for reading to integrate tasks might also consist of a prose text
and a non-prose document. Research is recommended in order to explore variables related to
As for texts used with productive reading to learn tasks, texts used for the reading to integrate
purpose could also be used to test other reading purposes with multiple-choice or short-answer
The following is an example of a reading to integrate question that combines a prose text and
Read the two texts below. Text 1 is a prose passage, and text 2 is a bar
graph. First, write a brief title for the bar graph. Next, write a summary of
the main information in the bar graph and explain how it is related to the
information in the passage.
Dimensions of the Energy Problem
Strictly speaking, no energy problem exists. The basic laws of physics
dictate that energy is conserved and can only be changed from one form to
another or into matter. Fuel, on the other hand, is the accumulation of matter and
therefore represents a store of energy. This energy is released in the form of
heat when the fuel is burned in chemical or nuclear reactions, which cannot be
reversed to regenerate the original fuel mass (at least not without the injection of
more energy than was originally released).
As a consequence, a fuel problem does exist. If the supply of fuel is finite,
not only will there be no energy supply when the fuel is exhausted, but also all
other processes that depend on it will cease. This will affect not only the obvious
energy consumers in the United States and the rest of the industrial West, but
even the most primitive societies, where the importance of oil-based fertilizer
supplies is growing.
Several factors combine to make the problem an urgent one. World
population is steadily increasing, which implies that the demand for energy will
also increase, although not necessarily in proportion. Social, economic, and
political pressure for economic expansion continues in industrialized countries.
This implies an increased energy input. The developing countries are becoming
aware that their economic position could be improved by increased energy
consumption, and they feel entitled to a larger share of the world’s energy
resources than they now receive. These pressures require that the world energy
supply be increased, particularly if the aspirations of some areas are to be met
without jeopardizing the living standards of others. Finally, it is now recognized
that the supply of the conventional fuels – coal, oil (petroleum), natural gas,
uranium, and fuel wood – is limited and insufficient to sustain present rates of
development for much longer. Although there may be debate about the exact
length of time available before the effects of a worldwide shortage become
apparent, there is agreement that such a shortage will occur. It is only a matter of
time; in the case of oil, for example, the debate is not about whether, but about
when oil production will peak.
Online Encyclopedia, 1995)
As in the case with reading-to-learn tasks, research will need to be carried out on task
comparability and scoring criteria.
Types of Response Formats
In a computer-based test, it is possible to include a much greater variety of response formats
than is possible in a paper-and-pencil exam. We recommend that the following response formats
be considered for use in the TOEFL 2000 reading test and integrated tasks:
• multiple choice
• open response formats
• click on a word, phrase, or sentence in the text or graphic
• click on and drag a word, phrase, or sentence in the text or graphic
• complete a chart, graph, or table
• create a chart, graph, or table
• extended written and/or spoken response
0 5 10 15 20 25 30
million barrels per day of oil production/consumption
Tasks using different types of response formats can be used to assess examinees’ abilities
with respect to the four purposes for reading, and some response formats may be better suited
than others to particular measurement objectives.
In redesigning the TOEFL reading test so that it includes measures of reading to find discrete
information, reading for basic comprehension, reading to learn, and reading to integrate
information across texts, we suggest that readers first be presented with one or more passages and
asked a series of reading to find information and reading for basic comprehension questions
representing a range of difficulty.
We suggest that these initial tasks include multiple-choice as well as open-ended machine-
scorable formats. They might include many of the formats that are used on the current paper-and-
pencil TOEFL reading test, as well as new computer-based TOEFL test formats such as the
insertion task format.
Once readers have completed the reading to find information and reading for basic
comprehension tasks, they could be given one or more reading to learn tasks that might involve
additional types of response formats. Reading to learn tasks might ask readers to organize
information from the text in a table or a schematic representing a rhetorical pattern.
To measure reading to integrate skills, the system might then present an icon which
readers would click on to see a second passage that would address the topic of the first
passage but in a different manner. At this point, readers could be asked to relate information
in the first passage to information in the second passage.
As with reading to learn tasks, readers’ performance on reading to integrate tasks would be
evaluated in terms of criteria that would permit exchangeability of scores across different tasks
varying in content, type of text, and reader purpose.
Linguistic Variables and Task Difficulty
Tasks for testing these various purposes for reading can be made more or less difficult by
varying certain linguistic parameters. We believe the following are an important subset:
• syntactic complexity,
• transition markers (cohesion),
• antecedent reference,
• modality (adverbs of attitude),
• amount of text, amount of time allowed,
• distances across text when cycle or integration is involved,
• competing linguistic distractors in the text environment,
• cohesion determiners (e.g., “We bought a camera. The lens was cracked.”), grammatical
relations as referents (back to subject or back to object), and
The linguistic variables are relevant to difficulty, however, only insofar as the task makes
them relevant, i.e., if a simple locating task is chosen to test a question of fact, the possible impact
of most of the linguistic variables is minimized. On the other hand, a task testing reading to
integrate information across multiple texts might be influenced by most of these variables.
4. Technological Considerations
The Role of Technology in Reading Comprehension
As a computer-based instrument, the TOEFL 2000 test will allow a greater range of
alternatives for measuring reading comprehension, particularly since the construct of reading as
defined in this paper encompasses the processing of both text and document materials inclusive of
electronic formats. As a construct, communicative competence in the 21st century involves the
processing of texts through the medium of computer technology. A broad definition of reading,
as well as writing, therefore includes the component of media literacy, which accounts for the
many visual and electronic forms of text that are present in our contemporary world. In fact, the
International Reading Association and the National Council of Teachers of English (NCTE &
IRA, 1996) recently included media literacy – specifically, “viewing” and “representing
language” – with the more traditional modalities of reading, writing, listening, and speaking in
their definition of the English language arts.
In the description of the construct of reading presented in this paper, each of the four reading
purposes aptly applies to electronic environments as well as to hard copy texts. In considering
reading in electronic environments, a particularly relevant dimension of the construct of reading
comprehension applies to the description of reading to integrate information across texts, since
this type of reading is defined through the use of multiple texts. When working online, readers
often use multiple screens of information, use texts plus graphics, and work with both electronic
and hard copy formats. Therefore, when working with multiple texts, readers need to be able to
infer which information generalizes across texts, which information from one text helps elaborate
or explain information from another text, and which sections of texts correspond to graphics on
the screen. An example of an authentic reading task would be reading several different online
texts in order to research a topic for a college paper.
In academic environments, information is disseminated in both paper and electronic formats,
and the recent reliance on the use of technology for both teaching and learning is increasing the
need for students to become proficient users of computer technology in accessing and reading
information. As information becomes more prevalent in electronic environments, users will have
greater choice in accessing texts, journals, and documents. In addition, as academic institutions
increasingly use computer technology to deliver instruction and for course requirements, students
will be expected to learn and be able to use a greater variety of computer environments.
Examples already include distance learning, online student registration, specialized Web sites
for programs and courses, e-mail for communicating with professors and students, and full libraries
of online information.
In many college settings, professors have their own Web sites, and students are expected to be
able to access course syllabi, course materials for lectures and activities, and other pertinent
course materials. Hansen and Willut (1998) provide an overview of the technologies that students
will be expected to use in the year 2000; however, their estimates of the use of technology
conservatively project that “courses using computer-based approaches to deliver the core
educational experiences will be in the minority in the year 2000” (p. 13). Recent accreditation
standards for state, regional, and national agencies now require the infusion of technology in all
coursework and require student proficiencies in computer technology.
Access to specialized Web sites, online libraries, bulletin boards, and e-mail provide reading
environments quite different from the environments of hard copy text and require a variety of user
proficiencies unique to the computer world. For instance, users need to be familiar with how to
read and use menus, icons, and navigational aids on the screen and need to have basic proficiency
in typing on a keyboard. Future computer designs that include voice recognition may reduce the
need for the user to have such computer proficiencies. Touch screens that eliminate the need for
using a cursor to make a selection may also reduce the need for users to be able to interact
through a keyboard or mouse.
The Reading Comprehension Interface
For the reading comprehension portion of the TOEFL 2000 test, we strongly recommend that
a reader-friendly interface be developed. In addition to single passage texts, the reading test is
likely to include multiple prose texts and texts integrated with pictures, maps, charts, graphs,
diagrams, and video. Because the experience of reading online is generally less appealing than
that of reading paper copy, every effort should be made to investigate and create the most user-
friendly interface and presentation design possible. The types of online information we would
like to draw on for the TOEFL 2000 reading test are summarized as follows:
• Single text – a short paragraph or a longer text;
• Multiple texts – short and/or longer paragraphs;
• Single text plus graphic – a short or longer paragraph accompanied by a graphic; and
• Multiple texts plus graphic – short or longer paragraphs accompanied by a graphic.
The types of graphics that could be utilized are as follows:
• Drawing – line drawing or schematic;
• Picture – picture to illustrate a text passage;
• Map – geographical, population, weather, or other type of map;
• Chart, table, or graph – chart, table, or graph that illustrates data; and
• Video – (if the technology supports such a capability) – video is a recommended source,
particularly when the video directly relates to the topics in a hard copy text.
In particular, we recommend the inclusion of navigational aids and icons.
Due to specific factors that influence the reading process, whether the task is in a hard copy or
electronic environment, the interface for the reading portion of the TOEFL 2000 test should have
the following design features:
• Clarity in resolution – The best state-of-the-art resolution features should be employed.
• Full text viewing – Examinees should be able to choose an icon that allows them to view
the entire text.
• Scrolling of text – So that test takers are able to access all the text, a scroll feature needs
to be included when the text is long.
• Magnification of the text – Examinees should be able to control a zoom feature to
magnify the entire text or a portion of a lengthy text.
• Speed of access – The TOEFL 2000 test should use state-of-the-art computers that
respond quickly when the user clicks on an option.
• Multiple text viewing – When readers are presented with multiple texts and/or graphics,
they should be able to switch from one text to the other very quickly and efficiently so
that the multiple texts are displayed without disrupting comprehension.
5. Research Agenda
The goals of the research agenda for the TOEFL 2000 test are to develop a test that is not
only reliable but also valid and fair. The design of a new test offers opportunities for integrating
test design and research that will contribute greatly to the validity of the test.
Test validation has been conceptualized in many ways, but at its heart it can be seen as “an
integrated evaluative judgment of the degree to which empirical evidence and theoretical
rationales support the adequacy and appropriateness of inferences and actions based on test scores
or other modes of assessment” (Messick, 1989, p. 13). Validity is a multifaceted concept.
Messick (1995) describes six aspects of validity – content, substantive, structural,
generalizability, external, and consequential – that can serve as criteria or standards for
measurement. Some of these aspects reflect traditional concerns such as whether content is
relevant to and representative of the domain, the generalizability of scores over tasks and across
time and among raters, and the relationship of scores to external measures and behaviors that are
or are not construct relevant. Other aspects, particularly substance and consequence, have
received greater emphasis in recent years than in the past.
The substantive aspect of validity concerns construct representation (Embretson, 1983), a
theoretical description of the information processes, strategies, and knowledge stores that underlie
task performance. With respect to construct representation, assessments can be described from
either the task perspective (What are the features of the task?) or the examinee perspective (What
processes, skills, strategies and knowledge do people use to solve problems?). Tasks, of course,
can be described in many different ways. By introducing a criterion, the relationship between
task features and difficulty, a distinction between critical and incidental task features can be
supported. The substantive aspect of construct validity is central to the TOEFL 2000 framework
because we expect to base test design on an understanding of task difficulty. A basic assumption
of this approach, one that needs to be substantiated through research, is that the features of the
task and examinee processes are interdependent (Chapelle, Grabe, & Berns, 1997). It should be
noted that construct representation has two distinct meanings. One, representativeness, relates to
coverage of a domain: Does the assessment represent a reasonable selection of content and
processes typical of the domain? The other, representation, refers to a psychological model of the
task in terms of features, knowledge, and processes involved in completing the task (Messick,
The consequential aspects of validity concern how test scores are used and the intended and
unintended consequences of these uses. One of the motivations for revising the TOEFL test is the
potential for positive washback; that is, a redesign of the test may have a positive effect on
teaching practice. However, a potential unintended negative consequence is reduced access to
higher education because of an increase in the cost of the test. Consequences such as these need
to be anticipated and their impact evaluated as part of the research process.
Because there are so many facets to validation, it is not an activity that should occur at the
end of the test design process. It should be an integral part of the process. The decisions that are
made in designing a test and the basis for these decisions all become evidence relevant to test
validity. Nevertheless, there are sequential constraints that need to be taken into account as to the
kinds of validity evidence that can be collected at various stages of test design.
The sequential nature of the test design process offers an organizing frame for laying out the
issues that can be addressed at each stage as well as one of a number of criteria for prioritizing
research. The stages of test design might be characterized as construct identification, prototyping,
pilot testing, and field testing. A brief description of each of these stages follows:
1. Construct identification – The construct to be assessed should be articulated, and the
ways that the construct might be operationalized should be proposed.
2. Prototyping – Prototypes of possible tasks and associated scoring rubrics should be
developed and evaluated. At the same time, research on tools to support the development
of tasks can be initiated, and preliminary consideration should be given to psychometric
3. Pilot testing – Those tasks that are thought to have potential for operational use are pilot
tested on small samples to identify potential problems and to clarify their relationship to
the constructs that are the objectives of the assessment. The outcome of this phase should
be preliminary specifications for task types and the manner in which test sections might
be assembled for testing in the next phase.
4. Field testing – Tasks that are deemed to be strong contenders for operational use are
assembled into possible operational sections and pretested on samples large enough to
support rigorous psychometric evaluation. The function of this stage is to finalize the
blueprint for item development and test assembly for operational pretests.
At each of these stages, operational, psychometric, and construct-related issues can be
addressed. Operational concerns focus on the feasibility of the proposed assessment. Questions
such as what kinds of hardware and software are needed, how much time and effort are required
to develop and score tasks, how much time the assessment requires, and whether the cost of the
assessment is reasonable, need to be answered. Psychometric issues include evaluating the
statistical properties of tasks and the appropriateness of different methods of scaling tasks and
generating scores. Finally, construct validation involves documenting that the tasks assess
construct-relevant rather than construct-irrelevant processes and skills.
Although the above framework suggests sequential stages, the iterative nature of the test
design process should be emphasized. For example, one would expect that the results of pilot
testing might dictate revisions to prototyped tasks. At any point in the design process, however,
the specific questions that can be answered are constrained by the number of task exemplars that
exist and the number of participants we want to involve in the evaluation.
With respect to the research agenda for the TOEFL 2000 reading comprehension test, some of
the questions that need to be answered at each stage of the design process are discussed below.
While some of the issues described are specific to reading assessment, others are more general
and also apply to the other modalities or to integrated tasks.
The goal of this stage is to produce documents that describe the constructs to be assessed,
indicate how they might be operationalized, and clarify the expected uses of the test. Although
the research that can be initiated at this stage is constrained because examples of potential tasks
do not exist, important topics that can be investigated include the following:
1. Literature reviews of theoretical descriptions of competency and empirical research
supporting these views should provide a basis for articulating the constructs to be assessed.
2. Research on how the test is currently used and the characteristics of the population taking
the test would be useful for documenting the appropriateness of the test for the kinds of
inferences users wish to make, for planning score reports, and for later evaluation of the
consequential impact of test revisions. For example, one question that could be addressed
is, what strengths and weaknesses do L2 students have, and how do these relate to
success or failure in academic settings?
3. Many research issues about the characteristics of texts for reading assessments have been
identified. These issues include identifying the range of text features characteristic of
academic texts, assessing the contribution of various features to text difficulty, providing
empirical support for the test blueprint, and developing tools to support efficient item
development. Research on the above topics would be facilitated by a systematic
collection of samples of text materials used by students in broad major fields and at
various types of institutions or identification of an existing corpus that could be used to
answer a number of questions about the range and characteristics of materials that
students encounter. The need for the following research activities is anticipated:
a) Analyze the rhetorical and linguistic features of authentic texts typical of the
domain of academic reading. This will provide information about the range of
registers most commonly involved in academic discourse. (See, for example,
Biber, 1986.) This information is needed to guide the selection of types of texts
that will provide appropriate content representativeness for the TOEFL 2000 test.
b) Choose a system for categorizing texts. An empirically supported multidimensional
system would be desirable.
c) Develop alternative measures of text complexity based on syntactic complexity
and/or informational density. While readability features of texts used in reading
tests in the past have not been shown to be directly related to item difficulty, it
seems to us to be possible that reading text characteristics may be more directly
related to task difficulty for the new types of texts and tasks we have proposed for
the TOEFL 2000 reading test.
d) Explore whether natural language processing (NLP) tools can be developed to
assess text difficulty and to tag text features for test development.
e) Determine what other aspects of vocabulary in addition to word frequency
contribute to difficulty.
f) Analyze the kinds of documents that students must understand and their
co-occurrence patterns with different types of texts.
g) Determine how various types of signaling influence the comprehension of text
4. Research on the variables controlling the difficulty of current TOEFL tasks should be
initiated and research on the variables controlling the difficulty of potential reading-to-learn
and reading-to-integrate tasks should be investigated as prototype items are developed.
5. Investigate the effects of controlling the reading time allowed and/or measuring reading
Exemplars of potential item types and scoring systems will be developed and authored for
computer administration. Alternative versions of tasks may need to be developed so that
questions about which formats best assess relevant constructs can be answered in subsequent
phases. The following research and development activities are needed.
1. Explore how to operationalize the new “Reading to Learn” and “Reading to Integrate”
constructs proposed in this reading assessment framework. Develop task exemplars that
vary the nature of the prompts, stimulus exposure time, memory demands of the task, and
the response formats.
2. Propose scoring systems for the tasks.
3. To ensure optimal screen design, summarize prior research on the impact of presentation
features on reading comprehension and consider this research in designing the interface.
4. Develop tutorials instructing examinees how to complete the prototype tasks.
5. As exemplars are developed, conduct preliminary observational studies of small groups of
L1 and L2 students to evaluate the tutorial and interface.
A small number of exemplars for each prototype task will be developed and pilot tested on
small samples of examinees to evaluate which tasks are most suitable for the population and
which conditions of administration are appropriate. With respect to the pilot testing, the following
activities will be necessary.
1. Develop and evaluate scoring systems for constructed response items.
2. In addition to conducting a preliminary evaluation of the statistical properties of the item
types, initiate construct validation research addressing a number of issues. This research
is needed to demonstrate that the skills and abilities that account for performance on the
test are related to the construct being assessed and not to other sources of individual
differences. The outcome of this phase should be tentative or alternative blueprints for
the assessment. The types of investigations that can be initiated at this point include
a) User acceptance – Do the constituents who will use the test results believe that the
tasks are appropriate?
b) Concurrent validity – Is performance on the tasks related to other, concurrent
indicators of linguistic competence such as placement in remedial classes, need for
tutoring, teacher evaluations, or performance on more complex criterion tasks?
c) Construct representation – Is there evidence that the processes and strategies that test
takers use to answer questions are construct relevant?
d) Impact of test-taker characteristics – Some characteristics, such as how long test
takers have studied English or how long they have been in the United States, are
construct relevant. Others, such as computer experience, are not. What is the impact
of construct relevant and irrelevant factors on performance?
e) Native speakers – How do they perform on the tasks?
f) Factors that affect task difficulty – In particular, what is the nature of the relationship
between reader purpose and task difficulty?
Test sections composed of task types that have been found to have strong potential for
inclusion on the new test would be pretested on large samples in operational-like settings to
determine statistical properties, to implement and evaluate scoring systems, and to evaluate
psychometric models. Topics and questions that must be researched are summarized below.
1. Ways of enhancing scale interpretability should be explored. These might include:
a) scale descriptions linked to proficiency descriptions of tasks; and
b) normative information about performance by native speakers applying to different
types of programs (two-year colleges, four-year colleges, graduate schools; sciences,
social sciences, humanities programs). The latter approach also has implications for
test fairness in that it would provide information about the range of performance by
2. Additional construct validation research that requires larger samples could be carried out.
This includes studies of:
a) Convergent and discriminant validity – Is performance on the tasks more highly
correlated with performance on other tasks that are thought to assess the same
construct and less correlated with performance on tasks thought to assess other
b) Construct representation – For example, is there evidence that the difficulty of the
tasks can be manipulated in systematic ways that are relevant to the construct?
c) Experience – Because it is expected that short-term experience will have an impact on
this type of test, it would be useful to document how L2 mastery, as measured by the
potential item types, changes over time and what factors facilitate or inhibit
d) Subpopulation differences – Studies should be carried out to determine if there are
sources of construct-irrelevant variance, such as gender, affecting performance on the
3. The appropriateness of different psychometric models for scoring and scaling items will
need to be evaluated.
The outcome of the test design process will be a blueprint for test development that will
specify in detail the numbers and types of tasks to be developed, the kinds of materials to be used,
the task features that should be manipulated to control task difficulty, the composition of test
sections, and scoring rubrics.
Many of the issues addressed in previous stages will need further exploration or replication
under operational conditions. In addition, there are some questions, such as the consequential
effects of the test, that cannot be answered until the test has been operational for a period of time.
6. A Better Reading Test
The existing TOEFL reading test is a good test by current standards. Designing a new test
that will constitute a significant improvement over the current one is therefore challenging. The
current TOEFL reading test has never been explicitly linked to any particular theory of reading or
reading construct, however. The most obvious improvement we can make over the current test,
then, is to articulate the construct we want to test and to link the proposed test design to that
construct. We believe that the proposed reader-purpose framework will allow for better
communication of the principles driving test design and test development, while at the same time
make it possible to overlay complementary frameworks driven by processing and task-based
views of reading.
Moreover, we believe that the reader-purpose framework expands the construct tested beyond
the level of finding discrete information and basic comprehension measured in the current TOEFL
reading test. It seems likely that tasks and texts intended to measure reading to learn and reading
to integrate information across texts will also tap into the rhetorical and discourse structure of the
texts in ways that make text readability more important than it is for reading to locate information
and reading for basic comprehension. In addition, the new types of texts and tasks associated
with reading to learn and reading to integrate information across texts are likely to raise the
ceiling of the test to allow us to discriminate better at higher proficiency levels.
One of the goals of the TOEFL 2000 project is to identify variables contributing to task
difficulty in order to improve the interpretability of test scores. In other words, we hope to
provide some information about what examinees can and cannot do with the English language
rather than merely report what score they received on the TOEFL reading scale. Research can be
designed to confirm whether variables thought to influence task difficulty actually do so. This
will then allow us to provide useful information to test users and examinees about the strengths
and weaknesses of an examinee’s performance. It will also provide critical information to test
designers and test developers that will allow them to make the most efficient use of examinee
time and of items and to create particular types of items to target certain proficiencies and goals.
For example, we hope to increase the importance of variables associated with the type of
information requested and to decrease the weight of variables associated with the plausibility of
the distractors, thus increasing construct validity.
It is our belief that improved communication about what is being tested and improved
interpretability of examinee performance will have positive washback effects as well. Assessing
reading with various types of prose and non-prose documents, and with writing, listening, and
speaking in the integrated skills tasks, broadens the construct being measured to more realistically
represent real-world language needs and language use. Research can be designed to investigate
washback effects on what examinees study and to determine whether the emphasis on
communicative learning increases once the new test is operational.
Finally, the proposed design is flexible enough that it does not preclude the introduction of
additional new item types at some future time. We believe that new item types would be likely to
include components that contribute to task difficulty in similar ways to those identified at the time
the TOEFL 2000 test is introduced. Identifying and focusing on variables that contribute to
difficulty in the design of the new test thus facilitates the process of updating the test from time to
Adams, M. J. (1990). Beginning to read. Cambridge, MA: The MIT Press.
Akamatsu, N. (in press). The effects of first language orthographic features on word recognition
processing in English as a second language. Reading and Writing: An Interdisciplinary
Axelrod, R., & Cooper, C. R. (1996). The St. Martin’s guide to writing. New York: St. Martin’s
Barker, T. A., Torgesen, J. K., & Wagner, R. K. (1992). The role of orthographic processing skills
on five different reading tasks. Reading Research Quarterly, 27, 346-345.
Baron, R. A. (1992). Psychology. Needham Heights, MA: Simon & Schuster.
Bates, E., & MacWhinney, B. (1989). Functionalism and the competition model. In B.
MacWhinney and E. Bates (Eds.), The crosslinguistic study of sentence processing
(pp. 3-73). Cambridge: Cambridge University Press.
Beers, B. I. (1983). World history: patterns of civilization. Englewood Cliffs NJ: Prentice Hall.
Berman, R. (1986). A cross-linguistic perspective: Morphology and syntax. In P. Fletcher &
M.Garman (Eds.), Language acquisition: Studies in first language development (2nd ed.).
Cambridge: Cambridge University Press.
Bernhardt, E. B. (1987). Cognitive processes in L2: An examination of reading behaviors. In J.
Lantolf & A. Labarca (Ed.), Delaware Symposium on language studies: Research on
second language acquisition in the classroom setting. Norwood, NJ: Ablex.
Bernhardt, E. B. (1991). Reading development in a second language. Norwood, NJ: Ablex.
Biber, D. (1988). Variation across speech and writing. New York: Cambridge University Press.
Biber, D. (1992). On the complexity of discourse complexity: A multidimensional analysis.
Discourse Processes, 15, 133-163.
Biber, D. (1993). An analytical framework for register studies. In D. Biber & E. Finegan (Eds.),
Sociolinguistic perspectives on register (pp. 31-56). New York: Oxford University Press.
Biber, D. (1995). Dimensions of register variation. New York: Cambridge University Press.
Bowey, J. A., & Francis, J. (1991). Phonological analysis as a function of age and exposure to
reading instruction. Applied Psycholinguistics, 12, 91-121.
Brewer, W. F. (1980). Literary theory, rhetoric, and stylistics: Implications for psychology. In R.
J. Spiro, B. C. Bruce, & W. F. Brewer (Eds.), Theoretical issues in reading comprehension
(pp. 221-239). Hillsdale, NJ: Erlbaum.
Britton, B. K., & Black, J. B. (1985). Understanding expository text. Hillsdale, NJ: Erlbaum.
Brown, T. & Haynes, M. (1985). Literacy background and reading development in a second
language. In T. H. Carr (Ed.), The development of reading skills (pp. 19-34). San Francisco,
Carpenter, P., Mikaye, A., & Just, M. (1994). Working memory constraints in comprehension:
Evidence from individual differences, aphasia, and aging. In M. Gernsbacher (Ed.), Handbook
of psycholinguistics. San Diego, CA: Academic Press.
Carrell, P. (1992). Awareness of text structure: Effects on recall. Language Learning, 42, 1-20.
Carson, R. (1962). Silent spring. Greenwich, CT: Fawcett Publications.
Carver, R. P. (1997). Reading for one second, one minute, or one year from the perspective of
rauding theory. Scientific Studies of Reading, 1 (1), 3-43.
Chapelle, C., Grabe, W., & Berns, M. (1997). Communicative language proficiency: Definition
and implications for TOEFL 2000. (TOEFL Monograph Series Report No. 10). Princeton,
NJ: Educational Testing Service
Chen, R. S., & Vellutino, F. (1997). Prediction of reading ability: A cross-validation study of the
simple view of reading. Journal of Literacy Research, 29, 1-24.