测试文本长度是阅读测试开发者需要考虑的重要因素。虽然学界对文本长度的重要性已达成共识，但对它的研究还远不够，正如Alderson（2000：108）所提醒的，文本长度值得深入研究，但它远未得到足够的重视。学者对测试文本长度的看法莫衷一是，有关实证研究结果相互矛盾。本研究将深入探究测试文本长度对英语阅读的影响，探讨何种长度的测试文本更能可靠、有效地评估考生的英语阅读能力。

本研究聚焦测试文本长度，即阅读测试中阅读文本所含单词总数，分为短篇文本（约350词）、中篇文本（约700词）、长篇文本（约1050词）3个层次，研究者在英美正式出版物或网站上选取文本特征相近，但长度不同的4篇文本（2篇长度分别为356和383词短篇、1篇长度为700词中篇和和1篇长度为1007词的长篇），并基于以上文本命制阅读理解试题。北京某高校的129名（分为高、中、低3个水平组，各水平组人数相当）非英语专业二年级学生参与了正式研究。本研究采取量化实验、有声思维、半结构访谈等量化与质性研究相结合的方式，从结果和过程的两个层面收集测试文本长度对英语阅读理解影响的数据和证据，获得以下发现：

（1）测试文本长度对考生阅读理解测试的成绩影响显著。具体而言，考生在长篇文本中得分显著低于中篇和短篇文本，但在中篇和短篇文本中的得分之间无显著差异。在各子技能得分率方面，除理解篇章大意子技能外，文本长度对其它6个子技能影响显著，即理解段落大意、理解明确表达的概念和细节、理解隐含表达的概念和细节、理解作者的态度和观点、根据上下文猜测词和短语的意思和理解句子、段落、篇章层面的相互关系，其影响呈现非线性、多层次的特征。在考生的阅读水平与其在不同长度文本中的得分相关性方面，就单篇文本而言，考生在短篇文本中的相关性不显著，而在中篇或长篇文本中的相关性显著；就不同长度文本搭配而言，考生在各种长度文本搭配中的得分与其阅读水平的相关度均高于单一长度文本。

（2）测试文本长度对受试者阅读理解过程有显著影响。结果显示，短篇文本所激发的低层认知加工频数多于高层认知，即受试者阅读短篇文本时，会偏重语言，对文本的词句层面进行精细化理解。而在阅读中、长篇文本时，受试者所采用的高层认知加工频数则要多于低层认知，即受试者在处理中、长篇文本时，会偏重意义，从整体上去理解语篇。同时，研究还显示，各子技能考查点在不同长度的测试文本中所激发的认知加工环节也有差异。除了在短篇文本中存在的应试技巧的现象外，受试者在中、长篇文本中即便处理局部命题时，也会把命题置于更为宏观的段落或整个篇章中去考虑，视野更加宽阔。随后的半结构访谈也显示在阅读不同长度的测试文本时，受试者的关注重点、阅读策略、阅读速度、难度感知等4个方面都有很大差异。就不同阅读水平的受试者而言，对其阅读过程分析显示，在短篇文本中，中、高水平的受试者无法充分发挥其在语码识别自动化和整体理解上的优势，而中、长篇文本则提供了发挥考生阅读能力，尤其是高阶阅读能力的空间，因此也就在一定程度上解释了受试者在中、长篇文本的得分与其阅读水平呈显著相关的原因。换言之，在很大程度上，短篇测试文本的区分力弱，而中、长篇测试文本更利于区分不同水平的受试者。

本研究的理论贡献在于深入探究了文本长度这一重要文本特征对阅读理解结果和过程的影响，对二语阅读测试中的文本长度选择提供了实证理据；通过对阅读认知过程的研究，基本明确了测试文本长度影响阅读的模式，从而解释了测试文本长度是如何影响理解的这个关键问题；基于本研究结果，结合相关研究，研究者还提出“测试文本长度阈值”假说：只有当测试文本长度超过一定数值，对读者的认知加工能力产生压力时，才会激发具有本质性差别的认知加工过程，理解难度发生显著变化。这一假说的提出证实了阅读测试引入达到一定长度文本的必要性，这是对文本长度研究领域一个重要的补充。在实践方面，本研究解决了困扰测试开发者的“文本长度怎么分”、“如何选择”等难题；为优化阅读理解测试评价手段提供了参考方案。本研究还发现较长文本能促使受试者启动多种阅读方式，为破解考查阅读流畅性的问题提供了可能性方案，即选择达到一定长度阅读文本。此外，本研究为大学英语四、六级考试在文本长度上的改革提供了来自阅读过程和阅读结果的理据支持，验证了改革的合理性与必要性。

In reading comprehension tests, text length, as one important text feature, needs to be considered by every test developer. Although its crucial role in reading tests has long been recognized, the relevant research has been far from sufficient, as reminded by Alderson, “Text length is a surprisingly under-researched area” (2000:108). Unfortunately, this situation has not been improved up to now: researchers have different views on the effect of test text length, and the empirical researches have arrived at conflicting results. The aim of this study is to investigate the effect of test text length on the test takers’ English reading comprehension in order to find out which length could provide more reliable and effective assessment for EFL college students’ reading ability.

In this study, test text length, as an independent variable, is operationalized as the number of words in one reading passage and is classified into three levels: short (about 350 words), medium (about 700 words) and long (about 1050 words). Two short texts (356 and 383 words), a medium length text (700 words) and a long text (1070 words), which varies in length but are similar in other text features, were selected from the American or British formal publications or websites. Based on them, a reading comprehension test was developed. Participants were 129 non-English major sophomores (made up of the students from 3 reading proficiency levels) in a key university in Beijing. Quantitative and qualitative data were collected from a reading test, thinking aloud tasks and semi-structured interviews.

The main research results are as follows.

(1) The test text length has a significant effect on the test takers’ reading performance. To be specific, the test takers’ overall scores in the long text are significantly lower than those in the short and medium length texts, though the scores in medium-length text are insignificantly lower than those in the short texts. In terms of reading sub-skills, except the understanding the main idea of the passage, text length is found to have a significant effect on the test takers’ performance in other six sub-skills, namely understanding the main idea of the paragraphs, understanding the explicitly expressed ideas, understanding the author’s attitude and argument and understanding the relationship at the levels of sentences, paragraphs and the whole passage. The specific pattern of text length effect on each sub reading skill is shown to be non-linear and multileveled. With regards to the correlation between the test takers’ reading proficiency levels and their scores in testing texts of different length, the correlation is found to be insignificant between the test takers’ reading proficiency and their scores gotten from the short texts, however, the correlation is significant in case of the medium-length and long texts. And in case of combination, the correlation index is shown to be higher in the different-length texts assortment than that in one particular length text.

(2) Significant differences are also found in the reading process of test takers’ dealing with the different-length test texts. It indicates that short texts are likely to elicit more lower-level cognitive processes than higher-level ones. In other words, when reading short texts, the test takers tend to engage in lexical and syntactic levels rather than a discourse level processing. On the other hand, when reading medium or long texts, the test takers are inclined to approach the texts globally, activating more higher-level discourse processes. Meanwhile, the cognitive processes prompted by every sub-skill test item are shown to be qualitatively different in the three length texts. For one thing, the longer texts seem to preclude the text-wise strategies to a large degree, that is, the test-wise strategies are only identified in the short texts instead of longer ones. For another, while ostensibly some test items are local in the longer texts, the test takers still tend to respond from a broader perspective and place the local propositions in a wider paragraph or text context to consider. The post semi-structured interview also reveals that the test takers’ primary concerns, reading strategies, reading speed and perceived difficulties were different among the three-length texts. With regards to the test takers of different reading proficiency levels, the reading cognitive process analysis indicates that in the short texts, medium and high-level readers’ advantages in lexical automaticity and discourse processing could not display themselves fully, while in the longer texts, the test-takers’ advantage in these aspects could do well and helps them outperform the lower-level group, which could explain why the significantly higher correlation between the test takers’ scores in longer texts and their reading proficiency. In other words, to some extent, the discrimination power of short texts is weak, while the medium-length and long texts could effectively distinguish test takers of different reading levels.

This study has some theoretical and practical contributions. Theoretically, it sets out to explore the test text length’s effect on the reading comprehension outcome and processes. The results could offer empirical evidence for the rational choice on text length in the reading comprehension tests. And the study has specified the way of test text length effect on reading cognitive process, which hence explains the critical question of how test text length affects reading. More importantly, based on the findings of the present and previous studies, the research has put forward a hypothesis called “Test Text Length Threshold”, that is, only when test text length reaches such a particular level that produces stress on cognitive processing capability could elicit a fundamentally different cognitive processing pattern and affect the comprehension difficulities significantly. This hypothesis not only offers reasonable explanations of the previous conflicting research results, but proves the necessity of introducing the texts whose lengths are above the threshold in the reading tests. Practically, the study provides a plan to optimize the present reading test methods, and helps to address the long-standing problems of “how to classify text length”, “how long the texts in the reading tests should be”. One of the findings that the longer texts would engage the test takers in reading in such a way to integrate different reading types suggests that it might be feasible to measure both reading fluency and reading accuracy by using longer test texts. In addition, the study offers evidence from the test takers’ performance and process to justify the reading text length reform in CET-4 and CET-6.