Newspapers have been a crucial of part people’s lives, with 52–85% of the people in different countries reading news more than once a day (Cabrera, 2020). Together with the development of technology, most aspects of our lives have been digitized, and the way we receive our daily news is certainly no exception. Researches have shown that people are giving up traditional, paper-based news and giving their favor to its digital, online counterparts (Cabrera, 2020; Pew Research Center, 2021a,b). In a recent research conducted on American news habit by Shearer (2021), 86% of United States adults got their news from a digital device (e.g., smartphone, computer, or tablet). These figures were significantly higher than television (68%), radio (50%) and print publications (32%) (Shearer, 2021). The main reason for this shift could be due to convenience. Compared to print newspapers or magazines, online publications are more easily accessible, could be read from anywhere and are supported by state-of-the-art technology which offers better reading experience. On top of that, most, if not all, online newspapers are eco- and reader-friendly, that is, they are free to read and do not harm any tree.

Due to the popularity of online newspapers, the ability to read and comprehend this type of publication has been viewed as a critical goal for second language learning, and the use of newspapers has always been emphasized in language teaching (Nation, 2013; Nation and Macalister, 2021). In fact, many language proficiency tests, like IELTS, have long incorporated newspaper and magazine articles in their reading component (Moore et al., 2011, 2015). As a result, it is crucial for English teachers and learners to be informed of the amount of vocabulary necessary to comprehend an online newspaper article, such a knowledge would strongly inform various decisions in goal setting, lesson planning and course book design. The aim of this manuscript is to offer an answer to such question, that is, to determine the number of words required to understand online newspaper or magazine articles.

The present study sets out to revisit Nation’s (2006) figures following the two major suggestions put forward by Schmitt et al. (2017): increasing sample size and employing up-to-date research methodology.

To re-examine the lexical profile of newspapers from the perspective of a larger sample size, the present study analyzed Davies’s (2016/2021) News on the Web (NOW) corpus, the largest corpus of English newspapers available. Besides the ultra-large sample size, the current study also employed the most comprehensive and up-to-date BNC/COCA wordlist (Nation, 2017). The word list contains twenty-five 1,000-word levels which reflects current English. In addition, the BNC/COCA is accompanied by four supplementary lists of proper nouns (Aaron, Greece, Grecian, Greenberry, and Waterloo…), marginal words (hm, huh, er, ah, and phew…), transparent compounds (aftershock, afterword, airbag, and powerboat…) and acronyms (PHD, UFO, and UDA…) (Nation, 2020), which allow more detailed analyses compared to the BNC list which is only accompanied by two supplementary lists of proper nouns and marginal words.

Moreover, the use of the BNC/COCA lists in lexical profiling research also contributes to the methodological shift in the field of vocabulary studies. As the most widely used vocabulary tests have long utilized the BNC/COCA lists as the source for their test items (McLean and Kramer, 2015; McLean et al., 2015; Webb et al., 2017), it would not take long for researches on most aspects of vocabulary knowledge and lexical development to follow. Therefore, it would be methodologically inconsistent to relate the lexical profile of a text based on the BNC lists to a study that reflected learners’ vocabulary knowledge of the BNC/COCA lists.

The study also responds to Webb’s (2021) call for more attention to the variation in lexical demands. In a recent manuscript, Webb (2021) expressed his concern that lexical profiling researches only “reflect the mean number of word families needed to reach a certain lexical coverage figure” and often ignore the fact that “each corpus is made up of a large number of texts, and there is likely to be a great deal of variation in the vocabulary of each text” (p. 286). It is true that we should not assume the same coverage to be reliably applied on different texts just because they belong to the same text genre, especially for newspapers. Researches have shown that the use of grammar and vocabulary could greatly vary among regions and generations (Chambers, 2000; Davies and Fuchs, 2015; Davies, 2021; Wan and Cowie, 2021). Therefore, it would be reasonable to hypothesize that the lexical demands of newspaper and magazine articles from different countries and periods of time bear certain degrees of distinction. The NOW corpus comprises data from online newspapers and magazines collected from twenty countries over a period of 11 years. The analysis of such corpus not only provides reliable figures on the lexical demand of newspapers, but also offers deep insights into the variation of the vocabulary knowledge required to comprehend English newspapers written in different countries and years.

In particular, the study seeks to answer the following questions:

The second and third columns of Table 1 present the coverage of each word level for the NOW corpus. The most frequent 1,000 word families in the BNC/COCA wordlist accounted for the greatest proportion of tokens, 72.48%. The coverage then dropped significantly to 10.27% at the second 1,000-word level. After the 2,000 word families level, the number of tokens as well as its coverage gradually decreased as the word frequency went down. Lower-frequency levels from the 5,000 level onward only accounted for less than 1% of the running words in the corpus, which generally highlighted the importance of high-frequency words to reading comprehension.

Another worth noting detail was the proportion of proper nouns, marginal words, transparent compounds and acronyms in the corpus. Proper nouns alone were found to make up of 3.91% of the tokens in the NOW corpus. The combined coverage of PN, MW, TC, and acronyms made up of 5.39% of the running words, which was close to the coverage provided by the third most frequent 3,000 word families in the BNC/COCA word list. These figures demonstrate the relative importance of being able to recognize and understand proper nouns as well as marginal words, transparent compounds and acronym.

The last two columns of Table 1 show the vocabulary knowledge needed to reach 95 and 98% coverage of online newspapers. The results from the analyses displayed two hypothesized scenarios: one assumed that all proper nouns, marginal words, transparent compounds and acronyms were easily understood or recognized, and one supposed that they were not. Since more than 5% of the tokens accounted by the four supplementary lists, it was more than certain that understanding 95% of the running words in online newspapers with the sheer knowledge of the twenty-five thousand word families in Nation’s (2017) BNC/COCA lists was impossible. However, if proper nouns, marginal words, transparent compounds and acronyms were assumed to be known, then the knowledge of 4,000 and 7,000 most frequent word families were necessary to achieve 95 and 98% coverage, respectively.

However, when looking at the NOW corpus from another angle, we could easily realize that the corpus was made up of newspaper and magazine articles from twenty different countries over a period of 11 years. Therefore, it may not be appropriate to judge the corpus’s lexical demand by its twelve billion tokens in combination, and the lexical profile of the corpus deserves a deeper investigation into its variation. Supplementary Appendix offers data for such analysis.

Supplementary Appendix provides data about the cumulative coverage of each sub-corpora including proper nouns, marginal words, transparent compounds and acronyms. Results from the analyses of the sub-corpora yielded interesting findings. Figure 1 is a graphic representation of Supplementary Appendix and offers visual support for the amount of words needed to achieve 85, 95, and 98% coverage for online newspapers in different countries. Due to space limitation, Figure 1 can only demonstrate a rough summary of the vocabulary demands of newspaper and magazine articles in different nations. Several “middle” numbers like 3,500 or 6,500 could be spotted in Figure 1, which did not necessarily mean that 3,500 or 6,500 word families were needed to understand 95 or 98% of the words in newspapers in certain countries. In fact, these “middle” figures signaled that a vocabulary knowledge of 3,000–4,000 or 6,000–7,000 word families was required for these coverage thresholds. This was due to the variation of vocabulary demands between different years or even between periods of time. Take Hong Kong as an example, although figures from 2012 to 2021 suggested that the 4,000 most frequent word families in the BNC/COCA lists were needed for 95% coverage, data from 2010 to 2011 showed us that it only took 3,000 word families to reach the same threshold.

It could be observed that 2,000 most frequent word families in the BNC/COCA wordlist covered 85–90% of the running words in online newspaper and magazine articles from all 20 countries. This highlighted the feasibility of using web-based newspapers as reading materials in English classes as well as the value of the BNC/COCA 2,000 to English learners. Although a one-size-fits-all threshold for 95 and 98% coverage could be said to be fictional, the data suggested that the vocabulary knowledge of the most frequent 3,000–4,000 word families in the BNC/COCA word list was necessary to gain 95% coverage. However, things became more complicated when it came to the 98% coverage. For some countries like Jamaica, New Zealand and Canada, the vocabulary knowledge at 6,000 level was enough to provide 98% coverage of web-based newspapers and magazines. Online newspaper and magazine articles from countries including Australia, Hong Kong, Ireland, Tanzania, and the United Kingdom seemed to be a little more demanding and required 6,000–7,000 for optimal comprehension. Online newspapers and magazines in the United States, South Africa, Singapore, Pakistan, Nigeria, Kenya, and Ghana needed the knowledge of the most frequent 7,000 word families to comfortably understand. For Malaysia and Sri Lanka, the required vocabulary knowledge for the 98% threshold was at 7,000–8,000 levels. Online newspapers and magazines written by publishers based in India and Philippines required readers to have a vocabulary knowledge at 8,000 level for optimal comprehension.

Bangladesh’ was the most unique country which required a range of vocabulary knowledge from 8,000 to 10,000 word families for 98% coverage. It was also worth noting that online newspapers and magazines in Bangladesh only needed a knowledge at 4,000 level for 95% coverage. The reason for the difference between the lexical demands for 95 and 98% coverage of Bangladesh’s newspapers might lie with the substantial proportion of words at the 13,000 level. Normally, lexical coverage at the 10,000 level and higher dropped below 0.2 or even 0.1 for most country sub-corpora in the NOW corpus. The same also went for the Bangladesh sub-corpus when the tenth, eleventh and twelfth 1,000-word level only represented the coverage of less than 0.2%. However, the coverage of the 13,000 level went up to 0.45–0.68% for the Bangladesh sub-corpora, which were relatively high for such a low-frequency word-level.

In answer to the first research questions, the vocabulary knowledge of the most frequent 4,000 families in the BNC/COCA list plus proper nouns, marginal words, transparent compounds and acronyms would reliably provide 95% coverage of the NOW corpus. This means that if 95% coverage was assumed to be sufficient for reasonable comprehension, then learning the most frequent 4,000 word families would be the ultimate learning goal for English learners in ESL and EFL contexts. As Laufer and Ravenhorst-Kalovski (2010) and Schmitt et al. (2011) suggested, 95% coverage of a text would result in acceptable comprehension, however, they also pointed out that the degree of comprehension at the 95% threshold is not really reliable and that 98% was truly the threshold for unsupported comprehension. If 98% were supposed to be the necessary threshold for text comprehension, then a one-size-fits-all answer would be nearly impossible to give. Although Table 1 showed that 7,000 most frequent word families in the BNC/COCA lists would be sufficient to provide 98% coverage for online newspapers and magazines, Figure 1 demonstrated that it was certainly not the case.

Still, if we were to take a broad view to the NOW country sub-corpora and consider Bangladesh as a special case, we could generally conclude that 4,000 and 8,000 most frequent word families in the BNC/COCA lists would reliably provide 95 and 98% coverage of the articles from online newspapers and magazines, which could be a rough answer to research question 1. The findings aligned really well with what Nation found in 2006 with the BNC wordlist. However, it is also worth noting that the 3,000-word level in some cases proved itself to be able to represent 95% coverage for online newspapers and magazines. Most importantly, a considerable proportion of data from the sub-corpora showed that the most frequent 6,000 word families can be a feasible learning goal to rely on. It is obvious that for some countries such as New Zealand, the readers’ vocabulary knowledge only needed to be at the 6,000 level to read online news and magazines for unsupported comprehension.

It is also interesting to compare the results to other researches that also employed the BNC/COCA lists. Specifically, online newspapers and magazines were found to share relatively similar lexical demands with English textbooks for EFL learners (Yang and Coxhead, 2020; Rahmat and Coxhead, 2021) and reading passages in international tests of English proficiency like TOEFL, IELTS, TOIEC, etc. (Kaneko, 2020). This proves that online newspapers and magazines could be a great source for English learners who are preparing for their IELTS or TOEFL tests. However, compared to academic books written in English (Hsu, 2018; Lu and Coxhead, 2020), newspapers were shown to be less lexically demanding, which is normal due to the nature of general and academic English.

Data from Figure 1 as well as Supplementary Appendix, demonstrated a “yes” answer to the second research question. In fact, it was really interesting to see that the most lexically demanding newspapers and magazines came from countries where English was a second or even foreign language like Malaysia, Sri Lanka, Philippines, India, and Bangladesh. On the other hand, in countries where people spoke English as a first language or had native-like English language proficiency such as Canada, Australia, New Zealand, Ireland, United States, United Kingdom, Singapore, and Hong Kong, online newspapers and magazines written in English seemed to be easier to read and understand. Certain explanations could be given to this phenomenon, one of which was the components of the BNC/COCA wordlist. As its name suggested, the corpora that were used to create Nation’s (2017) BNC/COCA frequency lists contained spoken and written texts primarily collected from American and British contexts. As a result, the BNC/COCA lists may have aligned better with the written texts from countries that have been heavily influenced by American and British Englishes. In other words, newspapers and magazines articles that had similar wording patterns to those of American or British written English showed better lexical coverage compared to other countries that had different wording patterns.

The findings would be even more interesting if we were to consider word frequency as an indicator of text difficulty. As Hashimoto (2021) and Stewart et al. (2021) discussed, there was a really strong relationship between word frequency rank and word difficulty. Therefore, it could be said that English newspapers from certain countries may pose greater or lesser challenges to certain English learners. International students and immigrants that have been studying British or American English may find these findings interesting since being able to understand local news could be a great way to establish a sense of connection and belonging to the local communities and networks of a country (Juang et al., 2018).

The study’s results would also of help for English teachers around the world, especially those who are thinking of using English newspapers in their own countries as teaching materials. In fact, articles from online newspapers and magazines could be a great source for English language teaching as they provide up-to-date and interesting information while offering a rich linguistic resources. Using newspapers as reading material could easily trigger learners’ interest and facilitate discussions, especially when they are about hot issues in the country or around the world. Teachers and course book writers would have different criteria when selecting teaching materials. But generally speaking, input resources selected for language learning should be lexically less challenging than what’s in the real world. For example, Tegge (2017) pointed out that songs selected by teachers were 1,000–2,000 word families less demanding than other songs on billboard chart. Collins’s (2017) study also indicated that reading passages in EFL textbooks were significantly easier to read than those appeared in standardized tests of English proficiency. If the lexical demands of the input texts have become the number one concern for lesson and material design, then the most obvious and maybe best practice would be to actively choose articles from English speaking countries including Canada, New Zealand, Australia, and the United Kingdom. Newspapers and magazines from ESL contexts like Hong Kong, Ireland, Jamaica and Tanzania could also be put into consideration when choosing reading materials since their articles showed relatively low lexical demands.

It is also noteworthy that even the lowest figures of lexical demands suggested that a word knowledge at 3,000 level was needed to read online newspaper and magazine articles without having to depend too much on dictionaries. Therefore, it is suggestive that the use of authentic articles from online newspapers and magazines in language courses could only be feasible for upper-intermediate or advanced learners. Language teachers should also make sure that their learners know the 2,000 most frequent word families in the BNC/COCA word list, the knowledge threshold where more than 85% coverage could be guaranteed. This could be done by using vocabulary tests that employed the BNC/COCA lists as the source for test items such as the New Vocabulary Levels Test (McLean and Kramer, 2015) the Listening Vocabulary Levels Test (McLean et al., 2015; Ha, 2021a) and Webb et al.’s (2017) Updated Vocabulary Levels Test. The vocabulary knowledge at 2,000 level generally ensures that learners could at least work with the material, of course with the support from teachers and/or more capable peers.

English teachers of advanced classes may use up-to-date newspaper articles as in-class reading activities where learners together read an interesting article and then discuss it. Language instructors can also assign learners to pick articles of their interest that reflect current situations around the world to read extensively, and then discuss what they have read with their peers when they come back to the class. Such practices might be especially suitable for English teachers of immigrants and refugees who would be in dire need of both the language and the updated information of the countries where they were currently based. However, it might be somewhat unrealistic to expect most immigrants and refugees to have knowledge of the most frequent 2,000 word families in the BNC/COCA word list.

The present study bears certain limitations that need to be addressed. As the study employed lexical profiler program accompanied by designed wordlists as the primary research methodology, it was unavoidably affected by the limitations of such approach (Nation and Webb, 2011).

The first limitation was the inability of lexical profiler programs such as RANGE to identify homographs [e.g., proceeds (meaning continues) and proceeds (meaning profits)]. Such a constraint may also have affected the classification of proper nouns since certain proper nouns bear the same spelling as other words (Gates, Walkers, Bush, etc.), leading to the difficult situation where manually adding these words to the list of proper nouns would cause severe conflicts in processing, and leaving them alone would result in these words being ranked as high-frequency words.

Second, lexical profiler programs such as RANGE could not count multiword items as single items. This was, in my opinion, one of the most serious flaws which most lexical profiling research that utilized the same research methodology have been suffering. As the classification of RANGE and other programs like AntWordProfiler was guided by word lists that contained primarily single-item words, they would read phrasal verbs and idioms such as out of the blue, out of the box, sleep on it, come across, sit up… separately and rank the components words of these phrases according to their designed frequency levels. Although most the component words of phrasal verbs and idioms are high-frequency verbs, understanding every single item in such phrases could not guarantee the comprehension of the phrase as a whole (Cornell, 1985; Gardner and Davies, 2007; Garnier and Schmitt, 2015).

The third point that deserves attention concerned how transparent and hyphenated compounds were treated. The present study adopted two assumptions that have been widely applied in vocabulary profiling research (Dang and Webb, 2014; Nurmukhamedov, 2017; Tegge, 2017; Yang and Coxhead, 2020; Nurmukhamedov and Sharakhimov, 2021; Rahmat and Coxhead, 2021) that transparent and hyphenated compounds could be easily understood by knowing the meaning of their component words, which could be problematic to certain extents. For example, whether or not transparent compounds such as aftershock, afterglow or absentminded could be understood by the sheer knowledge of after, shock, glow, and mind was actually a myth. Similarly, assuming a learner could understand sale-driven based on his or her knowledge of driven in the sentence: “The car is driven by Jack.” could be also be a questionable practice.

Besides the limitations concerning research methodology, there was another area that the present study could not address. Although the manuscript showed strong variations in the lexical demands between different countries, it only provided statistical arguments and did not address the issue from a socio-linguistic perspective. Therefore, future studies are encouraged to explore the link between cultural and social-economic factors of countries and their publications’ lexical profile.

The present study offers insights into the vocabulary load of the most popular sources of written English that people read every day. Its findings indicate that knowledge of the most frequent 3,000–4,000 word families plus proper nouns, marginal words, transparent compounds and acronyms could provide 95% coverage of the articles in online newspapers and magazines, which might be a degree of coverage for adequate comprehension and incidental vocabulary learning.

The study confirms Nation’s (2006) findings but emphasizes that coverage of newspaper and magazine articles varies greatly between countries, and that articles from English speaking countries are less lexically demanding than those in ESL and EFL contexts. The results also suggest that web-based newspapers and magazines could be good resources for language teaching and learning. However, it is advised that teachers and learners need to be very selective when choosing what to read as articles from certain countries were shown to be relatively more difficult to understand than others.

The data analyzed in this study is subject to the following licenses/restrictions: the corpora that support the findings of this study are available from Mark Davies. Restrictions apply to the availability of these corpora, which were used under academic license for this study. Data are available from https://www.english-corpora.org/ with the permission of Mark Davies. Requests to access these datasets should be directed to mark.davies@corpusdata.org.

The author confirms sole responsibility for study conception and design, data collection, analysis and interpretation of results, and manuscript preparation.

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.