Corpus Project Overview

This research project was managed by British Council China for the Chinese Basic Education Curriculum and Teaching Material Research Center (BECTMRC). The project sought to identify:

1.  the most commonly used medium to high frequency, age-appropriate language chunks that are presented in the updated 2021 New National English Curriculum (NNEC) (covering Grades 3 to 9) through comparison with commonly used, age-appropriate lexical chunks used by similarly aged native-speaking children in the UK.
2.  prominent gaps in high-frequency language within the New National English Curriculum (NNEC) that can be supplemented or included in future materials revision.

TransformELT, an independent language education consultancy based in the UK, was awarded the grant to conduct the research and publish the results. The principal researcher, James Thomas, created the Corpus of Native Youth English (CONYE23), a corpus of native speaking children’s output, drawing mainly on the CHILDES database and the age-appropriate sections of the BNC14 Spoken corpus for language produced by children. 

Large samples of language input written for the NNEC age group, (i.e. 9 to 15), were also collected, as children’s language output is strongly influenced by the language they encounter in written texts, when reading both for pleasure and for study purposes. Project outputs have been published in the form of an online database and a ‘book of chunks’, which can be accessed at the bottom of this report.

The research outputs will enable Chinese curriculum designers and materials writers for Grades 3 to 9 to verify their intuitions about the language used by similarly aged native-speaking children in the UK, and also to enhance the NNEC word lists with vocabulary items that will bring their teaching materials into closer alignment with native youth English.

Research process

To address both project objectives, a corpus of native-speaking youth English, CONYE23, was created, from which various multi-word units containing each of the words on the NNEC lists were uploaded into a database for further processing. During the life of the project, the output of the corpus research evolved to show the chunks in which the nouns, verbs and adjectives on the NNEC lists are used. 

The corpus output also enabled us to identify word families whose head words appear on an NNEC list, which simplifies their learning, as well as a significant number of word families not represented on the NNEC lists which would therefore require more elaborate teaching.

NNEC word lists

The NNEC word lists represent the minimum vocabulary requirement for school-aged children in China learning English. The lists are presented as lemmas – canonical or dictionary forms – without different parts of speech. There are 482 words on the Primary list and 1,628 on the Junior Secondary list. Very many of these words undergo conversion, i.e., they function in more than one part of speech. After downloading the lists of lemmas in all their parts of speech from the CONYE23 corpus, the primary school list has 778 lemmas and the junior secondary list, which includes the primary list, has 2,326.

Very many of these words also combine with each other to form collocates, chunks and other multi-word units. In the CONYE23 database there are many phrasal verbs, compound nouns and chunks which are combinations of words on the NNEC lists. 

Corpus of Native Youth English: CONYE23 

The Corpus of Native Youth English was created in 2023, using the corpus management tool, Sketch Engine. It contains 53,556,453 tokens – the number of individual words – which include 43,809,730 unique word forms. As each of the texts was collected, mainly from the internet, they were entered into an online database, and the 5,520 documents were tagged for genre, key stage, provenance (UK / US) and corpus (input / output), as well as school year and school subject, whenever this data was available. 

Key stages

In analysing the data, the most decisive category of metadata was key stages (KS), divisions in the British education system. The percentages represent their portions of the corpus. All of the data that was analysed was extracted from a subcorpus of KS1 to 3 (ages 5-14) which contains 31,023,143 tokens. KS1 was included because this represents the language that native-speaking children already know prior to the age at which Chinese children start learning English. Many of the texts for specific KS or age groups are in mixed categories, as they are recommended for multiple age groups. In these cases, the lowest KS was chosen for the metadata. 

Input and output subcorpora

Another major division of the corpus is its separation of input and output data. Input refers to the language that children are exposed to and output to the language they produce. 

The input subcorpus is the corpus of texts created for young native speakers of English, thereby embracing their receptive skills. It includes school subject resources, fiction, the transcripts of films as well as the subtitles of some age-appropriate films. While not every child will read and watch the same things, the basic assumption is that their authors have reason to believe that the language they choose to use will be comprehensible to their audience. 

For the output subcorpus it proved more challenging to obtain samples of language produced by young native speakers, to demonstrate their productive use of language. We planned to gather this data by inviting schoolchildren through their teachers to register anonymously on the CONYE website, and to paste in texts written as part of their schoolwork as well as  any writing they had done on their own. Given safeguarding legislation around child safety, special care was taken to prevent identification of the children. But for a number of reasons, this approach to collecting native-speaker children’s writing proved impractical. We therefore decided to turn to appropriate existing data: CHILDES and a subcorpus of the BNC Spoken (2014) to populate the output subcorpus. 

Resources

CHILDES

CHILDES, the Child Language Data Exchange System, includes a database of transcripts used for research into child language (MacWhinney and Snow 1990). The UK sections of the CHILDES English Corpus were downloaded from their website and processed accordingly. There are 11,712,579 tokens from CHILDES in the CONYE23 corpus. The age ranges in CHILDES correspond roughly to those of the key stages. 

BNC Spoken

BNC Spoken 2014 (Love et al. 2017) is a corpus of present-day spoken British English, gathered in informal contexts in the years between 2012 and 2016. It contains 10,495,185 words of transcribed content, featuring 668 speakers in 1,251 recordings. Among its metadata is the age groups of speakers, which was crucial for our research. The relevant sections of the corpus were downloaded and the many sentences containing taboo words were deleted. 

The CONYE23 database

This is actually a suite of relational databases, whose home database contains the NNEC primary and junior secondary school words. It facilitates generating lists of the words in each part of speech within these two levels. Beside each word in the Word List database, is a row of buttons: bigrams, collocations, grammar patterns and chunks. Clicking these buttons shows how the word is used syntagmatically in CONYE23. One of the original research questions focussed on chunks alone, but these databases furnish this richer palette of lists. The focus of the linguistic analysis was on the syntagmatic patterns in which the NNEC words participate. These are bigrams, collocations and chunks. The online database can be accessed below

Findings

Identifying prominent gaps in the NNEC

In order to identify prominent gaps in the NNEC, it was compared with a word list extracted from CONYE23. There are 43,425 nouns, verbs, adjectives and adverbs in its KS 1–3 subcorpus. The top 400 lemmas, many of which undergo conversion, were selected manually according to their relevance to essential ELT topics such as parts of the body, cultural and sporting activities and technology. The list can be accessed by frequency in CONYE23 and by alphabetical order. 

Collocation

The collocation database is based on the grammatical relationships of a word and its collocates. These syntactic relationships, along with grammar patterns, are an invaluable source of data for a lexical syllabus since they facilitate the teaching of clause structures that are governed by the lexicogrammatical properties of words. This accords with the psycholinguistic processes of language production. They instantiate syntactic pairings, such as the nouns that are the objects of a particular verb.

Grammar patterns

Grammar patterns can be thought of as extended colligations and are therefore properties of words. Grammar patterns are not sentence level syntactic structures such as conditionals, passive structures, etc. The grammar patterns in the database were not derived from the corpus; rather, they are a subset of the Collins COBUILD books of the grammar patterns that contain the patterns of many thousands of nouns, verbs and adjectives (Hunston & Francis 2000). The subset contains NNEC words only. This facilitates a systematic approach to vocabulary study that teaches syntactic patterns with the vocabulary in which they function. Students start by saying someone accepts something for someone; then they replace the placeholders with concrete references, such as the captain accepted the prize for her team.

Chunks

Chunks are by definition, semantically whole and require a different approach to deriving them from corpora. Corpus tools are not currently equipped to identify strings of words that are semantically whole. Nevertheless, using syntagms as an intermediary stage, it has been possible to derive many thousands of semantically whole chunks. 

Conclusion

The CONYE23 corpus and database of the multi-word units which instantiate how young native speakers use the words on the Chinese NNEC word lists demonstrate that they use words in the same patterns as adult speakers. This is inevitable as the precise meanings of words in any spoken or written text often result from their co-texts of collocation, colligation and grammar patterns. The majority of the chunks that are handed down to young native speakers remain in the language of adult native speakers. 

Some language produced for and by young native speakers is dropped as they mature linguistically, socially, physically and educationally. This includes discourse encompassing activities, games, toys, characters in stories, poems and songs, as well as various manifestations of asserting independence as they grow older. 

The database is a useful resource that will facilitate the study of existing texts and the writing of new texts to be used in teaching materials. This will enable more linguistically and pedagogically reliable approaches to the study of vocabulary. A well-structured syllabus that teaches collocation, grammar patterns and chunks recycles vocabulary in a way that leads students to grasp the roles of co-text and context in creating meaning.

This study has sought to bring to light the highly patterned nature of the language that young native speakers use. The properties of words which they acquire as they are exposed to language and develop their own use of language have the potential to become enriched content in foreign language teaching syllabuses. 

Sarah Mount, Alan Pulverness, James Thomas – July 2023

References

Hunston, S., & Francis, G. (2000) Pattern Grammar. Amsterdam, Philadelphia. John Benjamins.

Love, R., Dembry, C., Hardie, A. Brezina, V. and McEnery, T. (2017) The Spoken BNC2014: Designing and building a spoken corpus of everyday conversations. International Journal of Corpus Linguistics 22(3), 319-344

MacWhinney, B., & Snow, C. (1990) The Child Language Data Exchange System: An update. Journal of Child Language, 17(2), 457-472. doi:10.1017/S0305000900013866