The Academic Word List: A Foundational Resource for Academic English

The Academic Word List (AWL), developed by Averil Coxhead at Victoria University of Wellington, is a crucial tool for students, educators, and researchers seeking to enhance their proficiency in academic English. The AWL provides a structured vocabulary framework that bridges general English and specialized disciplinary language, making it a fundamental resource in higher education and English for Academic Purposes (EAP) instruction.

Composition and Structure of the Academic Word List

The AWL consists of 570 word families that appear frequently in academic texts but are not included in the most common 2,000 words of English. These words were identified through a comprehensive corpus-based analysis of 3.5 million words from academic texts across four broad disciplinary categories: Arts, Commerce, Law, and Science. The goal of the AWL is to provide learners with a lexicon that is applicable across disciplines, supporting both comprehension and production of academic discourse.

To facilitate learning and usage, the AWL is organized into 10 sublists ranked by frequency. The first sublist contains the most commonly occurring words in academic writing (e.g., analyze, concept, section), while the final sublist includes less frequent but still significant terms (e.g., adjacent, forthcoming, persistent). By structuring the list in this way, Coxhead’s work provides learners with a gradual and systematic approach to acquiring academic vocabulary.

Significance and Pedagogical Applications

The AWL has had a profound impact on academic writing, reading comprehension, and language instruction. For non-native English speakers, mastery of AWL vocabulary can significantly enhance their ability to engage with complex academic texts and articulate their ideas more precisely in written assignments. Additionally, for native English-speaking students, a strong command of AWL vocabulary contributes to the clarity and formality of academic writing.

The AWL is widely used in EAP programs, where it serves as a foundation for vocabulary instruction. It also informs curriculum development, assessment design, and self-directed learning strategies. For instance, learners preparing for high-stakes language proficiency exams such as IELTS and TOEFL benefit from explicit instruction in AWL vocabulary, as these words frequently appear in reading and writing sections of such assessments.

Potential Limitations

In their 2020 study, “Word lists and the role of academic vocabulary use in high stakes speaking assessments,” Smith, Kyle, and Crossley examined the effectiveness of various academic vocabulary lists, including the Academic Word List (AWL), in predicting performance on TOEFL speaking tasks. Their findings indicated only weak associations between the use of words from these lists and speaking scores, suggesting that the AWL and similar lists may not fully capture the lexical demands of academic speaking assessments. The authors recommend developing specialized vocabulary lists tailored to the specific requirements of academic speaking contexts.

https://www.wgtn.ac.nz/lals/resources/academicwordlist

Coxhead, A. (2000). A New Academic Word List. TESOL Quarterly, 34(2), 213–238. https://doi.org/10.2307/3587951

The University of South Florida (USF) Word Association, Rhyme, and Word Fragment Norms is a dataset compiled to study word relationships and cognitive processes in language. This database provides norms for word associations, rhyming patterns, and word fragment completion, making it a valuable resource for psycholinguistic research.

The database provides:

1. Word Association Norms

• Based on free word association tasks where participants were given a stimulus word and asked to provide the first word that came to mind.

• Provides forward associations (from cue to response) and backward associations (from response to cue), measuring how strongly words are linked in human cognition.

• Example: The stimulus “dog” might elicit responses like “cat” or “bark.”

• Contains approximately 5,000 stimulus words.

• Each word was presented to multiple participants (often 100+), resulting in over 72,000 unique word associations.

2. Rhyme Norms

• Includes lists of words that participants judge as rhyming with a given stimulus.

• Helps in phonological processing research and understanding how sound similarities influence memory and retrieval.

• Example: The word “hat” might elicit rhyming words like “bat” or “mat.”

3. Word Fragment Completion Norms

• Participants were given partial word fragments and asked to complete them.

• Used to study lexical access and retrieval processes.

• Example: Given “_ouse,” common completions might be “house” or “mouse.”

Relevance to Lexical Sophistication

The norms are a relevant measure Lexical Sophistication because they capture contextual distinctiveness, which assesses how diversely a word is used in different contexts .

Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of South Florida word association, rhyme, and word fragment norms. http://www.usf.edu/FreeAssociation/.

The ELP (English Lexicon Project) is a large, publicly available psycholinguistic database that provides behavioral norms for word recognition. It includes data collected from lexical-decision and word-naming tasks performed by native English speakers. The database contains response latencies, accuracy rates, and other relevant linguistic information for a large set of words.

The database includes measures

• Lexical-Decision Latencies: Measures the time it takes for participants to decide whether a given string of letters is a real English word. This is the so-called lexical decision task and often appears in the literature as LDT.

• Word-Naming Latencies: In the speeded naming task, subjects are presented with a visual word (or sometimes a nonword) and are asked to name the word aloud as quickly and as accurately as possible.

• Response Accuracy: Tracks how often participants correctly identify words in lexical-decision and word-naming tasks.

• Coverage: Contains data for approximately 40,481 real words and an equal number of nonwords used in lexical-decision tasks.

• Data Source: Collected from 816 native English-speaking subjects across different institutions.

The ELP is commonly used in psycholinguistics, cognitive science, and natural language processing to study word recognition, reading comprehension, and lexical access. It serves as a valuable resource for research on word frequency effects, lexical processing, and language development.

We use the TAASSC tool of Kristopher Kyle and hence inherit his operationalization of VAC. This is pretty straightforward, Kyle operationalizes a VAC in his thesis as

Verb argument constructions in TAASSC are defined as a main verb and all direct dependents that verb takes.

He notes that there are some limitations of this definition

Furthermore, the definition of a verb argument construction was largely determined based on the features analyzed by the Stanford Neural Network Dependency Parser (Chen & Manning, 2014). While this approach was straightforward and likely reduced error rates, it is possible that distinctions between VACs were made that were not appropriate. For example, a VAC (e.g., subject-verb-object) that includes a subordinating conjunction (i.e, because) was counted as a separate VAC type from its non-subordinated counterpart.

There are some important details hidden in the definition. It clearly depends on the dependency parse produced by the Stanford Neural Network Dependency Parser [we used version 3.5.1 – the last version that uses the SD tagset] but the Stanford Neural Network Dependency Parser produced three different dependency parses!

The representations are

1. Basic Dependencies [ type=”basic-dependencies”]

Basic dependencies provide a simple, straightforward representation of the syntactic structure of a sentence. Each word in the sentence is linked to its head word via a grammatical relation, forming a tree structure.

2. Collapsed Dependencies [ type=”collapsed-dependencies”]

Collapsed dependencies simplify the representation by collapsing certain prepositional and conjunctive relations into a more direct dependency, often improving downstream processing tasks. Specifically, it converts prepositional phrases into direct relations.

3. Collapsed and Propagated Dependencies [type=”collapsed-ccprocessed-dependencies”]

This representation extends the collapsed dependencies by further propagating conjunct dependencies. If a verb is connected to a conjunction (and, or), the dependency relations are duplicated for each conjunct.

Kyle uses the third of these to operationalize his VACs. He makes other modifications to the tree too before producing the definitive list of VACs. I hope to examine these in detail with examples. If this was done in Kyle’s thesis I can’t find it.

Explanations of the dependency parsing is in the Stanford Dependencies Manual.

Together with my colleagues J. Elliott Casal and Christopher Stewart, I am submitting a paper looking at the use of VACs by Generative AI. We use the excellent TAASSC tool of Kristopher Kyle to capture these VACs. We used the TAASSC version 1.3.8 to extract VACs.

A Verb Argument Construction (VAC) refers to a linguistic structure in which a verb and its associated arguments (such as subject, object, and indirect object) form a conventionalized pattern that conveys a specific meaning. The concept of VACs is rooted in Construction Grammar (Goldberg, 1995, 2006), which posits that grammatical constructions—combinations of words with particular syntactic and semantic properties—are fundamental units of language.

VACs are important because they reveal how speakers of a language systematically associate verb meanings with particular syntactic structures. For example, the ditransitive construction (“X gives Y Z”) inherently expresses transfer, regardless of the verb used (e.g., “give,” “send,” “tell”). Likewise, the caused-motion construction (“X causes Y to move Z”) conveys movement (e.g., “throw the ball into the box,” “push the chair out the door”). These constructions shape how verbs function in different syntactic frames, impacting language acquisition, processing, and variation (Ellis & Ferreira-Junior, 2009).

Importance of VACs in Linguistics:

1. Language Learning and Acquisition

VACs play a crucial role in first- and second-language acquisition. Research shows that learners acquire constructions holistically before abstracting verb-specific rules (Ellis, 2002). This supports the usage-based model of language learning, where exposure to frequent constructions facilitates acquisition.

2. Syntax-Semantics Interface

VACs bridge syntax and semantics by encoding meaning through syntactic structures. Goldberg (2006) argues that meaning is not solely derived from individual verbs but also from the construction itself, meaning that verbs “inherit” meaning from the VACs they appear in.

3. Computational and Corpus Linguistics

VACs have been widely studied in corpus linguistics and natural language processing (NLP) for verb classification, semantic role labeling, and machine translation (Stefanowitsch & Gries, 2003). By analyzing large corpora, researchers can identify probabilistic patterns in verb usage.

4. Cross-Linguistic Comparisons

VACs vary across languages, making them important for typological studies. Some languages rely more on argument structure constructions than verb morphology to express meaning (Levin, 1993). Understanding these variations informs linguistic theory and translation studies.

Key References:

• Ellis, N. C. (2002). Frequency effects in language processing: A review with implications for theories of implicit and explicit language acquisition. Studies in Second Language Acquisition, 24(2), 143–188.

• Ellis, N. C., & Ferreira-Junior, F. (2009). Construction learning as a function of frequency, frequency distribution, and function. The Modern Language Journal, 93(3), 370–385.

• Goldberg, A. E. (1995). Constructions: A Construction Grammar Approach to Argument Structure. University of Chicago Press.

• Goldberg, A. E. (2006). Constructions at Work: The Nature of Generalization in Language. Oxford University Press.

• Levin, B. (1993). English Verb Classes and Alternations: A Preliminary Investigation. University of Chicago Press.

• Stefanowitsch, A., & Gries, S. T. (2003). Collostructions: Investigating the interaction of words and constructions. International Journal of Corpus Linguistics, 8(2), 209–243.

Most academic software projects that I work with are very different from code developed by professional developers. There is frequently no unit testing. I work with projects that fall into three broad classes:

1. local programs.
2. web apps that are light clients that do almost nothing except call an API.
3. web apps that do extensive client side work and that interface with external databases.

Testing is normally done by live testers. This is useful for ensuring that the system runs and identify UX issues but testers may not identify issues where the code functions but delivers results which are not entirely correct. Live testing is also not easily replicable. Absent unit testing we are frequently faced with doing end-to-end testing. Even with unit tests there are several reasons to try to build an end-to-end test suite. An end-to-end test suite in its simplest form is a collection of inputs and expected outputs. For testing CLIs and APIs this is normally just a spreadsheet of input parameters and expected output. For GUI apps and web apps this means a list of actions within the app and then an expected output and will require a testing framework (such as selenium or BrowserStack) . There are several reasons to build such a test suite

1. there is unlikely to be a proper software requirements specification for an academic project. A end-to-end test suite explains what the researcher understands as correct functioning. As the “specifications” evolve we add new test lines that capture the new functioning and revise the old ones as necessary.
2. as with classic test suites, as we make maintenance or performance changes to existing code we can check that the code continues to function as expected.
3. these test suites serve as excellent performance test suites for profiling tools.