ARTICLES

 

Academic Word Families in Online English Dictionaries

 

Akademiese woordfamilies in aanlyn Engelse woordeboeke

 

 

Geraint Paul Rees

Translation and Language Sciences, Pompeu Fabra University, Barcelona, Spain (geraintpaul.rees@upf.edu) (https://orcid.org/0000-0002-9204-8073)

 

 

---

ABSTRACT

The concept of the word family has been widely employed in research on vocabulary in the teaching and learning of foreign and second languages. The underlying assumption being that once learners know one member of a word family, they can recognise other members. Empirical research supports this vis-à-vis receptive knowledge of inflectionally related wordforms. However, studies of academic writing indicate that using appropriate derivative forms of a known word is challenging, suggesting a need for dictionaries with morphological support for writers. Traditionally, in paper-based dictionaries, this need could not be fulfilled due, in part, to space constraints. This study aims to establish if it is met in five online English dictionary websites. It analyses the treatment of seventy-four academic wordforms which academic writers have been shown to have difficulty deriving when presented with the related base word. Results indicate good coverage of the derivative forms across the dictionary websites examined but inconsistency within and between resources in the way in which forms are treated. Differences include the status as entries or subentries and the provision of writing support features such as examples, grammar patterns, and collocation information. Finally, changes to the treatment of derivatives to better serve academic writers are suggested.

Keywords: academic writing, derivative forms, lexicography, morphology, online dictionaries, vocabulary acquisition, word families, writing support

---

OPSOMMING

Die woordfamilie-konsep is reeds wyd in woordeskatnavorsing in die onderrig en aanleer van vreemde en tweede tale ingespan. Die onderliggende aanname word gemaak dat wanneer leerders een lid van 'n woordfamilie ken, hulle ook ander lede kan herken. Empiriese navorsing steun hierdie aanname ten opsigte van reseptiewe kennis van fleksieverwante woordvorme. Studies van akademiese skryfwerk toon egter dat die gebruik van toepaslik afgeleide vorme van 'n bekende woord 'n uitdaging bied, wat daarop dui dat daar 'n behoefte aan woordeboeke met morfologiese steun vir skrywers bestaan. Tradisioneel kon, deels weens ruimtebeperkings, nie aan hierdie behoefte in papiergebaseerde woordeboeke voldoen word nie. In hierdie studie word beoog om vas te stel of daar in vyf aanlyn Engelse woordeboekwebtuistes wel hieraan voldoen word. Die hantering van vier-en-sewentig akademiese woordvorme waarmee akademiese skrywers sukkel om afleidings daarvan te vorm wanneer hulle die verwante basiswoord teëkom, word geanaliseer. Die resultate dui op goeie verteenwoordiging van die afgeleide vorme in die woordeboekwebtuistes wat ondersoek is, maar toon ook teenstrydighede binne en tussen hulpbronne t.o.v. die metode waarop die vorme hanteer word. Verskille sluit die status as inskrywings of subinskrywings en die voorsiening van skryfhulpkenmerke soos voorbeelde, grammatikale patrone en kollokasie-inligting in. Ten slotte word veranderings aan die hantering van afleidings voorgestel om akademiese skrywers beter van hulp te kan wees.

Sleutelwoorde: akademiese skryfwerk, afgeleide vorme, leksikografie, morfologie, aanlyn woordeboeke, woordeskatverwerwing, woordfamilies, skryfhulp

---

 

 

Introduction

Over the last three decades, the term 'word family' has been used in language teaching and vocabulary research to describe the categorisation of wordforms based on their inflectional and derivational morphology. The construct has been adopted enthusiastically in research on vocabulary in English language teaching. A key factor motivating the concept of word family (henceforth WF) was a desire to provide guidelines for the treatment of morphologically related wordforms in lexicography and language teaching (Bauer and Nation 1993). The starting point for this study is a list of seventy-four wordforms frequently used in academic contexts. It comprises sixteen basic wordforms and their derivatives. Empirical research has shown that L2 users have difficulty producing the WF members (i.e., related wordforms) of these forms in writing (Schmitt and Zimmerman 2002). This study aims to establish how well L2 users of the "Big Five" English dictionary websites are supported when producing these problematic forms by examining their treatment on these websites.

Word families and levels. The idea motivating WFs is that wordforms can be grouped based on their inflectional and derivational morphology. These groups can then be organised into levels. Table 1 reproduced from Bauer and Nation (1993: 254) shows the levels for the WFs develop, wood and bright.

 

 

An increase in WF level entails greater formal or semantic irregularity. At Level 1, each form represents a distinct word (i.e., one word = one family). At Level 2, inflected forms with the same base are grouped. The idea being that a learner who can recognise and use develop or any of its inflected forms could recognise and use the base or any of its other inflected forms. From Levels 3 to 6 eight criteria determine the level of an affix and its derived wordform (Bauer and Nation 1993: 256).

1. Frequency (generalisability): Affixes at lower levels occur in many word-forms. For example, the Level 2 inflectional affixes -s, -ed, -ing, are common to all English verbs. In contrast the affixes pre- and re- are far less generalised.

2. Productivity: The possibility of the affix forming new wordforms. Inflectional affixes -s, -ed, -ing frequently produce new forms with the base of any verb. Whereas since -ful is far more selective in the nouns and verbs it combines with, it produces far fewer wordforms.

3. Predictability: The extent to which the meaning of the word created by affixation can be predicted from the meaning of the base and the affix. For example, -ly attached to an adjective X, typically means 'in X manner', thus is highly predictable. In contrast, -ful when attached to nouns does not always produce word with predictable meaning (e.g., awful weather ≠ awe inspiring weather).

4. Regularity of the written form of the base: At lower levels, removing the affix leaves the base intact, at higher levels orthographic changes to the base are evident (c.f. red +ness and impose + ition).

5. Regularity of the spoken form of the base: At lower levels removing the affix leaves it phonologically intact, at higher levels phonological accommodations are evident. For example, removal of the Level 6 affix -ify from mystify gives myst; not a free base in its spoken form.

6. Regularity of spelling of the affix (allomorphy): For example, pre- has one written form, while in-, im-, il-, and ir- are allomorphs of in-.

7. Regularity of the spoken form of the affix (allomorphy): The extent to which the phonological form of the affix is predictable. For example, although the Level 1 affix -ed has three spoken forms, these are predictable.

8. Regularity of function: The extent to which the affix attaches to a base of a particular word class and produces a word of a particular class. For example, -ship always combines with nouns to produce nouns.

By applying the criteria above, Bauer and Nation (1993) produced the list of affixes in Table 2. Two levels are omitted: Level 1 where each wordform is treated as a different WF, and Level 7 where items have classical roots and affixes that are not found in the sample of WFs in this study.

 

 

Word families and language teaching and learning. The usefulness of WFs for language teaching relies on the assumption that once learners know one member, they can recognise others. This has been termed relational knowledge (Tyler and Nagy 1989). Some empirical research supports this for L1 readers and in-flectionally related wordforms. However, that derived forms are generally acquired after inflected forms suggests they pose greater problems (Berko 1958). For L2 users, the assumption of relational knowledge is more uncertain. Even proficient L2 users find using suitable derived forms of a known word challenging.

Studies on L2 writing or vocabulary acquisition suggest that learners find derivational morphology challenging. A longitudinal study of English vocabulary acquisition involving three L2 English postgraduate students in the UK indicated gaps in participants' morphological repertoire, particularly regarding the formation of adjectives and adverbs. Schmitt (1998) suggests that morphological errors become fossilised since two of the three participants made little progress producing morphologically related forms over an academic year. Another study of the English word association and grammatical suffix knowledge of 95 secondary and undergraduate students of English in Japan found participants gained 330 words on average over an academic year but could only produce 15% of the possible derivatives (Schmitt and Meara 1997). Similarly, in a study of TOEFL vocabulary involving 30 learners taking English language courses in preparation for undergraduate study in the UK, participants could only produce derivates in all four major word classes for 12 of 180 possible target words (Schmitt 1999).

Research focusing on productive knowledge of derivational morphology among learners is rarer. Schmitt and Zimmerman's (2002) carefully designed study examined the productive knowledge of 106 L2 English students who comprised two groups: One undertaking pre-sessional and undergraduate English language courses at universities in the US and the UK, and another an MA in English language teaching at a university in the UK. Participants were given 16 prompt words for which they were asked to complete gapped sentences by producing derivative forms of the prompt word from the four major word classes (noun, verb, adjective, and adverb). Participants produced only 50% of the derivative forms permissible. Although the presumably more proficient MA group performed better, knowledge of derived word-forms was still partial even for words which participants felt they knew well. This demonstrates a need for dictionaries that support written production of derived wordforms.

Word families and dictionary making. WFs were posited to help lexicographers treat morphology in a principled and consistent way. Bauer and Nation (1993) criticise the inconsistent treatment of derived forms as entries and sub-entries in several general-purpose English dictionaries from the late 1970s and 1980s. They are not alone in highlighting this issue. However, much research has focused on affixes themselves rather than the derivative forms produced by affixation. For example, Stein (1985) highlights different policies on the positioning of affixes in the indexes of several MLDs. Considering dictionaries as writing aids, it makes little sense to focus on affixes themselves rather than derivative wordforms produced by affixation. Writers are unlikely to ask, 'What word can I form with -ize?' but will likely query the use of a particular word, for example, 'How do I use philosophize in a sentence?'

There is some consensus on the treatment of wordforms derived by affixation. To be included, a derivative form must be established enough to occur above a certain frequency (De Caluwe and Taeldeman 2003; Stein 1985). Semantic predictability is another important consideration: "The more the meaning of a combination is assumed to be inferable from the meaning of its constituents listed in the dictionary and the process of formation itself, the stronger the likelihood that it will not be listed as a dictionary item" (Stein 1985: 38).

Analyses of entries for derivative forms reveal diverse interpretations of these criteria. In an examination of eight monolingual English desk dictionaries including MLDs, Stein (1985) highlights inconsistent definition of -ish derivatives from adjectives designating colour, and inconsistent treatment of derived forms as either main entries or run-ons. Similarly, De Caluwe and Taeldeman (2003) demonstrate inconsistent treatment of wordforms derived from water in the Woordenboek der Nederlandsche Taal (Van Sterkenburg 1992: 115), noting that some are listed as separate entries or lemmas and others within the headword water.

WFs were posited to remedy these inconsistencies. The idea is that as formal and semantic irregularities increase with higher-level word families, they require "more attention" from the lexicographer (Bauer and Nation 1993: 255). Bauer and Nation suggest ignoring regular, semantically transparent word-forms at Level 1; listing those created by inflection affixation at Levels 2 and 3 as non-defined sub-entries and treating higher-level items as main entries.

WFs and electronic lexicography. Electronic lexicography has been suggested as the solution to the inconsistent treatment of derived forms. Firstly, ostensibly freed from space constraints of paper dictionaries, electronic dictionaries have the potential to include information on all the derived wordforms in a language¹. Secondly, unbound by the alphabetical index, they could offer several routes to the derivative wordform (De Caluwe and Taeldeman 2003).

Regarding the first point, De Caluwe and Taeldeman (2003) stress the importance of not overwhelming the reader with information: "it is not the intention to confront the reader with an interminable amount of information, but this should be possible if the reader so desires" (De Caluwe and Taeldeman 2003: 121). Regarding access structure, they sketch an example of how an onomasiological query for "the fact/quality of being long" (De Caluwe and Taeldeman 2003: 123) might proceed in an ideal dictionary. With reference to Elektronisches Lernerworterbuch Deutsch-Italienisch/Dizionario Elettronico per Apprendenti Italiano-Tedesco (ELDIT), Ten Hacken, Abel and Knapp (2006) present a detailed example of how derivative forms can be treated in electronic dictionaries.

Aims. Lexicography has changed significantly since Bauer and Nation's guidelines were published. Many space and alphabetical ordering constraints of paper dictionaries have been mitigated in online resources. These could feasibly accommodate calls from research on WFs in language teaching for greater writing support for L2 English with derived forms. Accordingly, this study aims to investigate how derivatives are represented in online English dictionary websites consulted by learners. It will answer the following research questions:

1. How well are derivationally related members of WFs covered by dictionary websites with online monolingual English dictionaries?

2. To what extent are they treated in a way which facilitates use in writing?

 

Methodology

In this section, the dictionary websites examined are discussed along with the reasons for their selection. Next, the sample of 74 derived wordforms shown to be problematic for L2 English users is presented and the process Schmitt and Zimmerman (2002) used to obtain this list explained. Finally, the categories and procedure used in this analysis are given.

Dictionary websites examined. This study examines the treatment of morphological behaviour on five popular English dictionary websites (Cambridge, https://dictionary.cambridge.org/ (CAM); Collins, https://www.collinsdictionary.com/ (CD); Longman, https://www.ldoceonline.com/ (LONG); Macmillan, https://www.macmillandictionary.com/ (MELD); and Oxford, https://www.oxfordlearnersdictionaries.com/ (OX)².). The versions examined were those live in December 2022.

Monolingual Learners Dictionaries (MLDs) are the obvious place to investigate morphological information for learners. However, the migration from paper-based dictionaries to online dictionaries complicates this assumption. Of the "Big Five" monolingual English dictionary makers, only Longman and Mac-millan offer direct access to their MLDs. LDOCEonline.com also gives access to the Longman Business Dictionary³ (LBD). Access to the MLDs of Cambridge, Collins, and Oxford is offered via portals which aggregate content from several different dictionaries. For example, the collinsdictionary.com entry for precision collates data from Collins COBUILD (COBUILD), Collins English Dictionary (CED)⁴, Webster's New World College Dictionary (Agnes 2010) (WNWCD4), and other ancillary sources. This study investigates the data presented by each portal rather than focusing only on entries from MLDs since, while dictionary researchers are cognisant of different dictionary types and their target users, many end-users, particularly those at lower proficiency levels, simply want to get the job done. It would be strange if an end-user disregarded information from collinsdictionary.com because it came from CED not COBUILD.

Productively challenging academic word families. Schmitt and Zimmerman (2002) judge 74 wordforms acceptable responses to gapped sentences based on sixteen prompt words. These represent an ideal sample with which to investigate the treatment of morphological information in online English dictionaries. The sixteen prompt words were selected from Coxhead's (2000) A New Academic Wordlist (AWL). This lends content validity since many English dictionary users, including those shown to have problems with derivative forms in the research discussed above, work in academic contexts.

To obtain the list of 74 acceptable derivate wordform responses, Schmitt and Zimmerman (2002) first extracted all listed derivatives from four learners' dictionaries⁵. Secondly, they used frequency information from the BNC1994 to remove infrequent derivatives. Finally, they elicited responses from 36 L1-Eng-lish university students to the same gapped sentence prompts used by the non-native speakers. In arriving at their list of acceptable responses, they prioritised this final step. Table 3 shows WFs containing the basic and related wordforms along with their word class and WF level in parenthesis.

 

 

There are, at least, two notable points about this list. Firstly, Schmitt and Zimmerman (2002) treat accessed, assumed, authorized, released and surviving as adjectives. However, the first four could reasonably be verbs and surviving could be a verb or a noun. This is a frequent dilemma in English lexical analysis with no satisfactory answer (Hanks 2013). There are cases where these items are used as verbs and others where they are used as adjectives (Frankenberg-Garcia, Rees and Lew 2021). The analysis procedure below accounts for this. Secondly, the AWL has received criticism for ignoring discipline-specific differences in meaning, not accounting for the role of collocates in conditioning meaning and being based on the outdated A General Service List (West 1953) (Hyland and Tse 2007; Rees 2021). However, Schmitt and Zimmerman (2002) suggest these are words learners in academic contexts often need to produce. Their standout finding of partial knowledge of derivative forms demonstrates that L2 English users struggle to produce these words. Consequently, these are words for which they could conceivably consult a dictionary for guidance.

Procedure. A search for each of the seventy-four problematic derived forms is conducted on the five dictionary websites. The analysis of the results proceeds in two stages: Stage 1 records whether a wordform is covered; Stage 2 records whether the wordform is treated in a way that supports writing. Namely, whether examples and/or grammar and collocation information are provided. Except for MELD, the websites offer access to several different individual dictionaries. To imitate typical user behaviour, default settings for English language searches are used, and only those dictionaries from which data is presented on the initial results page are considered.

Stage 1: Analysis of coverage. Evaluating dictionary coverage involves not only judging if an item is covered, but also how it is covered. This study distinguishes between main entries and sub-entries. Across all the dictionary websites, in main entries the target word is listed as a headword. Sub-entries are more diverse. CAM and OX do not use sub-entries for derived forms. CD often lists derived forms as sub-entries as part of the main entry for the base form (Figure 1). On CD, derived forms are often, simultaneously, presented at the foot of the main entry for the base form under the heading "Derived forms" (Figure 2). In many LONG entries, "Word families" containing derived forms are shown at the top of the results page. On LONG, derived forms are sometimes presented as sub-entries (Figure 3). MELD often lists derived forms at the foot of the main entry for the base word under the heading "Derived word" (Figure 4). For ease of comparison, all these variations in sub-entry presentation are labelled 'sub-entry' here.

 

 

 

 

 

 

 

 

Additionally, the websites' response to searches for rare wordforms differs. CAM sometimes uses placeholder examples retrieved automatically from a corpus. If no standard entry can be found, MELD occasionally redirects the user to an example from its crowd-sourced OPEN DICTIONARY. In this analysis, placeholder and crowdsourced examples are treated as coverage provided examples are relevant to the target word. CD, LONG, MELD, and OX redirect the user to the more common wordform (e.g., philosophic redirects to philosophical). The common and rarer forms are considered interchangeable.

The treatment of words with ambiguous word classes, principally -ed affixes in the sample, varies within and between the dictionary websites. Searches for wordforms presumed adjectives by Schmitt and Zimmerman (2002) ending in -ed often redirect to the entry page for the verb in MELD and OX which use different pages for word classes, and to the main-entry page in covering both noun and verb in the other resources. If a sub-entry exists on these pages for the adjectival form, this form is recorded as being covered (e.g., authorize in LBD). Occasionally, traditional examples or corpus lines (automatically generated, occasionally incomplete sentences) illustrate an adjectival use even though the adjectival sense in not explicitly covered (e.g., release in CAM: "To what extent the rural sector absorbs the released labour is not clear"). In these cases, the presence of a relevant example or corpus line is noted for stage two of the analysis.

Stage 2: Analysis of support for written production. A key assumption of this study is that examples and information about typical grammatical and collocational behaviour support productive use of the wordforms. While there is much research about what constitutes a good dictionary example (Kilgarriff et al. 2008) and the optimal number of examples for supporting production (Frankenberg-Garcia 2015; Ptasznik 2023), here analysis is limited to noting the presence or absence of examples.

In this study, typical combinations containing grammatical words (i.e., prepositions and determiners) are labelled grammar patterns while typical combinations of lexical words are labelled collocations. This policy is maintained irrespective of how these combinations are labelled on the dictionary websites. For example, combinations of grammatical words often appear in the collocation dictionary sections of the websites. The theoretical debate about the difference between collocation and grammar pattern is irrelevant for most dictionary users. However, information about the lexical items which co-occur with a particular wordform, and their syntactic configuration is useful for writers.

By aggregating the number of entries with writing support features such as examples, grammar patterns, and collocation information and dividing this by the total number of items from the sample covered, a writing support score can be calculated. This score gives an approximation of how well a resource supports users with the sample items when writing.

The overall writing support score is the ratio (R) of the sum of items with examples (E), grammar patterns (G), and collocations (C) for the sample items covered by the dictionary (N). To reflect the diversity in syntactic behaviour of word classes and the varying degrees of difficulty they could present writers, four writing support scores are reported:

- Overall score

- Score with adverbs excluded

-- Score with adjectives excluded

-- Score with adverbs and adjectives excluded

To calculate the exclusive scores, the sum for items of the included word classes (T) are first weighted (W) representing their proportion of the total sample items covered by the dictionary (N):

The exclusive score, a ratio, is then calculated using this weighting.

The coverage statistics and writing support scores indicate how well users of the dictionary websites are supported when seeking to use the problematic derivative forms in writing. For a more detailed impression, it is necessary to examine which items have writing support features.

Although examples can provide information about grammar patterns, here analysis focuses on semantics. Namely, whether derived wordforms missing examples are sufficiently semantically regular for a user to infer their meaning and use. This study does not differentiate between exemplification styles employed in the dictionaries. However, it is noteworthy that CAM, CD, and LONG occasionally present examples automatically extracted from corpus lines. When relevant to the target word, these are counted.

Comparing items with and without grammar pattern and collocation information by word class across the dictionary websites provides a clearer impression of how well users are supported when writing the problematic forms. Although users can intuit collocation and grammar patterns from examples, only those instances where the dictionary compiler intentionally highlights these aspects are considered. Common strategies include presenting salient collocations or grammar patterns in bold in examples (all dictionary websites examined) and/or separating common collocates with slashes (e.g., LONG (Figure 3) and OX) and displaying information from the publisher's collocation dictionary for certain searches. Additionally, LONG occasionally provides links to fuller entries for salient collocations and grammar patterns; CD, MELD, and OX display common idioms for some of the sample, while CD includes COBUILD grammar patterns.

 

Results and discussion

Coverage. The impression of inconsistent treatment of derivative forms reported in previous research is not immediately supported. Most items in the sample are covered by the five websites. The mean number of items treated per website (N = 74) is 66.4 with a standard deviation of 3.64. The overall coverage of the sample items did not differ significantly by dictionary website, X² = 3.194; df = 4; p < .05.

 

 

A high degree of coverage was expected, the sampling criteria ensured target items were used reasonably frequently and widely. Indeed, inclusion in a dictionary was one of the three criteria Schmitt and Zimmerman (2002) used to select the permissible responses to their gapped sentence exercise.

Greater difference is apparent in how words are treated. OX and CAM cover all sample items as main entries, while CD, LONG, and MELD use sub entries for around one-sixth of the items. This suggests that although CAM and meld cover a greater number of items overall, OX and CAM provide better writing support than the other resources. Further analysis of the entry contents is needed to substantiate this.

Table 4 shows the eleven wordforms which are missing from at least one website. Only three items are absent from all websites: authoritive and authoritively, infrequent spellings of authoritative and authoritatively, and traditionize a rarer verb meaning 'to make into a tradition'.

 

 

The treatment of coherency, a more infrequent form of coherence, and philosophic, a more infrequent form of philosophical, is inconsistent. The former is absent from CAM, MELD, and OX, the latter not found in CAM. Except for philosophic in CAM and CD, searching for these wordforms redirects the user to the page for the more frequent form. Once there, the infrequent form is listed after "also" (LONG and OX) or "or" (MELD). The first entry when searching for philosophic on CD is a COBUILD entry stating: "Philosophic means the same as philosophical" with a hyperlink to philosophical. Since both wordforms are wholly interchangeable, this redirection strategy seems sound. For resources where the infrequent forms are not listed, the alphabetic proximity of these items to their counterparts means that users may select the relevant form from the alphabetical listing presented when a search produces no exact results. Searching for philosophic in CAM produced a placeholder consisting solely of corpus lines for philosophic.

The treatment of forms with the Level 3 affix -ness, liberalness (only present in CD) and preciseness (absent from LONG and OX), may be inconsistent. For example, preciseness is in CAM but not liberalness. It may be that liberalness was considered too infrequent for inclusion⁶ or its inclusion may be an oversight given the productivity (almost any adjective + -ness produces an acceptable noun) and semantic regularity (meaning "'property of being X', where X is the base adjective" (Carstairs-McCarthy 2018: 78)) of this suffix. However, as these are the only two -ness forms in the sample, care must be taken not to overgeneralise.

The omission from CAM, MELD, and OX of accessibly, an adverb formed with the Level 3 affix -ly could suggest inconsistent coverage. However, the presence of the thirteen other -ly adverbs from the sample suggests another factor, possibly frequency, plays a role.

The wordforms ethnicity and minimization are notably absent from LONG. There are four other occurrences of -ity, and two other occurrences of -ation sample wordforms covered by the website. Since words formed with -ity often have a specialised meaning which "may be hard to deduce" (Bauer and Nation 1993: 275), the omission of ethnicity is unfortunate. The omission of minimization here is surprising given its frequent semi-technical uses. While the omission of these words formed with often challenging Level 4 affixes could be a simple mistake, it may still inconvenience users.

Beyond coverage, there is less consistency in the way sample items are treated across the websites. One source of confusion is the ambiguous status of -ed and -ing forms which can be analysed as either adjectives or participle forms and in the case of -ing also as nouns. Schmitt and Zimmerman (2002) label the -ed forms (accessed, assumed, authorized, released, and selected) and the -ing form (surviving) as adjectives.

Table 5 shows searches for these -ed and -ing forms give inconsistent results. All sites redirect searches for the items accessed and released to access and release (v). No adjectival senses of these items are given. The adjective sense of selected is a sub-entry of the verbal sense from the LBD. The adjective assumed is listed as a main entry in CD and OX. The adjective authorized is present as a main entry in all the dictionaries except MELD. The adjective surviving is present as an entry or sub-entry in all dictionaries except OX. However, there are examples and collocations for the verbal entry which could be analysed as adjectival.

 

 

Some of these deficiencies are mitigated, intentionally or otherwise, by features of online dictionaries. Problems with corpus methods in lexicography often stem from inaccuracies in part-of-speech tagging (Frankenberg-Garcia, Rees and Lew 2021). Many methods tend to treat -ed forms as verbs rather than adjectives. This may explain the tendency to treat these forms as participles in the dictionaries. However, it also means that some of the corpus-derived examples in verbal entries could be analysed as adjectives. For instance, the example provided for the fifth sense of release (v) in OX: "The newly released files reveal [...]". This is more apparent still in automatically retrieved examples from corpora. For example, in the entry for release in CAM: "To what extent the rural sector absorbs the released labour is not clear" and "There are only a few landraces and very old released varieties available."

Helpful features include the alphabetical index adjacent to entries on all websites except LONG. For example, on CAM's page for assume, the user is presented with adjectival uses: assumed debt, assumed liabilities, assumed name in the 'Browse' box at the bottom of the entry. For years, liberation from the constraints of the alphabetical index has been regarded positively (c.f., De Schryver 2003). However, this feature can mitigate a methodological deficiency in electronic lexicography. Predictive text searches also help users find adjectival senses. For example, in MELD typing assumed predicts assumed name which is listed as a discrete entry. MELD also contains a crowd-sourced example containing an adjectival use of authorized, authorized push payment. An example of a crowd-sourced element potentially resolving a deficiency, albeit a relatively minor one, in a professionally produced dictionary.

The prevalence of homographs in English is problematic for electronic lexicography. Table 6 indicates the word class initially displayed when searching for a homographic item. Dictionary search engines cannot determine the user's intended word class. The basic form's ordering might reflect the compilers' view of the primary form or merely the frequency of word classes in the corpora used.

 

 

This coverage analysis provides insights into how members of derivationally related WFs are treated in online monolingual English dictionaries. Overall coverage statistics suggest reasonably consistent treatment of the WF members sampled. Inconsistencies include: the omission of forms with the morpheme -ness (liberalness and preciseness) which could be justified by its formal and semantic regularity, inconsistent treatment of rare wordforms which have more frequent equivalents (coherency and philosophic), and the ambiguous word class of -ed and -ing wordforms. These minor inconsistencies may not have an impact on the user. Furthermore, electronic lexicography methods both contribute to and mitigate such inconsistencies.

Writing support. The coverage analysis above suggests that members of derivationally related WFs are well covered on the websites examined (RQ1). However, to establish the extent to which they are treated in a way which facilitates productive use in writing (RQ2) a finer-grained analysis is necessary. A key assumption here is that examples, grammar patterns, and collocation information help writers. Another assumption is that the six rare forms with more frequent counterparts can be disregarded since it is likely that users will look up the more frequent counterpart.

The writing support scores in Table 7 suggest that OX provides the most comprehensive writing support for the problematic wordforms, closely followed by CAM and CD. MELD's score is notably lower than the others. This relation holds for the exclusive scores. However, caution is needed when interpreting differences in such a small sample. These scores indicate inconsistency in how the sample is treated across the websites examined. A closer examination of the individual components of writing support (examples, grammar patterns, and collocation information) confirms this impression and elucidates differences in sample treatment within dictionary websites.

 

 

The proportion of items with examples (Figure 6) differs significantly by website, X² = 30.068; df = 4; p < 0.001. Both OX and CAM provide examples for 97% of the items they cover. Items missing examples are the ambiguous word class forms accessed (CaM and OX), assumed (CaM), and selected (OX). Since the dictionaries treat them as verbs and provide examples for the verbal senses, they effectively offer examples for all items they list. CD provides examples for 88% of items covered. Again, two ambiguous class items (accessed and released) lack examples. LONG provides examples for 86% of items it covers including accessed, assumed, and released. The outlier here is MELD where 70% of items covered have examples.

 

 

There is clear inconsistency in the provision of grammar patterns on the websites examined (Figure 6). The proportion of items with this information differs significantly by website, X² = 14.2796; df = 4; p < 0.006. Overall, CD leads providing information for 60% of items covered. LONG provides grammar information for 53% of items covered, followed by OX (49%). CAM provides grammatical information for 39% of covered items, MELD for 36%. The syntactic behaviour of different parts-of-speech poses different degrees of challenge for writers. However, this trend persists when adverbs are excluded. For example, with a coverage statistic of 72% CD is notably higher than OX (55%), LONG (53%), and CAM (51%), and considerably more so than MELD (42%). When adjectives and adverbs are excluded, OX has the highest statistic (72%) followed by CD (71%) and CAM (70%); LONG covers 67% of noun and verb items, with MELD lower at 55%.

The proportion of items with collocation information (Figure 6) differs significantly by website, X² = 12.192; df = 4; p < 0.05. OX leads by providing collocation information for 68% of items covered, followed by CD (59%) and CAM (56%), then MELD (45%), and finally LONG (41%).

Since many users can induce information about grammatical patterns and collocational behaviour from dictionary examples and corpus lines, the relative absence of grammar patterns on CAM and OX is perhaps mitigated by their comprehensive example provision. This is reflected in the overall writing support score.

From the broad view adopted so far, considerable variation in the provision of writing support features between dictionaries is apparent. The following three sub-sections provide a finer-grained analysis of this variation.

Examples. As Table 8 indicates, after ambiguous word class forms, -ly adverbs are the wordforms most frequently missing examples. In general, they are semantically regular "Xly means 'in an X fashion' for any adjective X" (Carstairs-McCarthy 2018: 20). This general rule applies to coherently, ethnically, minimally, and persistently (all lacking examples in MELD). However, accessibly, authoritatively, ideologically, and philosophically are edge cases. For instance, without an example learners lacking deep relational knowledge could conceivably make the erroneous connection philosophy → philosophical ('related to philosophy') → philosophically (in a 'manner related to philosophy') rather than the prototypical meaning "in a way that calmly accepts a difficult situation" (CAM).

 

 

The high degree of productivity and semantic regularity of the affix -ness which generally means "'property of being X', where X is the base adjective." (Carstairs-McCarthy 2018: 78) could explain the omission of liberalness from all resources except CD and preciseness from LONG and OX, and the omission of an example for liberalness (CD) and preciseness (CD and MELD). However, the presence of examples for these items in the other resources suggest their creators do not share this assumption of relational knowledge.

The lack of examples for liberalization (MELD) and minimization (CD and MELD) can be explained by the generalizability of -ation. However, as with the absence of minimization from LONG, both wordforms have a specialised meaning frequent in academic context (e.g., "He is a longtime proponent of his country's economic liberalisation." (CAM); "cost minimization" (OX)). The absence of liberalize from MELD is notable for the same reason (e.g., "They will work with a view to further liberalize the investment regime" (CAM)). Like the absence of an example for philosophically discussed in the coverage analysis above, the absence of an example for philosophize from MELD is problematic as it does not typically mean 'to create philosophy' rather "to talk for a long time about subjects such as the meaning of life" (CAM). An example could also demonstrate that, in contrast to many words derived with the affix -ize, it is intransitive. The following examples from CAM for the entries for the -ize forms sampled illustrate complementation patterns well:

I authorized my bank to pay her £3,000.

They have plans to liberalize the prison system.

We must minimize the risk of infection.

Students, she complained, had nothing better to do than spend whole days philosophizing about the nature of truth.

The provision of examples for wordforms derived with -ity is also problematic. The missing example for accessibility in LONG is surprising. Firstly, because examples exist in the other dictionaries and, secondly, because it has a specialised yet frequent sense: "how easy something is to reach, enter, use, etc. for somebody with a disability" (OX). Additionally, examples for ethnicity and liberality are missing from MELD. As Schmitt and Zimmerman (2002) show, the extent to which productive knowledge of these words is easily predictable from productive knowledge about their base is questionable. When the base has two or more senses this assumption of relational knowledge entails a further assumption: that the user knows which sense is relevant to the derivative. For example, the definitions below come from CAM: (1) and (2) define ethnic, (3) defines ethnicity. The relation between (1) and (3) is immediately apparent. The relation between (2) and (3) requires some mental gymnastics.

1. relating or belonging to a group of people who can be seen as distinct (= different) because they have a shared culture, tradition, language, history, etc.:

2. seen as different or interesting because of coming from a culture or tradition that is not Western:

3. a large group of people with a shared culture, language, history, set of traditions, etc., or the fact of belonging to one of these groups:

Examples for the WF members coherence and cohere are notable omissions from MELD. This may stem from an assumption that learners have the relational knowledge to make the connection to the adjective coherent. This is particularly questionable in the case of coherence as although the Level 5 affix -ence is reasonably regular, it is not frequent (Bauer and Nation 1993: 260).

Regarding the provision of examples, there is clear inconsistent treatment across dictionaries, and in the case of CD, LONG and MELD, within dictionaries. Barring the ambiguous word class items, in CD and LONG the sub-entry status of items may be an explanatory factor for, or a consequence of, the missing examples. However, in MELD both main- and sub-entries lack examples.

Grammar patterns. A comparison of items with (Figure 7) and without (Figure 8) grammar pattern information suggests inconsistent writing support between and within websites.

 

 

 

 

As discussed, the need for grammar pattern information varies by word class. All sampled adverbs given grammatical support in CD have main entry status. Their grammar patterns come from COBUILD. Many first appeared in the 'extra-column' of the paper dictionary (Hands 2018) and were migrated online. Wordforms lacking grammar pattern support occur as "derived words" and sub-entries in other Collins dictionaries such as Collins English Dictionary and Webster's New World College Dictionary. Similarly, entry status explains the presence of grammatical information for adverbs in LONG. Those with support are the "Sentence adverbs", inevitably and traditionally, and precisely. The latter is followed by the interrogative pronouns how/when/where. All adverbs lacking grammar support in LONG, except liberally, are sub-entries. In MELD, precisely is also listed followed by how/when/what and in OX it followed by because. All other sampled adverbs in the latter two dictionaries lack grammatical pattern information.

Adjectives selecting prepositions (accessible to, liberal with, minimum of etc.) are treated fairly consistently. Inconsistencies occur in CD, LONG, and OX, which mark typical word order for some adjectives (e.g., "precise [adj NOUN]") but not others with the same order (e.g., coherent). Dictionaries that do not indicate this order (e.g., CAM and MELD) offer less detailed yet more consistent treatment.

All sampled verbs in CD have grammar pattern information. Patterns for cohere are absent from CAM, LONG, and MELD. Of the -ize affixed verbs, only liberalize has pattern information in CD, while patterns for philosophize are absent in LONG, MELD, and OX. As discussed, grammar pattern information may be useful for learners wishing to use philosophize as it is a rare example of an intransitive verb derived with -ize which frequently occurs with the prepositions of or about, as documented in CAM and CD. Similarly, cohere with is a typical pattern given in OX and CD.

Nouns are derived using a greater variety of affixes than other word classes. Table 9 shows the sample nouns included on each website and whether they have grammar pattern information. The overall impression is one of inconsistent treatment within and between dictionaries.

 

 

Wordforms without grammar pattern information are predominantly derived by affixation using -ity and -ation. Those that do have grammar patterns can be analysed as the base wordforms or are often the most frequent member of their family according to Schmitt and Zimmerman's (2002) counts. The usefulness of grammar pattern information for these items to writers can only be ascertained by direct empirical research. However, it is notable that producing these word-forms posed problems for Schmitt and Zimmerman's (2002) participants.

Some items missing grammatical patterns exhibit similar grammatical behaviour to those which have them. For example, assumption that appears in all resources while inevitability that is absent form CAM and MELD. This suggests a need for grammatical pattern information for many items missing it. Like the provision of examples, many of the wordforms without grammatical pattern information were treated as subentries, irrespective of their word class.

Collocation information. Unlike closed classes or phrasal categories that constitute grammar patterns, the range of potential collocates is limitless. Variation in typical collocates presented for a given base between resources is expected due to variation in corpus composition. Consequently, this analysis of collocation information must adopt a broad focus.

The provision of collocation information does not follow the general trend for writing support in the dictionaries examined. Notably, LONG rather than MELD provides collocation information for fewest items. However, differences exist across word classes.

Collocation information is absent for three out of twenty-three noun items (Table 10) in all resources: ethnicity, liberalness, and preciseness. Five items (accessibility, liberality, liberalization, and minimization in CAM; and inevitability in COD) only have it in one resource. In contrast, seven items are absent from one resource (minimum and selection from COD; authorization, coherence, ideology from LONG; and philosophy and precision from MELD).

 

 

All eleven verb items have collocation information in at least one resource (Table 11). Although, for cohere and philosophize, this information is only provided by CAM. This is problematic because it assumes relational knowledge with other family members. Three resources lack information for liberalize (CD, LONG, and MELD) and minimize (CAM, CD, and OX). As with examples, some academic writers might benefit from collocation information about these semi-technical terms.

 

 

Ostensibly, provision of collocation information for adjectives is less comprehensive than for nouns and verbs (Table 12). However, seven of the items missing collocation information are ambiguous word class items treated as verbs. Moreover, three resources lack information for ideological (CAM, LONG, and MELD) three for authoritative (LONG, MELD, and OX), two for coherent (CAM and LONG) and philosophical (LONG and MELD).

 

 

Provision of collocation information for adverbs is the least comprehensive of all word classes (Table 13). Information is provided for precisely in all resources except CAM. CD also provides information for liberally and persistently, OX for authoritatively and selectively, and CAM for minimally. Two factors may explain this sparse coverage: Firstly, the suffix -ly is extremely semantically regular "Xly means 'in an X fashion', for any adjective X." (Carstairs-McCarthy 2018: 20), so presumably lexicographers assume users can use the -ly adverbs in production by connecting them to their knowledge of the adjective base. Secondly, users are unlikely to start a collocation search using an adverb: "It would not make sense for a writer to initiate a collocation query from an adverb (e.g. 'what words can I use with primarily?')" (Frankenberg-Garcia et al. 2019: 28).

 

 

This analysis of grammar support features for derivative forms suggests examples, grammar patterns, and collocations work independently when supporting writers. This is unlikely; writers may take information simultaneously from all three sources. If one feature (e.g., grammar pattern information) is unavailable they may rely more heavily on another (e.g., examples). Future analysis of writing support would benefit from a model reflecting this relationship.

 

Conclusions

This study aimed to investigate the treatment of academic WFs on five English dictionary websites frequently used by learners. It was motivated by a belief that the members of these WFs should be treated in a way that facilitates learners' written production. Two factors prompted this belief: Firstly, research demonstrating that when given a basic prompt wordform, academic writers struggle producing derivative forms from the same WF. Secondly, the removal of space restraints in electronic resources, which hypothetically allows more detailed coverage of derivatives than paper-based dictionaries.

Overall, the five websites examined cover most items in the sample of challenging wordforms. This good coverage contrasts with findings on paper-based dictionaries. However, as in previous research, there is considerable variation in the treatment of derivative wordforms within and between resources.

The quantity of writing support features varies greatly across websites. Although MELD covers a high proportion of sample items, it provides fewer examples, grammar patterns, and collocation information than the other resources. Within resources, the reasons for inclusion or exclusion of items and their related writing support features are not always clear. For certain affixes, this may be due to assumptions about generalisability of their semantic or syntactic behaviour. These assumptions may be misguided since empirical research suggests writers do not always connect bases and derivatives formed by suffixation even with highly generalisable and productive affixes. Occasionally, (e.g., ethnicity, liberalization), analysis of the excluded wordforms suggests their semantic relationship to the base is idiosyncratic. Alternatively, their relative frequency in corpora used in compilation may explain exclusion. Further investigation here would be beneficial.

Further research could also mitigate limitations restricting the generalisability of these conclusions. Important limitations relate to the 74 problematic word-forms investigated. Not only is this sample small, but its items are also morphologically limited containing a relatively narrow range of suffixes. Future research should investigate forms created via prefixation (e.g., with co-, in-, re- etc.) if producing these is found to be a problem for writers.

Practical considerations for dictionary makers. Assumptions about users' relational knowledge of WF members should be reevaluated. Instead of assuming that writers can connect the base, the affix and derivative meaning, dictionary makers should aim for more complete treatment of derivatives. Electronic resources, unrestrained by the physical restrictions of paper-based dictionaries, could offer users fuller entries for derivative forms. However, compiling dictionary entries costs money. Deprived of income from sales of paper dictionaries, it is unlikely that publishers will invest in this. Nonetheless, as seen with corpus lines and collocation lists, methods from electronic lexicography can, sometimes inadvertently, offer a solution.

 

Endnotes

1 For a more nuanced view, see Lew (2011) who makes a distinction between the potentially unlimited storage space for lexicographic data and more limited presentation space on the user's screen.
2 Macmillan English Dictionary online was shut down on June 30th, 2023.
3 The edition of the LBD from which the entry is taken is not specified.
4 The editions of COBUILD and CED from which the entries are taken are not specified.
5 The dictionaries mentioned are Cambridge International Dictionary of English (Procter 1995), COBUILD English Learner's Dictionary (Sinclair 1989), Longman Dictionary of English Language and Culture (Summers 1992), and Oxford Advanced learner's Dictionary of Current English (Crowther 1995).
6 This would be surprising; "we checked the frequency of these derivatives in the BNC and considered eliminating those that had very low frequency counts or did not exist in the corpus." (Schmitt and Zimmerman 2002: 156)

 

References

Agnes, M. (Ed.). 2010. Webster's New World College Dictionary. Fourth Edition. Cleveland, Ohio: Wiley.         [ Links ]

Bauer, L. and P. Nation. 1993. Word Families. International Journal of Lexicography 6(4): 253-279.         [ Links ]

Berko, J. 1958. The Child's Learning of English Morphology. Word 14(2-3): 150-177.         [ Links ]

Carstairs-McCarthy, A. 2018. An Introduction to English Morphology: Words and Their Structure. Kindle Edition. Edinburgh: Edinburgh University Press.         [ Links ]

Coxhead, A. 2000. A New Academic Word list. TESOL Quarterly 34(2): 213-238.         [ Links ]

Crowther, J. (Ed.). 1995. Oxford Advanced Learner's Dictionary of Current English. Fifth Edition. Oxford: Oxford University Press.         [ Links ]

De Caluwe, J. and J. Taeldeman. 2003. Morphology in Dictionaries. Van Sterkenburg, P. (Ed.). 2003. A Practical Guide to Lexicography: 114-126. Amsterdam: John Benjamins.         [ Links ]

De Schryver, G.-M. 2003. Lexicographers' Dreams in the Electronic-Dictionary Age. International Journal of Lexicography 16(2): 143-199.         [ Links ]

Frankenberg-Garcia, A. 2015. Dictionaries and Encoding Examples to Support Language Production. International Journal of Lexicography 28(4): 490-512.         [ Links ]

Frankenberg-Garcia, A., R. Lew, J.C. Roberts, G.P. Rees, and N. Sharma. 2019. Developing a Writing Assistant to Help EAP Writers with Collocations in Real Time. ReCALL 31(1): 23-39.         [ Links ]

Frankenberg-Garcia, A., G.P. Rees and R. Lew. 2021. Slipping Through the Cracks in e-Lexicography. International Journal of Lexicography 34(2): 206-234.         [ Links ]

Hands, P. 2018. COBUILD Design and Layout: Changes over the Last 30 Years. Collins Dictionary Language Blog. https://blog.collinsdictionary.com/language-lovers/cobuild-design-and-layout-changes-over-the-last-30-years/ [30 July 2023]

Hanks, P. 2013. Lexical Analysis: Norms and Exploitations. Cambridge, MA: MIT Press.         [ Links ]

Hyland, K. and Polly Tse. 2007. Is There an "Academic Vocabulary"? TESOL Quarterly 41: 235-253.         [ Links ]

Kilgarriff, Adam, Miles Husák, Katy McAdam, Michael Rundell and Pavel Rychlý. 2008. GDEX: Automatically Finding Good Dictionary Examples in a Corpus. Bernal, Elisenda and Janet DeCesaris (Eds.). 2008. Proceedings of the 13th EURALEX International Congress, Barcelona, 15-19 July 2008: 425-432. Barcelona: Institut Universitari de Lingüistica Aplicada, Universitat Pompeu Fabra.

Lew, R. 2011. Space Restrictions in Paper and Electronic Dictionaries and their Implications for the Design of Production Dictionaries. Banski, Piotr and Beata Wójtowicz (Eds.). 2011. Issues in Modern Lexicography. München: Lincom Europa.

Procter, P. (Ed.). 1995. Cambridge International Dictionary of English. Cambridge: Cambridge University Press.         [ Links ]

Ptasznik, B. 2023. More Examples May Benefit Dictionary Users. International Journal of Lexicography 36(1): 29-55.         [ Links ]

Rees, G.P. 2021. Discipline-Specific Academic Phraseology: Corpus Evidence and Potential Applications. Charles, M. and A. Frankenberg-Garcia (Eds.). 2021. Corpora in ESP/EAP Writing Instruction: Preparation, Exploitation, Analysis: 32-54. London: Routledge.         [ Links ]

Schmitt, N. 1998. Tracking the Incremental Acquisition of Second Language Vocabulary: A Longitudinal Study. Language Learning 48(2): 281-317.         [ Links ]

Schmitt, N. 1999. The Relationship between TOEFL Vocabulary Items and Meaning, Association, Collocation and Word-class Knowledge. Language Testing 16(2): 189-216.         [ Links ]

Schmitt, N. and P. Meara. 1997. Researching Vocabulary through a Word Knowledge Framework: Word Associations and Verbal Suffixes. Studies in Second Language Acquisition 19(1): 17-36.         [ Links ]

Schmitt, N. and C. Zimmerman. 2002. Derivative Word Forms: What Do Learners Know? TESOL Quarterly 36(2): 145-171.         [ Links ]

Sinclair, J. (Ed.). 1989. COBUILD English Learner's Dictionary. London: Collins.         [ Links ]

Stein, G. 1985. Word-formation in Modern English Dictionaries. Ilson, R. (Ed.). 1985. Dictionaries, Lexicography and Language Learning: 35-44. Oxford: Pergamon.         [ Links ]

Summers, D. (Ed.). 1992. Longman Dictionary of English Language and Culture. Harlow: Longman.         [ Links ]

Ten Hacken, P., A. Abel and J. Knapp. 2006. Word Formation in an Electronic Learners' Dictionary: ELDIT. International Journal of Lexicography 19(3): 243-256.         [ Links ]

Tyler, A. and W. Nagy. 1989. The Acquisition of English Derivational Morphology. Journal of Memory and Language 28(6): 649-667.         [ Links ]

Van Sterkenburg, P. 1992. Het Woordenboek der Nederlandsche taal: Portret van een taalmonument. The Hague: Sdu.         [ Links ]

West, M. 1953. A General Service List of English Words. London: Longman, Green & Co.         [ Links ]