Patterns, Constructions, and Local Grammar: A Case Study of ‘Evaluation’

Abstract

This article takes as its starting point the analysis of adjective complementation patterns and sets this in the context of other studies of phraseology, especially Construction Grammar. The article proposes that a large number of meaning–pattern combinations can be identified as constructions. This endeavour assists and is assisted by the derivation of a local grammar of evaluation. The grammar is made explicit in 22 ‘Analyses’, grouped into five main categories. It includes discussion around the process of mapping meaning on to pattern, the consideration of borderline cases, and the debate around naming of elements. This contributes to a comparison of different approaches to phraseology, and in particular to the place of Construction Grammar in relation to more output-oriented approaches. Suggestions for the application of this approach to language teaching are offered.

1. INTRODUCTION

In this article we offer an updated reinterpretation of the notion of grammar patterns (Hunston and Francis 1999) in terms of Construction Grammar (Goldberg 1995, 2006). We argue that each of the meaning–pattern combinations identified in Francis et al. (1996, 1998)¹ can be regarded as a construction, yielding approximately 1,000 constructions at the same level of specificity. Furthermore, as the component elements of each construction can be annotated with functional labels, those constructions that perform an identifiable speech act function can be interpreted in terms of a local grammar (Barnbrook 2002). The semantically labelled constructions, we argue, can be applied to the development of resources for language teaching and may have further applications to the automatic processing of text. This argument is illustrated with a case study of the language function of evaluation. Specifically, the article proposes that the complementation patterns of adjectives (Francis et al. 1998) can be used to identify evaluative constructions and that these constructions in turn may be annotated to derive a local grammar of evaluation.

The article is organized as follows: following this introduction, the key terms used in the article are defined, and examples of previous research given; the study that underpins this article is then reported, and 22 analyses around the concept of evaluation are proposed; the relationship between pattern, construction, and local grammar is then discussed in more detail, and potential applications for the study offered. The article ends with a conclusion pointing to future directions.

2. DEFINITIONS AND PREVIOUS RESEARCH

In this section the terms ‘pattern grammar’, ‘construction grammar’, ‘local grammar’, and ‘evaluation’ are defined and some of the previous research in these areas is summarized.

Pattern grammar (Francis 1993; Hunston and Francis 1999; Hunston 2015) is an approach to the grammar of English that generalizes from the patterning of individual words as observed through concordance lines from a large corpus of general English (cf Sinclair 1991, 2004). It was developed originally to encapsulate the grammatical behaviour of items in a learners’ dictionary (Sinclair 1995). Although the concept of a grammar pattern can be used to describe any words, the most cited grammar patterns specify the complementation of verbs, nouns, and adjectives. The grammar pattern coding used in Sinclair (1995) and subsequently in Francis et al. (1996, 1998) uses abbreviated symbols to stand for word classes or clause types. For example, it expresses verbs, nouns, and adjectives, or the groups of which they are head, by ‘v’, ‘n’, and ‘adj’, that-clauses by ‘that’, and to-infinitive clauses by ‘to-inf’. In cases where the pattern includes specific words rather than classes, these are conventionally indicated in italics. Mostly they are the prepositions ‘at’, ‘for’, ‘with’, etc. A string of symbols identifies the pattern, with the node word in capitals. For example, the pattern V n to-inf specifies that the verb (V) is followed by, and governs, a noun phrase (n) and then a to-infinitive clause (to-inf), as in … told us to go home. The pattern N from n indicates that the noun is followed by, and governs, a prepositional phrase beginning with from, as in … recovery from the steep recession … The pattern it v-link ADJ that indicates that the adjective (ADJ) is preceded by an introductory it and a link verb and is followed by a that-clause, as in It now seems certain that St Paul’s Cathedral will not be surrounded … . In total, about 200 grammar patterns are identified (see Francis et al. (1996, 1998) for more exemplification). These books are available from 2018 as an online resource (www.collinsdictionary.com).

Grammar patterns relate to form only, unlike, for example, the Corpus Pattern Analysis proposed by Hanks (2013) and developed in the Pattern Dictionary of English Verbs (PDEV) project (www.pdev.org.uk). For example, whereas the entry for the verb ENCOURAGE in PDEV distinguishes between ‘humanencourageshuman’ (e.g. She laughed and encouraged him) and ‘eventualityencourageseventuality’ (e.g. a lack of public transport encouraged drink-driving), this distinction is not made in the pattern grammar nomenclature and both instances are coded V n (‘verb followed by noun phrase’). This means that the grammar patterns are less informative than the PDEV entries. On the other hand, the grammar patterns offer a level of generality associated with a ‘grammar’, and further semantic information is given in two pattern grammar resource books (Francis et al. 1996, 1998) (see footnote 1). In these publications, the words that occur with each pattern are listed in groups based on shared meaning. For example, the pattern V n to-inf lists 219 verbs divided into 12 groups, including two groups connected with verbal processes (e.g. ask, tell; encourage, urge), one connected with ‘causation’ (e.g. cause, compel, oblige), and one connected with ‘helping’ (e.g. aid, enable, help). The identification of the words in each pattern is based on lexicographical work undertaken as part of the Cobuild project in the 1990s (cf Sinclair 1995), though the online grammar pattern resource (see footnote 1) includes substantial updating (cf Francis 2015). Groups were identified on the basis of a ‘common sense’ and largely atheoretical approach to word meaning (Hunston and Francis 1999).

As another example, the pattern it v-link ADJ that lists 245 adjectives divided into eight groups relating to: ‘likelihood’, ‘obviousness’, ‘desirability’, ‘undesirability’, ‘importance and necessity’, ‘interest and surprise’, ‘relevance’, and ‘other’. It is immediately apparent that all the specified meanings relate to the domain variously termed ‘stance’, ‘attitude’, or ‘evaluation’. Indeed, it is found that the majority of adjectives identified as governing complementation patterns have evaluative meanings, and thus the case study in this article relates to evaluative meaning. Many of the adjectives covered by our analysis, such as happy, said, astonished, afraid, appear also in studies of Affect (Martin and White 2005; Bednarek 2008). The ‘Affect’ category in Martin and White’s taxonomy distinguishes personal emotion from appraisal of a target, the latter being covered by Judgement and Appreciation. In those cases where the adjective expressing emotion is complemented by a further element, as in Anne was afraid that John would soon be sent abroad, two analyses are possible: ‘emotion + stimulus’ or ‘evaluation + target’. For the purposes of a case study of evaluation, where only adjectives with complementation patterns are being considered, the second analysis is more relevant, though the first remains a valid alternative.

Turning now to construction grammar: this is an approach to the description of language patterning that has much in common with pattern grammar but that grew up within the traditions of Cognitive Linguistics rather than in the traditions of Corpus Linguistics, and until recently there has been little dialogue between the two (though see Ellis et al. (2016) for an exception). Corpora are increasingly used as evidence for constructions as they are for patterns, but whereas patterns are perceived as purely observational phenomena, constructions are an attempt to model the mental representation of language. Dąbrowska (2015), for example, offers construction grammar as a valid alternative to universal grammar, and Ellis et al. (2016) use corpus evidence to demonstrate the acquisition of verb complementation constructions by learners of English.

Constructions are a matching of form and meaning at all levels of generalization. The most basic definition of a construction includes the proviso ‘some aspect of its [the linguistic pattern’s] form or function is not strictly predictable from its component parts or from other constructions’ (Goldberg 2006: 5). Examples would include idioms such as ‘jog someone’s memory’ (ibid.). However, it is also proposed that ‘patterns are stored as constructions even if they are fully predictable as long as they occur with sufficient frequency’ (ibid.). This permits the pattern–meaning combinations proposed in this article (see below) to be candidate constructions.

An important aspect of constructions is that although typical lexis can be identified in each construction (see Stefanowitsch and Gries 2003; Gries and Stefanowitsch 2004 for extended discussion), meaning belongs to the construction rather than to the lexis. Goldberg (2006: 6) illustrates this with examples such as She smiled herself an upgrade, where the meaning ‘make something happen that is of benefit to oneself’ is construed by the construction ‘verb oneself something’ rather than by the verb SMILE. Bencini and Goldberg (2000) test the effects of verb and construction on the perception of sentence meaning and conclude that construction has the greater effect. Some studied constructions are of a high level of specificity, such as the ‘accident waiting to happen’ construction (Stefanowitsch and Gries 2003); others are very general, such as the ‘interrogative’ construction or the ‘ditransitive’ construction (Stefanowitsch and Gries 2003; Goldberg 2006). The multi-level approach of construction grammar is both a benefit and a disadvantage. On the positive side, all of lexis and grammar can be described in a single model, without the need for an elaborate system of grammatical levels or ranks (as, for example, in Halliday’s (1985) model). Constructions might even be said to respond to Hasan’s (1996) vision of lexis as the most delicate grammar and certainly coincide with Sinclair’s vision of a description of English that does not presuppose a division into lexis and grammar (Sinclair 1991: 3) or with many of Hoey’s observations of lexical priming (Hoey 2005). On the negative side, the number of potential constructions is vast, and a listing of them all seems an impossible task. Studies of constructions tend to treat specific examples which are convincing in terms of the concept of ‘construction’ but which do not progress towards a systematic description of a language (though see Wible and Tsao (2017) for a proposal for how this systematicity might be achieved).

Of particular interest to this article are what might be called the ‘mid-level constructions’ (that is, neither very general nor very specific) such as the ‘verb someone into doing something’ (or causative ‘into’ construction) investigated by Wulff et al. (2007), which are very like grammar patterns. Indeed, a number of studies (Mateu 2005; Hiltunen 2010) have presented candidates for constructions that are indistinguishable from patterns. On the other hand, it is clearly not the case that ‘construction’ is directly equivalent to ‘pattern’. For example, as shall be illustrated further below, the ADJ at n pattern includes examples such as Those new to the area were always astonished at the vivid crimson of the earth, which might be said to represent a ‘reaction at’ construction with 45 adjectives listed in Francis et al. (1998), and examples such as She was not very good at writing letters, which might be said to represent an ‘(un)skilled at’ construction with 30 adjectives listed in Francis et al. (1998). This article offers a way of integrating pattern and construction; it proposes, not that each pattern is a construction, but that each meaning–pattern combination is a construction. This would suggest that the lists of grammar patterns to be found at www.collinsdictionary.com provide evidence for approximately 1,000 constructions at a given level of specificity. We argue that this goes some way to addressing the drawback to construction grammar suggested above. The candidate constructions we propose, however, are based on corpus investigation alone; we have no evidence as to whether they are stored as constructions in the minds of speakers.

This article also makes extensive use of the concept of local grammar. A local grammar, as the term is used in this article, is always a grammar of a discourse function. (This distinguishes these local grammars from Sinclair’s (2007/2010) suggestion for a local grammar of a word.) It is therefore closely related to performative speech acts. One of the first local grammars in this sense was Barnbrook’s (2002) pioneering local grammar of the definitions used in the Collins Cobuild Student’s Dictionary (CCSD) (Sinclair 1990). Other examples include grammars of requests (Su 2017), apologies (Su and Wei in press), disclaimers in company reports (Cheng and Ching 2016), and Affect (Bednarek 2008). In all these studies, a recurring sequence of forms is identified, and functional labels are mapped on to that sequence. The task of the researcher, then, is to specify the function, the way(s) in which that function is realized (as lexis and grammar), and the functional labels needed to annotate the representative examples. Barnbrook (2002: 135–6), for example, identifies four types of definition in the CCSD and 17 sub-types—an illustration, incidentally, of the heuristic value of local grammar identification. The functional labels he employs include ‘Definiendum’ (the defined word or phrase), ‘Definiens’ (the explanation or definition), ‘Hinge’ (a grammatical operator linking the Definiendum and the Definiens), and ‘Co-text’ and ‘Matching Co-text’ (additional explanatory elements mirrored in the two halves of the definition). Table 1 gives an example: the CCSD entry for life imprisonment (Barnbrook 2002: 173).

Table 2 shows an example from Cheng and Ching (2016), demonstrating the mapping of the functional labels (‘Creator of disclaimer’, ‘Thing denied’, ‘Restriction on denial’, and ‘Hinge’) on to the pattern elements (‘noun group’, ‘verb’, ‘to-infinitive clause’, etc.).

These instances also illustrate a key point about local grammars: they depend upon the identification of the sentence being analysed as an instance of the chosen function. For example, a sentence with the same grammatical structure as the one in Table 1, such as the invented When criminals are sentenced to life imprisonment, they are sent to a high-security prison, does not have the function of ‘definition’, and therefore the labels used by Barnbrook are not appropriate (‘a high-security prison’ is not the Definiens and ‘life imprisonment’ is not the Definiendum). This is an obvious restriction on the usefulness of local grammars for the automatic extraction of information in text, and indeed for language teaching. For Barnbrook, this is not an issue, as his corpus consists only of definitions from the CCSD. Cheng and Ching (2016) start by manually identifying all disclaimers in their corpus; in doing so they identify a restricted set of vocabulary items (such as obligation, commitment, reflect) which could be used to target disclaimers in a larger corpus that had not been pre-processed in this way.

As noted above, local grammars of the type pioneered by Barnbrook account for the meaning elements involved in performing a speech act: giving a definition, making an apology or a request, or disclaiming responsibility. In these cases the selection of local grammar terminology is justified by the speech act being employed. The concept of local grammar has been adopted more broadly, however (e.g. Warren and Leung 2016), in particular by Bednarek (2008) to describe the reporting (as well as the performing) of Affect (Martin and White 2005). Bednarek starts with the patterns of adjectives, nouns, and verbs used to report Affect and derives a local grammar expressed as a series of analyses, of which the first line in Table 3 is an example. As noted above, the analysis in this article focuses on the alternative ‘evaluation of target’ interpretation, and so relabels this example as shown in the final line in Table 3.

This leads us to the last in this list of definitions: the term ‘evaluation’ is used in this article to mean the expression of an attitude towards an entity (person, object, proposition, or situation). Unlike the expression of Affect, which may or may not have an explicit cause or trigger, evaluation, as used here, is always the evaluation of something. This accords with Thompson’s (2010: 402) view that ‘appraising must have a target’. The discourse function of evaluation has received increasing research interest in recent years, in part because it has a range of applications, from modelling for students how stance is expressed in academic discourse (Hyland 2005; Biber 2006), to quantifying positive and negative judgements of products from millions of online comments (Turney 2002; Su 2016), to identifying ideological stance in news reports (Partington et al. 2004; Bednarek 2016). Under various guises (‘stance’, ‘appraisal’, ‘sentiment’, for example), it has been studied using diverse methods including corpus searches for specific words or phrases (Conrad and Biber 2000; Hyland and Tse 2005a, b), qualitative discourse analysis (Martin and White 2005), and methods that combine the two (Charles 2006; Fuoli 2012; Partington et al. 2013; Trnavac et al. 2016; Partington 2017). Evaluative meaning is notoriously difficult to pin down, being cumulative (Hunston 2011: 3–4), often implicitly expressed (Martin and White 2005), and subject to embedding and nesting (Partington et al. 2013). Inevitably, local grammars of evaluation target only the most explicit expressions of that meaning; in this article, only evaluation which is expressed by adjectives occurring with complementation patterns is analysed. Countering that limitation, we can assert that this local grammar is based on a complete listing of all adjective complementation patterns in English and the listing of about 2,500 individual adjectives (Francis et al. 1998).²

To recapitulate the argument of this article: we use the notion of pattern grammar to propose form-meaning pairings, thereby contributing to research into construction grammar. More specifically we propose evaluative constructions, based on the lists of adjective patterns given in Francis et al. (1998). These constructions can be parsed and annotated with labels that relate them to the function of performing or reporting evaluation, thereby forming a local grammar of evaluation and contributing to research into evaluative meaning and its application.

3. METHOD: FROM PATTERN TO CONSTRUCTION

The data for the study are taken from the list of 44 adjective complementation patterns in Francis et al. (1998), which briefly comprise:

1. Adjectives followed by a that-clause, to-infinitive clause, wh-clause, or -ing clause (e.g. be amazed that; be cheap to (build); be aware how; be lucky (having)

2. Adjectives followed by a prepositional phrase (e.g. be good at; be heavy on; be liable to; be generous with)

3. Patterns with it (e.g. it is interesting that; it is fashionable to; find it absurd that)

4. Patterns with there (e.g. there’s nothing good about …)

The rationale for basing the study on adjective complementation patterns has been given above. The aim of the study is to account for examples for each of the adjectives and each of the complementation patterns in the Adjectives component of Francis et al. (1998), excluding only the minority of adjectives that do not express evaluative meaning. We proceeded pattern by pattern and group by group. For example, we find that the ADJ at n pattern has three meaning groups, with these rubrics (Francis et al. 1998: 428–30):

1. The ‘nervous’ group: These adjectives indicate that someone reacts to a situation or to an idea in some way, for example, by being surprised, happy, or unhappy. For example, aghast; agog; alarmed; amused; anxious; appalled; ashamed; astonished; astounded… (34 adjectives in total)

2. The ‘angry’ group: These adjectives indicate that someone is angry about a situation or an idea. For example, angry; annoyed; disgruntled; exasperated; furious; incensed… (12 adjectives in total)

3. The ‘good’ group: These adjectives indicate that someone does something well or badly. For example, adept; bad; brilliant; clever; competent; effective; efficient; excellent… (30 adjectives in total)

It is clear that whereas the first two groups share the meaning of ‘react to a situation’, the reaction being alarm, amusement, shame, surprise, or anger, the third group expresses a very different meaning. In other words, the form ‘ADJ at n’ matches with two meanings, depending on whether the adjective is of the ‘reaction’ type or of the ‘(un)skilled’ type. Thus, two form-meaning pairings, or constructions, are proposed, one with the meaning of ‘react at’ and the other with the meaning of ‘skilled at’. These might be designated the ‘reactive at’ construction and the ‘(un)skilled at’ construction. The distinction is supported by the exercise of local grammar analyses, that is by the mapping of meaning elements on to the examples; thus the work of building a local grammar facilitates the identification of construction. Examples (1) and (2) illustrate how the constructions differ in terms of the meaning–form mapping.

1. Phillip’s parents were annoyed at not being told the full story earlier

2. Some teachers may be adept at introducing their pupils to grammatical concepts

Example (1) illustrates the ‘reactive at’ construction and reports an evaluation carried out by Phillip’s parents, whereas Example (2), illustrating the ‘(un)skilled at’ construction, performs an evaluation (by the speaker) of ‘some teachers’. In each case the Evaluation is indicated by the adjective (annoyed and adept), but in Example (1) the Target is the object of the preposition ‘not being told the full story earlier’, whereas in Example (2) it is the subject of the clause ‘some teachers’. The construction exemplified in Example (1) may therefore be annotated as ‘Evaluator–Evaluation–Target’, whereas that exemplified in Example (2) is annotated as ‘Target–Evaluation–Action’ (the Action label will be discussed further below).

In the research reported in this article, this procedure has been repeated for each of the 44 patterns and for each meaning group in each pattern. Although the meaning groups are helpful in distinguishing types of meaning, it is borne in mind that they were compiled originally simply to present the adjective listings in a rational way; we have not considered ourselves bound by the groups in proposing constructions.

The ‘reporting’/‘performing’ distinction is important in all the patterns examined, and indeed most patterns can be interpreted in terms of a ‘person reacts to target’ construction and a ‘target is evaluated’ construction, though with different frequencies in terms of type. In the pattern ADJ that, for example, 9 of the 12 meaning groups (107 adjective types out of 115) represent the ‘person reacts to target’ construction, but in the pattern ADJ to-inf only 5 of the 17 groups (82 adjective types out of 260) do. In these cases, a large number of meaning groups can be said to instantiate the same construction; in ADJ that, for example, the ‘surprised’, ‘angry’, ‘horrified’, ‘glad’, ‘certain’, ‘aware’, ‘anxious’, ‘agreed’, and ‘consistent’ groups may be subsumed under the concept of ‘reaction’. In other cases, each group seems to demand a separate analysis. For example, the pattern ADJ for n can be interpreted as six constructions:

1. The ‘reactive for’ construction. For example, The people are impatient for change; We are grateful for being alerted…. The adjectives are found in meaning Group 3: desperate, eager, hopeful, impatient, ready, etc., and meaning Group 13: apologetic, grateful, guilty, sorry, thankful.

2. The ‘proxy for’ construction. For example, She was afraid for her son. The adjectives are found in meaning Group 7: afraid, concerned, fearful, worried, and meaning Group 8: ambitious, delighted, glad, happy, sad, sorry, thrilled.

3. The ‘purposive for’ construction. For example, Cylinder mowers are ideal for use on ornamental lawns. The adjectives are found in meaning Group 1: adequate, appropriate, brilliant, excellent, fine, good, great, ideal, inappropriate, wrong, etc.

4. The ‘specifying for’ construction. For example, The event is not suitable for children under ten; His team is ready for action; Modern facilities are not necessary for success; The hotel is convenient for the airport. The adjectives are found in meaning Group 1: suitable, unsuitable, etc., meaning Group 2: available, open, prepared, ready, ripe, etc., meaning Group 10: critical, crucial, essential, necessary, vital, and meaning Group 11: convenient, handy, inconvenient, practical, useful, etc.

5. The ‘affected for’ construction. For example, Sunshine is good for you. The adjectives are found in meaning Group 5: advantageous, bad, beneficial, costly, damaging, good, healthy, unfortunate, etc., and meaning Group 9: compulsory, mandatory, obligatory, optional.

6. The ‘reason for’ construction. For example, He is famous for his witty approach to design. The adjectives are found in meaning Group 4: celebrated, famous, legendary, notable, notorious, well-known, etc.

It will be noted that meaning Group 1 appears under two constructions, distinguishing between ‘onions are suitable for making into soup’ (the ‘purposive for’ construction) and ‘onions are not suitable for children under two’ (the ‘specifying for’ construction). Meaning Groups 6 (responsible for, etc.), 12 (pushed for time, etc.), and 14 (bound for Boston) are not included because they do not represent evaluative meaning as defined here.

As noted above, identifying local grammar meaning element labels contributes to the distinction between constructions. This can in turn be used to organize the very large number of constructions that is the consequence of this method of analysis; those patterns that share a local grammar analysis are grouped together. This is the next stage in the methodology. The aim is to arrive at as few analyses as possible, where possible fitting several patterns into the same analysis. As a consequence, there is rarely a one-to-one correspondence between pattern and analysis. The outcome of the procedure is a set of analyses, each annotated with labels contributing to a local grammar of evaluation.

The procedure followed here is unusual in two ways. Unlike most studies of evaluative language, original corpus analysis has not been carried out, and we are reliant on previous corpus research for our data. Secondly, we have chosen to proceed pattern by pattern rather than word by word in mapping meaning on to form. We believe there are advantages to these innovations. By using the outcome of previous research, we are able to take into account of many more individual words than is possible in other methods.³ Focusing on one pattern at a time throws the distribution of meaning elements across formal elements into sharp relief and facilitates the task of developing the local grammar, again enabling us to achieve greater coverage in our schema.

4. RESULTS: EVALUATIVE CONSTRUCTIONS AND A LOCAL GRAMMAR OF EVALUATION

We present the results of our investigation in a set of tables (Analyses 1–7). Each analysis brings together a number of constructions, each construction formed of an adjective complementation pattern and some of the sets of adjectives that are used with it. For example, Analysis 2a comprises 16 constructions, each consisting of a pattern and some of the sets of adjectives used with each pattern. For the pattern ADJ at n, for example, three such sets comprise the construction, with other sets contributing to a construction shown in Analysis 4a. It must be added, however, that this alignment of pattern and construction is open to debate. In Analysis 1, for example, it would be possible to propose a single construction, consisting of all the patterns summarized as it v-link ADJ clause. Pending further debate, then, the argument in this article is that the pattern plus selected sets of adjectives comprise the construction.

For reasons of space, the tables are kept as short as possible. The additional online resources give fuller tables, including all relevant patterns and kinds of evaluative meaning, though not all the adjectives listed in Francis et al. (1998). Each adjective in the tables represents other similar adjectives. For example, the online Analysis 1 lists seven adjectives in the pattern it v-link ADJ that. Each adjective stands in for the meaning group it comes from, so all the 243 adjectives listed with that pattern are accounted for by the analysis. All the examples in the tables are taken from Francis et al. (1998), though some have been shortened. In presenting the patterns and their coding we move from the most straightforward cases to the more complex or marginal. It will be noted that where the pattern includes a prepositional phrase, the preposition should strictly speaking be considered a Hinge rather than part of another element. For example, in They were keen on the idea of education, the Target of the reported evaluation is the idea of education, the Evaluator is they, and the Hinges are were and on. To avoid over-complicating the tables, however, the preposition is placed in the same column as the phrase that follows it.

The first set of examples (Analysis 1) includes only two substantive elements: the Evaluation (i.e. the evaluative adjective) and the Target (i.e. the entity or situation being evaluated). These examples perform an act of evaluation by the speaker and involve patterns with it and there. These patterns are well known as key indicators of overt evaluation, and all instances of these patterns fit the same analysis. These might be described as ‘the evaluative it construction’ and ‘the evaluative there construction’.

The second set (Analyses 2a–2c) reports evaluation by an Evaluator. Analyses 2a and 2b include the same elements—Evaluator, Evaluation, and Target—but the various patterns place the Evaluator in either subject (Analysis 2a) or object of preposition (Analysis 2b) position and the Target likewise in either subject (Analysis 2b) or object of preposition (Analysis 2a) position. Analysis 2c includes a further element: Proxy and represents the ‘proxy for’ construction mentioned above. Unlike Analysis 1, only some adjectives in each pattern fit this analysis. The proportions involved vary. For the pattern ADJ in n, for example, only a few adjectives (e.g. interested, confident) fit the analysis, but for the pattern ADJ of n, at least 70 adjectives do. In Analysis 2a, the adjectives are those which have been discussed in other contexts as realizing Affect. In some cases, as well as evaluation of the Target by the Evaluator, evaluation of the Evaluator by the speaker is implied. For example, he was dismissive of the idea reports ‘his’ feelings towards the ‘idea’, but also performs an evaluation of ‘him’ (see also Hunston 2011: 140). The more obvious examples of this layered evaluation are highlighted in italics in Analysis 2a, both here and in the online tables. However, it must be noted that the presence or absence of such multi-layering is not clear-cut and some subjective judgement is necessary here.

The examples in Analyses 3a and 3b also report, as opposed to perform, evaluation. Like the examples in Analysis 1 they include an introductory it, in object position in Analysis 3a and in subject position in the less common Analysis 3b. As well as the Evaluator, Evaluation, and Target elements they include an indicator (thought, see, regard) of the act of evaluation, labelled here Evaluative_act. Analyses 3a and 3b have the same elements but in a different order. Note that the patterns v it ADJ that and v it ADJ to-inf are used with verbs such as think (e.g. thought it curious that) and also verbs such as make (e.g. made it curious that). The patterns only fit this analysis when the verb is of the ‘think’ type.

We now turn to examples that present greater challenges in terms of their analysis, and where more extensive discussion is necessary. We first look at examples where, arguably, what is evaluated is an action rather than a person or thing. Analysis 4a shows the first set of these. There are a number of possible interpretations of these examples, each with a slightly different emphasis. These can be explained using possible paraphrases:

1. Example: I was daft going into management. Paraphrase: ‘I went into management and this action was daft’. Possible preferred coding: Actor + Evaluation + Action (where Actor + Action = Target)

2. Example: Mr Gates has been hugely successful in creating a world-beating business. Paraphrase: ‘Mr Gates has been successful and the reason is that he has created a world-beating business’. Possible preferred coding: Target + Evaluation + Reason

3. Example: She was good at raising money. Paraphrase: ‘She was skilful, but only in respect of raising money’. Possible preferred coding: Target + Evaluation + Restriction

Our proposed compromise between these possibilities is to have a simple coding of Target + Hinge + Evaluation + Action for each example, but to note that the Target is the Actor of the Action, and that the Evaluation covers ‘Target … Action’, as indicated in Analysis 4a. This analysis is somewhat contentious. Where the pattern involves a verb, either in a clause (e.g. ADJ to-inf: foolish to ignore them) or in an -ing clause following a preposition (e.g. good at raising money), the interpretation of Actor + Action is an obvious one. Analysing examples where the preposition is followed by a noun phrase (e.g. I was bad at Maths) in the same way is less secure. For the sake of consistency, however, I was bad at Maths is treated here as I was bad at doing Maths, hence fitting the same analysis.

There are a number of borderline cases which are excluded from this analysis. For example, the pattern ADJ in n includes a group of adjectives such as beneficial, helpful, useful, valuable (as in Celery seed extracts are helpful in the treatment of arthritis). The prepositional phrase indicates an action that the evaluated Target participates in, but as the action is performed by someone other than the Target, these are not seen as fitting this analysis. We also exclude examples such as Secrets are destructive of relationships (in the pattern ADJ of n), as although there is an action (‘secrets destroy relationships’), the action is indicated by the adjective, not by the prepositional phrase. These examples are assigned to Analysis 5 (see below).

As is well known (Francis et al. 1998: 404–5), the pattern ADJ to-inf can be used as in Analysis 4a, where the subject of the main clause is the same as the understood subject of the to-infinitive clause. For example, in We would be foolish to ignore them, ‘we’ is the implied subject of ‘ignore them’. The pattern can also be used as in Analysis 4b, where the subject of the main clause is the goal or object of the to-infinitive clause. For example, Watches have become more attractive to look at implies ‘someone looks at watches’. In the Analysis 4b examples, again, we face a dilemma of coding and again this can be exemplified with paraphrases:

1. Example: These machines are fiddly to clean. Paraphrase: ‘We clean the machines and the process is fiddly’. Possible preferred coding: Goal + Evaluation + Action (where Action + Goal = Target)

2. Example: Watches have become more attractive to look at. Paraphrase: ‘Watches are attractive, but only in respect of their physical appearance’. Possible preferred coding: Target + Evaluation + Restriction

Again we compromise with the coding Target + Hinge + Evaluation + Action, this time noting that the Target is the goal of the action, and that the Evaluation covers ‘Target … Action’. Note that, as in Analysis 4a, the action may be nominalized in the noun phrase following the preposition (e.g. use in for use on).

We now turn to the set of adjective–pattern combinations that present the most challenging situation. In the labelling shown in the previous tables, there is considerable uniformity in the mapping of semantic elements on to grammar pattern ones. This can be exemplified by looking at the v it ADJ that pattern in Analysis 3a. Francis et al. (1998: 506–9) list no fewer than 147 adjectives occurring with this pattern. They represent a variety of types or parameters of evaluation, including ‘good’ (e.g. effective), ‘bad’ (e.g. dreadful), ‘(un)true’ (e.g. plausible), ‘(un)usual’ (e.g. extraordinary), ‘important’ (e.g. essential), ‘(un)likely’ (e.g. certain), and ‘evident’ (e.g. clear).

Whatever the parameter, however, they all fit Analysis 3a. Similarly, in Analysis 2a, there is a variety of prepositions, and therefore constructions, but the mapping remains consistent. When carrying out the analysis of patterns, however, we encountered a great many instances where there is a Target and an Evaluation and then some other element that is less easy to identify at an appropriate level of generality or granularity. This difficulty arises with respect to adjectives followed by a propositional phrase. Consider, for example:

1. Police were vague about the gunman’s demands

2. Cream is also helpful against a dry flaky skin

3. The death penalty has proven worthless as a solution to crime

4. Success is achievable by anyone willing to work hard

5. It was not fair on them

6. The language is similar to Turkish

7. She felt drunk with strange emotions

In each case the role of the underlined element could be said to be specific to the adjective and the preposition: the topic of the vagueness in Example (3); the specific target of the cream in Example (4); the respect in which the death penalty is worthless in Example (5); the achiever of the success in Example (6); the people affected by the lack of fairness in Example (7); the similar language in Example (8); the cause of the feeling in Example (9). One solution is to propose a cover-all term, such as ‘Specifier’, or ‘Scope’. Another is to attempt a finer-grained analysis that would still achieve an element of generalizability. Analysis 5 (online) shows our proposed solution, which includes the general ‘Specifier’ label for some cases, but proposes more specific labels where these are possible. The underlined element in Example (3) is labelled ‘Topic’, in Example (4) it is ‘Specifier’, in Example (5) it is ‘Role’, in Example (6) it is ‘Actor/Method’, in Example (7) it is ‘Affected’, in Example (8) it is ‘Comparator’, and in Example (9) it is ‘Cause’.

Our final sets of evaluative examples (Analyses 6a–6c) account for a small number of less frequent patterns that combine it patterns with prepositional phrases (e.g. It is vital for him that he returns home soon) and where the mapping is once again straightforward.

Finally, there are a number of adjective–pattern combinations where the evaluation shades into other elements. For example, in she is adamant in her refusal, the adjective adamant offers an intensification of ‘her refusal’ rather than an evaluation of it (see Analysis 7a). In Its forests were abundant with wildlife, the adjective abundant quantifies the wildlife, in general terms, while still, arguably, assessing this as a positive characteristic of the forest (see Analysis 7b). There is, however, an overlap here between quantity and evaluation. The examples slow to learn and not big on tact could be included under Analysis 4a and 5, respectively. Beyond these scenarios we are outside the scope of evaluative meaning. For instance, a large number of adjectives followed by with or in indicate possession or presence, as in Every surface is scattered with photographs, and there are adjectives that behave rather like modal auxiliaries, such as liable to, as in The house is liable to problems.

5. DISCUSSION: PATTERNS, CONSTRUCTIONS, AND LOCAL GRAMMARS

The starting point for this article was a set of forms, specifically, adjectives and the complementation patterns that are dependent on them. These forms can be designated as ‘grammar patterns’. It has been proposed that the various combinations of pattern and meaning can be interpreted as constructions, though whether they are stored as such by language users remains to be investigated. In some instances, as noted above, there is a one-to-one correspondence between pattern and construction, as in the it v-link ADJ that pattern or ‘evaluative it’ construction. In most instances, however, there is a one-to-many correspondence, as in the ADJ at n pattern (the ‘reactive at’ construction or the ‘(un)skilled at’ construction) or the ADJ for n pattern, for which six constructions were proposed above. For the most part, the adjectives occurring with these patterns/constructions are evaluative in meaning, and it was hypothesized that it would be possible to draw generalizations about the mapping of evaluative meaning elements on to the various adjective patterns, leading to a local grammar of evaluation.

In the formulation of a local grammar, a number of meaning elements have been proposed. These are listed in Table 4. The elements in italics (from Role onwards) could be said to be finer-grained sub-divisions of the Specifier element. A total of six main analyses have been proposed, though there are 13 actual tables, and one analysis (Analysis 5) could be divided into eight separate tables. This is a manageable number and suggests that the right level of granularity has been achieved. We are confident that the analyses between them account for the vast majority of adjective + pattern combinations recorded in Francis et al. (1998) that have an evaluative meaning and that are therefore evaluative constructions, even though space permits the inclusion of a relatively small number of example adjectives in our tables.

We have stated above that one of the benefits of developing a local grammar is that it acts as a heuristic—a way of paying close attention to all instances of a given set of patterns. It also draws attention to the multiplicity of evaluative constructions that can be proposed based on adjectives and their complementation. Individual cases have been commented on above, but we summarize those comments and extend them here:

Patterns with it are highly predictable in the mapping of semantic elements on to formal ones (see Analyses 1, 3a and 3c, 6a–6c). Constructions of a more general or more delicate kind can be proposed, with the most general being ‘it is evaluation (prepositional phrase) clause/phrase’ (Analysis 1, 6a–6c), ‘THINK it evaluation clause’ (Analysis 3a), and ‘it STRIKE someone as evaluation that’ (Analysis 3b).

Where the adjective expresses Affect, then evaluation is reported rather than performed, with the subject of the clause realizing the Evaluator and the element following the adjective realizing the Target, or in rarer cases the Proxy (see Analyses 2a and 2c). The choice of clause type or preposition (happy about, angry at, annoyed that, etc.) depends on the adjective and the degree of nominalization. In the discussion above, it has been assumed that each meaning–preposition combination comprises a construction (the ‘reactive about’ construction, the ‘reactive at’ construction, and so on). A more general interpretation is that there is a form expressed as ‘Person + BE + Affect + Preposition + Entity’ or ‘Person + BE + Affect + clause’ which matches the meaning of ‘reaction to target’, comprising a single construction. These interpretations are not inconsistent but suggest that constructions exist at various levels of delicacy (Halliday 1985; Wible and Tsao 2017).

Then there are some patterns which realize only a small number of meaning possibilities (see also Su 2015) and therefore comprise a small number of constructions. Examples of these are:

1. The pattern ADJ to-inf is sometimes used with Affect adjectives, in which case it conforms to the situation discussed in the previous paragraph and appears in Analysis 2a. Where the adjective is not an Affect one, the pattern performs evaluation of an action or situation, as in We would be foolish to ignore them (‘We ignore them’; ‘That action is foolish’) or The party looks certain to win the election (‘The party will win the election’; ‘That situation is certain’) (see Analysis 4a). Where the subject of the main clause is not the understood subject of the to-infinitive clause, an action or situation is still evaluated, but the paraphrase must capture the difference in Actor, as in These shows are cheap to make (‘We make shows’; ‘Doing so is cheap’) or He was excellent to work with (‘We worked with him’; ‘That was an excellent situation’) (see Analysis 4b). Here, though, the consistency or reliability of the analysis comes into question. It could be argued that He was excellent to work with evaluates ‘He’ as ‘excellent’ and that to work with is a Specifier (as in Analysis 5). The line between the two interpretations is extremely blurred.

2. The meaning of the pattern ADJ about n seems to be governed by the meaning of about as an indicator of topic. This is true whether the adjective is one of Affect, so that the topic is also the Target, as in They were enthusiastic about the idea, or a non-Affect one, so that the subject of the clause is the Target and the prepositional phrase is a Topic (where the assumed action is thinking or speaking, as in The police were vague about the gunman’s demands) or a Specifier (as in Janet could not afford to be cavalier about money). Two constructions can be proposed: one expressed as ‘Person + BE + Affect + about entity/situation’, paraphrasable as ‘Person evaluates entity’; and one expressed as ‘Person + BE + Adjective + about entity/situation’, paraphrasable as ‘Person has/expresses an attitude/behaves towards entity, and I evaluate that attitude/behaviour’.

3. The pattern ADJ at n contributes to two constructions, again depending on whether the adjective expresses Affect. These are illustrated by: she felt guilty at having been spared and she was good at raising money.

Many patterns, however, are interpretable as a multiple set of constructions, depending on the adjective used with them. They also therefore occur in a range of analyses. The ADJ for n pattern is one example, as discussed above. Another is the pattern ADJ with n, which occurs in Analysis 2a (I was angry with them, where them is the Target), Analysis 2b (The tomato has remained popular with gardeners, where gardeners is the Evaluator), Analysis 4a (She was adept with her hands, where she … her hands arguably construes an action), and in several sections of Analysis 5: The first lady is busy with charity work (charity work is Specifier); Sales figures were comparable with those at previous exhibitions (those at previous exhibitions is Comparator); He was very patient with children (children is Affected); The valleys are ablaze with colour (colour is Cause). It also appears in the intensifying and quantifying patterns in Analyses 7a and 7b: Her voice was breathless with excitement; Its forests were abundant with wildlife.

Finally, in some cases, the configuration-pattern mapping, or construction, is consistent only if the pattern is further restricted. For example, as noted above, the patterns v it ADJ that and v it ADJ to-inf fit Analysis 3a only when the verb is of the ‘think’ type, as opposed to the ‘make’ type.

The local grammar we have proposed allows us also to ask whether the meaning distinctions proposed by other approaches to evaluative language are supported by this study. In particular, we can interrogate the Affect–Judgement–Appreciation model of Appraisal (cf Su and Hunston in press). The distinction between Analyses 2a and 5, which depends on the identification of the adjective concerned as ‘reaction’ or ‘opinion’ does support the unique position of Affect (see also Bednarek 2008). In most cases, however, neither the target-type nor the parameter of evaluation, both crucial to the Judgement–Appreciation distinction (Su 2015), are identified through pattern/construction alone.

There is potential for the identification of evaluative constructions with adjectives to contribute to resources for language teaching. An ambitious aim would be to contribute to a ‘constructicon’ (cf Fillmore et al. 2012) for learners, listing the combinations of lexis and grammar available in a given language to perform particular functions such as evaluation. For example, the examples shown here as Analysis 2a can be summarized for learners as a series of ‘slots’: ‘person + feels + emotion towards + thing’. The possibilities in each slot can be enumerated: be, feel, became, seemed, etc., in the ‘feels’ position; and the various adjective + preposition–clause combinations found in the ‘emotion towards’ position. Such a resource would combine elements of a dictionary, a pattern grammar, and a thesaurus.

Less ambitiously, the pattern grammar resources (Francis et al. 1996, 1998) can be used to derive teaching materials aimed at prompting learners to produce the various constructions identified. For example, the following prompts can be used to elicit examples of the ADJ about n pattern/‘reactive about’ construction:

1. ‘I described my idea’ + ‘John was enthusiastic’

2. ‘I wanted to meet some friends’ + ‘Ann was not keen’

3. ‘There was a terrible mess’ + ‘Robin was cheerful’

Learners would be asked to produce:

• 1a John was enthusiastic about my idea

• 2a Ann was not keen about meeting friends

• 3a. Robin was cheerful about the terrible mess

The levels of complexity involved in different constructions can also be exploited. For example, the prompt:

4 ‘the paintings were sold’ + ‘Jen was unhappy’

can be rephrased simply using the ADJ that pattern/‘reactive that-clause’ construction:

4a Jen was unhappy that the paintings were sold

or using the more complex nominalization (‘were sold’ → ‘sale’) necessitated by the preposition:

4b Jen was unhappy about the sale of the paintings.

Such activities promote awareness of the potential of adjective complementation and flexibility in using a variety of constructions. Other applications, such as using adjective complementation patterns in the automatic retrieval and parsing of evaluative meaning in naturally occurring text (Wiebe et al. 2005), remain an exciting but unexplored potential.

6. CONCLUSION

This article has argued that patterns, constructions, and local grammars are mutually supportive when deriving a comprehensive description of a set of linguistic resources such as those associated with evaluative meaning. These three approaches to language are all based on the analysis of naturally occurring language. They share a concern for patterning that supersedes a lexis/grammar divide. They all focus on alignments between form and meaning. The starting point for the article was language form and comprised the 40 adjective complementation patterns identified in Francis et al. (1998). A key proposal in the article is that the groups of adjectives listed for each pattern in that publication can be reinterpreted as constructions because they represent a matching of form and meaning. The number of constructions linked to each pattern ranges from 1 (it v-link ADJ that) or 2 (ADJ at n) to 6 (ADJ for n) or more (ADJ with n). The consequence is a very large, even unwieldy, number of constructions altogether. The identification of semantic elements within each construction, mapping meaning on to form, assists in distinguishing constructions and also contributes to the specification of a local grammar of evaluation. As a result, the large number of constructions can nonetheless be analysed using a relative small number of analyses (22, grouped into five main categories).

The language resources of explicit evaluation have been used as a test case for the reinterpretation of pattern grammar in terms of construction grammar and the contribution of both to the derivation of a local grammar. Because we can be confident that all adjective complementation patterns have been considered and analysed, we offer a comprehensive local grammar of the function of evaluation as expressed using such resources, joining other pragmatically driven local grammars (Su 2017; Su and Wei in press). The resources used to express evaluation, both explicitly and implicitly, are extensive, however (Martin and White 2005, Hunston 2011), and this local grammar can be only very partial. Perhaps its main contribution, as in the work by Su (2017, Su and Wei in press), is to specify the meaning elements involved in the evaluative act.

There is considerable scope for expanding this work. As noted above, the pattern grammar resources (Francis et al. 1996, 1998) include about 200 different patterns, complementing adjectives, nouns, and verbs. If each pattern can be interpreted as five constructions, which based on the work reported here seems a reasonable estimate, then 1,000 constructions of a similar level of specificity would have been identified. It remains the case that this identification is based on observation alone and does not address the question of whether such constructions are represented in the minds of language users. That question would be answerable by empirical work of a kind not undertaken here (but see Ellis et al. 2016). Other future research could include the quantification of lexis occurring in each of the proposed constructions, leading to the identification of collostructions and the measurement of collostructional strength (Stefanowitsch and Gries 2003; Gries and Stefanowitsch 2004).

This article has also discussed briefly the potential pedagogical applications of this local grammar approach. These have focused on the design of teaching materials that aim at developing a flexible language repertoire. In addition a thesaurus-like ‘constructicon’ has been proposed for use by language learners and teachers.

Footnotes

References