4.1 Introduction

One of the aims of the study described in this book was to undertake at least some annotation of semantic roles. Doing so requires a consideration of the available ways of doing this, and of the different purposes of the role annotation in different approaches. Similar debates are to be found in, for example, the valency grammar literature (Herbst and Götz-Votteler Reference Herbst and Götz-Votteler2007; Herbst and Uhrig Reference Herbst and Uhrig2019). The term ‘valency’ has a longer history than ‘pattern’ and is the term preferred by some of the approaches discussed in this chapter. Herbst (Reference Herbst, Herbst and Götz-Votteler2007: 20) distinguishes between syntactic and semantic valency. Syntactic valency is equivalent to ‘grammar pattern’, identifying the form of the arguments that accompany and categorise a verb or other lexical item (Matthews Reference Matthews, Herbst and Götz-Votteler2007: 3). Semantic valency relates to the meaning of those arguments. Götz-Votteler (Reference Götz-Votteler, Herbst and Götz-Votteler2007) usefully sets out some options:

• identify the semantic roles, which constitute ‘a more or less fixed set’ such as Agent or Patient;

• identify the semantic components of the items occupying those roles, such as animacy;

• subclassify the arguments into semantic categories, such as ‘person, animal, object, or force’;

• describe the arguments in verb-specific terms, specifying subgroups, such as ‘a group of people such as an army’ as the Agent of the verb ATTACK.

Herbst and Uhrig (Reference Herbst and Uhrig2019) describe a project to reinterpret valency patterns as constructions, and so to designate the valency patternbank of English into a constructicon of English. In this enterprise, the verbs and their arguments become elements or ‘slots’ in constructions.

In this chapter, four traditions of semantic element annotation are briefly described and compared. They are: Corpus Pattern Analysis – CPA (Hanks Reference Hanks2013), Systemic Functional Grammar – SFG (Halliday and Matthiessen Reference Halliday and Matthiessen2014), Local Grammar – LG (Barnbrook Reference Barnbrook2002), and FrameNet (Fillmore et al. Reference Fillmore, Johnson and Petruck2003). Following this survey, the approach to semantic annotation taken in this study is described.

When comparing approaches to semantic annotation, a number of parameters of difference may be identified (Hunston and Sui Reference Hunston and Sui2024). One such difference is between the semantic label as input to the identification of the unit and the semantic label as the output or consequence of the identification of the unit. For example, in CPA, category labels such as ‘Human’, ‘Entity’, or ‘Proposition’ are used to account for an example being identified as belonging to a given unit. That is, the labelling is part of the input to the unit. In FrameNet, on the other hand, the label indicates the role that the element plays in the Frame that the unit is part of. That is, the label is the output of the unit, explaining what the unit means. Example (1) illustrates this comparison. The CPA labels are ‘Human’ and ‘Entity = valuable’; the annotation might be glossed as ‘if you find the verb ACCEPT and noun phrases meaning “human being” and “a valuable entity”, then you have found an example of this unit’. FrameNet identifies this example as belonging to the Receiving frame and labels the Frame Elements ‘Recipient’ and ‘Theme’. The annotation might be glossed as ‘if an example belongs to this frame, then its elements may be interpreted as having these roles’.

1. (1)

   	She	accepted	a glass of wine. (BNC)
   CPA	Human	accepted	Entity=valuable
   FrameNet	Recipient	accepted	Theme

A second difference is between schemas which allocate examples to specific contexts or meaning areas, and those which take a more general view of meaning. Local Grammar, for example, selects instances with specific rhetorical functions and allocates element labels commensurate with those functions. Systemic Functional Grammar, on the other hand, applies more general labels. Example (2) illustrates this comparison. The LG labels draw on the specific rhetorical function of the example, Apologiser and Apologisee (drawing on Su and Wei Reference Su and Wei2018), while the SFG labels simply treat the example as an instance of a verbal process (drawing on Halliday and Matthiessen Reference Halliday and Matthiessen2014).

1. (2)

   I	want to	apologise	to the rest of you.
   LG	Apologiser	Apologising	Apologisee
   SFG	Sayer	Process	Receiver

The reason for the different labelling systems is that each approach has a distinct purpose. Corpus Pattern Analysis aims to define precisely the various meanings a word can have, in terms that are similar to valency grammar (Herbst et al. Reference Herbst, Heath, Roe and Götz2004). In practical terms, the patterns can be used in automatic sense disambiguation: if an instance of the verb ACCEPT is found preceded by a noun phrase belonging to the class ‘Human’ and followed by a noun phrase belonging to the class ‘Entity of a valuable kind’ (see example (1)), then the verb belongs to this specific sense and no other. Systemic Functional Grammar seeks to establish the range of resources available to the speaker of a language (see Chapter 5). It distinguishes process types, such as ‘verbal process’, and establishes potential participant types as a way of establishing how that process type operates (see example (2)). FrameNet identifies the role that each element of a target utterance has within a situation or Frame; for example, the subject of the verb ACCEPT in example (1) is the Receiver in a situation in which one person gives something to another person. FrameNet specifies how Frames can be expressed. That is, it does not ask ‘what meanings can this word make?’, as CPA does, but asks ‘what are the different ways that this meaning can be made?’ Local Grammar does something similar, but focuses on rhetorical functions, such as apology (see example (2)), rather than situational Frames. This difference in purpose makes it difficult for one schema to exactly replicate the terminology from another schema.

4.2 Corpus Pattern Analysis

The CPA project led by Hanks (Hanks Reference Hanks2013) led to the compilation of the online Pattern Dictionary of English Verbs (PDEV) (accessed at pdev.org.uk). The purpose of the project was to distinguish word meanings by identifying ‘patterns’ consisting of structural elements and semantic sets. In Hanks’s words: ‘A pattern consists of a valency structure, together with sets of preferred collocations’ (Hanks Reference Hanks2013: 92). Each pattern is associated with ‘a particular meaning or implicature’ (Hanks Reference Hanks2013: 92). A distinction is made between prototypical uses, or ‘norms’, and unusual or creative uses, or exploitations. Hanks distinguishes his approach from that of Hunston and Francis (Reference Hunston and Francis2000), noting that ‘their framework does not assign a central role to lexical collocations in determining what counts as a pattern’ (Hanks Reference Hanks2013: 6). He points out, correctly, that this means that patterns, in the Francis et al. (Reference Francis, Hunston and Manning1996, Reference Francis, Hunston and Manning1998) sense, do not distinguish between word meanings unless two meanings have different syntactic patterns. He gives the example of fire a gun and fire an employee, which in Pattern Grammar terms are both coded V n, but which in PDEV are accounted for by two separate patterns (Hanks Reference Hanks2013: 6):

• pattern 1 ([Human] fire [Firearm]) and

• pattern 5 ([Human1 = Employer] fire [Human2 = Employee])

(The representation of the PDEV patterns is somewhat simplified here. Single square brackets are used instead of double ones, and some alternatives, such as ‘either a person or an institution can fire an employee’ are omitted.)

Central to Hanks’ work is the notion of lexical sets. In the fire a gun/employee examples, for instances, the lexical sets ‘Human’ and ‘Firearm’ are used. This means that identification of the pattern is dependent on (a) the identification of the valency pattern V n and (b) the recognition that the Subject and Object nouns belong to the class Human or the class Firearm. Discussing how lexical sets are arrived at, Hanks uses the example of the verb URGE (Hanks Reference Hanks2013: 117). One meaning of this verb occurs in examples such as (3).

1. (3) We urged our horses forward. (BNC)

Hanks notes that the most frequent noun in the object slot is horse, but that common sense dictates that other animals used as transport, such as camels, might also be urged, even if there is little corpus evidence to that effect. On the other hand, inanimate entities which human beings control, such as vehicles, cannot truly be urged, though examples can be found of somewhat whimsical uses such as example (4).

1. (4) He urged the ship onwards. (enTenTen21)

In Hanks’ terms, example (4) would be an instance of exploitation, that is, a metaphoric extension of the central usage of URGE. There is, then, a fuzzy lexical set that contributes to this meaning of URGE and that might be given the name [Steed]. Hanks refers to the criterion that the urged entity should possess consciousness as ‘intensional’, distinguished from an ‘extensional’ criterion that simply lists all occurring collocates (Hanks Reference Hanks2013: 118).

In PDEV, the meaning of each pattern is explained by means of an implicature. This explains rather than defines the pattern. To illustrate this, Table 4.1 shows 7 selected patterns from the 21 patterns listed in the PDEV entry for the verb TELL. They have been chosen to illustrate a range of valency patterns and lexical sets. Because the implicatures are sometimes a little difficult to interpret, being expressed formulaically, a gloss has been added in some cases.

Table 4.1Selected patterns of TELL from PDEV

Table 4.1:Long description

Pattern 1. Four elements: 1. human or institution or information source 2. tell 2. human 2 or institution 2. quote or that-clause or wh-clause. Implicature: human or institution or information source informs human or institution quote or that-clause or wh-clause. Gloss: Person tells person that something is the case. Or for example, Newspaper tells people what events took place. Pattern 5. Four elements: 1. human or institution or information source 2. tell 2. human or institution 4. of anything equivalent to a topic. Implicature: human or institution or information source informs human or institution about some facts concerning any topic. Gloss: Person tells person of any topic. Pattern 6. Four elements: 1. human or institution 2. tell 2. human or institution 4. to-infinitive. Implicature: human or institution orders, instructs, or advises human or institution to-infinitive. Gloss: Person tells person to do something. Or for example, Government tells Civil Service to do something. Pattern 8. Three elements: 1. inanimate thing or event 2. tell 2. story. Implicature: inanimate thing or event is a reason for believing a story. Gloss: for example, the evidence told the story of the crime. Or for example, His departure told its own story. Pattern 12. Three elements: 1. human 2. tell 2. joke. Implicature: human says something that is intended to be funny. Pattern 14. Five elements: 1. human 2. tell 2. anything 4. from 5. anything. Implicature: Human is able to decide the difference between entity 1 and entity 2. Pattern 21. Four elements: 1. human schoolchild 2. tell 2. tale 4. on or against human schoolchild. Implicature: Schoolchild 1 reports to teacher or authority some action done by schoolchild 2 that is against a rule, and by doing so breaches a code of honour among schoolchildren. Gloss: One child tells tales on another child.

The valency or syntactic patterns shown in Table 4.1 are as would be expected. We see the grammar patterns V n (pattern 8), V n from n (pattern 14), V n of n (pattern 5), V n that (pattern 1), and V n to-inf (pattern 6). What is more interesting are the lexical sets specified for each pattern. In pattern 1 the Subject may be either a human being, or an institution, or an information source, and the Object may be a human being or an institution. In patterns 12, 14, and 21 the Subject may only be a human being. In pattern 8 the Subject may be an inanimate entity (such as evidence) or an event (such as his departure). The Object is the word story, with obligatory modifiers. Most interestingly, the implicature given for pattern 21 (‘one child tells tales on another’) specifies elements that are not present in the pattern itself: a teacher or other authority figure to whom the information is given; an action that the second child has done; and a rule that is broken by that action. It also includes the information that to ‘tell tales on someone’ is regarded as a wrong thing to do by other children.

There is a strong case to be made for regarding the patterns recorded in PDEV as compatible with the notion of constructions, in the sense that each word pattern is uniquely associated with a meaning or implicature. Pattern 21 in Table 4.1 is a particularly good example of a construction, as its meaning is non-compositional, that is, not predictable from the form alone. The other patterns are more predictable and less striking, but can also be interpreted as constructions without difficulty.

Hanks’ notion of exploitations can also usefully be applied to constructions, even though Hanks takes the word as central and asks ‘what unusual uses of this word occur?’, whereas researchers such as Goldberg start from the construction and ask ‘what unusual words occur in this construction?’. Both might conclude, however, that examples such as ‘he urged the ship onwards’ or ‘she smiled herself an upgrade’ are examples of exploitation of a linguistic norm.

However, CPA presents a challenge to the description of constructions described in this book, raising the question of whether constructions should be annotated in terms of the semantics of their contributing elements. Because the preference here is to annotate the ‘output’ function rather than the ‘input’ category, labels of this kind have not been used. However, some information about lexical sets has been included as part of construction descriptions. Here are some examples:

• Pattern: V n -ing. Cx 10: the ‘spend time doing something’ construction. Description: A person spends time doing something. The NP following the verb indicates the period of time. Example: We spent the day exploring. (BNC)

• Pattern: V on n. Cx 23: the ‘build on’ construction. Description: A person or entity uses another person or entity as a basis for increase or improvement. Example: The proposed draft protocol builds on existing pollution controls. (BNC)

• Pattern: V n about n. Cx 6: the ‘know a lot about something’ construction. Description: A person knows or learns about something. The NP following the verb indicates an amount. Example: She knew a lot about birds. (BNC).

We now turn to the schemas that code the output of patterns, starting with SFG (Halliday and Matthiessen Reference Halliday and Matthiessen2014).

4.3 Participant Roles in Systemic Functional Grammar

It will be explained in Chapter 5 that SFG separates three roles that the clause plays: it constitutes an exchange or interaction between speaker and hearer (the interpersonal metafunction); it is the carrier of a message, comprising ordered pieces of information (the textual metafunction); and it construes a situation or set of events (the experiential metafunction). Within each metafunction, the speaker has a range of alternate resources available to them, such as interacting with interrogative or declarative mood, or beginning the clause in a marked or unmarked way. Of most relevance here is the experiential metafunction, which interprets clauses in terms of how they represent the social and physical world. The main distinctions made are between process types, and part of the way that process types are recognised is by the participant roles they occur with. For example, a clause built around a mental process verb phrase will have a Senser and a Phenomenon, whereas a clause with a verbal process verb phrase will have a Sayer and a Message. The main process types proposed by Halliday (Reference Halliday and Lyons1970) are material, mental, and relational. These reflect the three major ways of construing the world: as a set of events and happenings (material processes); as a set of perceptions and thoughts (mental processes); and as a representation of connections between things, including connections between reified events and happenings (relational processes). The other process types are verbal, behavioural, and existential.

The identification of process types and their associated participant roles constitutes a large part of what make this kind of grammar ‘functional’. Halliday and Matthiessen explain that traditional labels such as ‘Subject’ and ‘Object’ in fact conflate different kinds of information (Halliday and Matthiessen Reference Halliday and Matthiessen2014: 79–82). They go into considerable detail in examining the varying relationships that can exist between the verb element in a clause and the Subject and Object elements. Examples taken from Halliday and Matthiessen (Reference Matthiessen2014: 226–307), are given here:

• The lion caught the tourist: Actor-Process-Goal (p226)

• They played a game of tennis: Actor-Process-Scope (p238)

• I gave my love a ring: Actor-Process-Recipient-Goal (p239)

• They washed the plates clean: Actor-Process-Goal-Attribute (p238)

• Mary liked the gift: Senser-Process-Phenomenon (p248)

• The gift pleased Mary: Phenomenon-Process-Senser (p248)

• Mice are timid creatures: Carrier-Process-Attribute (p267)

• Mr Garrick played Hamlet: Identified-Process-Identifier (p277)

• He told me the truth: Sayer-Process-Receiver-Verbiage (p306)

• He praised my teaching ability: Sayer-Process-Target (p307)

One aim of SFG is to account for all combinations of Process and Participant in a language. Of necessity, this means that the labels used are somewhat general and relate to overall distinctions of use rather than to context-specific distinctions. One response to this is to propose sets of labels that are functional in a different way (Hunston and Sinclair Reference Hunston, Sinclair, Hunston and Thompson1999: 79), linking the labelling of elements of clauses to the rhetorical function of the clause. The term ‘Local Grammar’, taken from Gross (Reference Gross and Hoey1993), is used to describe this approach, described in the next section.

4.4 Local Grammar

The difference between the participant roles used in SFG and the element labels in LG might be illustrated with an invented example based on one used by Sinclair (personal communication): ‘A dog is a beloved companion’. The indeterminacy of the meaning is captured in the possibility of two codings in SFG: this is either Carrier-Process-Attribute, meaning that a dog has the characteristics of being lovable and companiable, or it is Identified-Process-Identifier, meaning that the identifying description of a dog is as a loved companion. Sinclair essentially rephrases this distinction in terms of the function of the clause: it is either an evaluation of a dog or a definition of a dog. In one, the constituent elements might be coded ‘Target + Evaluation’; in the other, it might be ‘Definiendum + Definiens’. For a reader to know which coding is correct, they have to use common sense to decide whether ‘a beloved companion’ evaluates or defines ‘a dog’.

When a LG is devised, therefore, it proposes semantic role labels that are specific to a given rhetorical function. The labels are relevant only to clauses that perform that function. Local Grammars are useful in determining what roles are relevant to a given function, how those roles are expressed and what options are available for doing so, and in making it possible to quantify the occurrence of the various options in a corpus. Although each LG study is comprehensive in scope, inevitably only a handful of language functions have received attention of this kind, so that Gross’s (Reference Gross and Hoey1993) initial ambition of extending the description to the whole of a language is currently not within reach.

Barnbrook’s work on the LG of definitions (Barnbrook Reference Barnbrook2002, see also Barnbrook and Sinclair Reference Barnbrook, Sinclair, Sinclair, Hoelter and Peters1995, Reference Barnbrook, Sinclair, Ghadessy, Henry and Roseberry2001) is the most extensive and comprehensive study of this kind. The aim was to identify and quantify the different formats of definition used in the Collins Cobuild Student’s Dictionary (Sinclair Reference Sinclair1990). Barnbrook (Reference Barnbrook2002: 135–136) identifies 17 types of definition, divided into four groups. The types range in frequency (in the CCSD) from over 10,000 instances to just 14. A finite set of terms is used to label the elements, such as ‘Definiendum’ (that which is defined), ‘Definiens’ (the defining element), ‘Superordinate’ (part of the defining element, that assigns the defined thing to a class), ‘Discriminator’ (that which separates the defined thing from others in the class), and ‘Hinge’ (which joins parts of the definition to each other).

Barnbrook’s general approach has been adapted by other researchers and applied to other functions: evaluation (Hunston and Sinclair Reference Hunston, Sinclair, Hunston and Thompson1999; Hunston and Su Reference Hunston and Su2019); disclaiming (Cheng and Ching Reference Cheng and Ching2018); apologising (Su and Wei Reference Su and Wei2018; Su Reference Su2020); thanking (Su Reference Su2018); data commentary (Zhang et al. Reference Zhang, Jiang and Zhang2024); defining (Zhang and Su Reference Zhang and Su2023); and exemplifying (Su et al. Reference Su, Zhang and Chau2022). Much of this work is relatively small scale, but a paper by Yu et al. (Reference Yu, Li, Su and Fuoli2024) applies a larger-scale approach using Large Language Models. Their paper illustrates the difficulty in quantifying types of rhetorical patterns. In Barnbrook’s case, the definitions were all located in a dictionary and are identified as definitions in the dictionary database; in the case of naturally occurring apologies, the apologies have to be found first, and only then can they be parsed. Inevitably it is only the more explicit apologies that are identified and analysed.

Table 4.2 gives some examples of LG coding.

Table 4.2Examples of Local Grammars

Table 4.2:Long description

Definition example: When a student graduates, he or she has successfully completed a degree course… This is divided into five and each element is annotated: When equals Hinge; a student equals Co-text; graduates equals Definiendum; he or she equals Match; has successfully completed a degree course equals Definiens. Data Commentary example: Our correlations in table 7 show that better governed countries have… This is divided into five and each element is annotated: Our correlations equals Data; in equals Hinge; table 7 equals Graphic; show equals Act; that better governed countries… equals Interpretation. Exemplification example: This negotiation process actually involves complex mechanics. Schiffirin (1987: 24-28), for example, proposes a five-plane model which involves… This is divided into four and each element is annotated: This negotiation process actually involves complex mechanics equals Exemplified; Schiffrin (1987: 24-28) equals Exemplification - supporting research; for example, equals Indicator; proposes a five-plane model which involves… equals supporting statement.

4.5 FrameNet

FrameNet is one of the most detailed and influential approaches to the description of form and meaning in language. A clear account of its principles, outputs, and working methods can be found in publications such as Fillmore et al. (Reference Fillmore, Johnson and Petruck2003) and Fillmore (Reference Fillmore, Herbst, Schmid and Faulhaber2014), and the data and other resources can be found at framenet.icsi.berkeley.edu. FrameNet is based on Frame Semantics (Fillmore Reference Fillmore1976), which proposes that the meaning of a word is dependent on the real-world situation or concept in which that word occurs:

There are a number of important concepts in FrameNet:

The frame. There are hundreds of frames listed on the FrameNet website (framenet.icsi.berkeley.edu/frameIndex). Some indicate objects (e.g. Containers, Clothing). Some relate to recognisable events (e.g. Commercial_transaction, Cooking_creation). Some are more abstract (e.g. Causation, Conduct). While many relate to physical actions, many relate to speech (e.g. Citing, Chatting) or to emotions or attitude (e.g. Candidness, Coming_to_believe). Some frames are clearly more specific versions of other frames. For example, there are 34 listed frames that specify a type of cause, such as Cause_benefit_or_detriment, Cause_to continue, and Cause_to_land, or Cause_to_perceive. Relations between frames are formalised using a variety of concepts, of which the most transparent are ‘inherits from’, ‘is inherited by’, and ‘uses’ or ‘is used by’. For example, the Coming_to_believe Frame inherits aspects of the Event Frame and is inherited by the See_through Frame. It uses the Mental_activity Frame and is used by the Falling_for and the Turning_out Frames.

The lexical unit. Every word analysed in the FrameNet project is said to evoke one or more frames, on which its meaning depends. A word and its associated sense comprise a lexical unit (LU). For example, the verb buy belongs to two LUs, one evoking the Commerce_buy Frame (which inherits from the Getting Frame), and one evoking the Fall_for Frame (which uses the Coming_to_believe Frame). An example of the first would be ‘I bought a new coat’, while an example of the second would be ‘I don’t buy that idea’. The FrameNet website offers information about each LU. This comprises the valency patterns that the LU is found in, and examples that are coded for Frame Elements. For example, the LU buy, in the Commerce_buy Frame, is shown to occur in about 30 valency patterns (an approximate number because as of March 2024, the entry is listed as not yet complete). Using the pattern grammar conventions, these include, for example: V n at n (‘buy an item at a price’); V n from n (‘buy an item from someone’); V n (‘buy something’); V n for n (‘buy something for someone’; ‘buy something for an amount of money’).

Frame Element. A crucial aspect of the FrameNet project, and the one most relevant to this chapter, is the identification of Frame Elements (FEs). Frame Elements are the semantic roles identified for each frame, and are divided into Core and Peripheral Elements. All the examples listed under each LU are coded for the Elements they exemplify. The Commercial_transaction Frame has four core Elements: BUYER, GOODS, MONEY, and SELLER. It has three peripheral Elements: MEANS, RATE, and UNIT. In instances where Core Elements do not occur, they are implied. For example, the Revenge Frame has five core Elements: AVENGER, INJURED_PARTY, INJURY, OFFENDER, and PUNISHMENT. This means that every description of an act of revenge evokes a scenario with these elements. However, many examples of this frame leave implicit one or more of these Elements. For example, I’ll get even with you for this (from the FrameNet website entry for GET EVEN) includes the Elements AVENGER (I), OFFENDER (you), and INJURY (this), but the INJURED_PARTY and the PUNISHMENT are left implicit.

Both Construction Grammar and FrameNet were pioneered by Fillmore, and have shared origins (Fillmore et al. Reference Fillmore, Lee-Goldman, Rhomieux, Sag and Boas2012). Fillmore (Reference Fillmore, Herbst, Schmid and Faulhaber2014) explicitly explores the connections between them. As shown in Chapter 2, others have sought to use the notion of frames in building a network of constructions: a constructicon. Perek (Reference Perek, Ziem, Willich and Michel2025) comments that ‘frame semantics is often considered the semantic component of construction grammar’, adding that both see a ‘continuity between lexical and constructional meaning’. Taking this argument further, Perek finds it logical to ‘represent constructional meaning using the same theoretical apparatus as lexical meaning i.e. semantic frames’. Perek and Patten (Reference Perek and Patten2019) and Perek (Reference Perek, Ziem, Willich and Michel2025) describe a Constructicon of English which draws on the Pattern Grammar resources (Francis et al. Reference Francis, Hunston and Manning1996) to derive constructions that are presented and annotated with FEs taken from FrameNet. Perek and Patten (Reference Perek and Patten2019) give an example of the ‘Communication “V that” construction’, which uses the form of the V that pattern and has FEs belonging to the Communication Frame. A further 14 more specific frames are identified that directly inherit from or use the Communication Frame and which occur with verbs identified as occurring with V that in Francis et al. (Reference Francis, Hunston and Manning1996). Examples are: Communication_response; Request; Reasoning; and Reporting. One such frame, Statement, itself has a further seven frames that directly inherit or use it, including Telling, Reveal_secret, and Predicting.

Consideration was given in the study described in this book to following a similar procedure and to apply FE annotations to the identified constructions. This proved impossible to do consistently, mainly because the constructions had been identified without reference to the FrameNet framework. As a consequence, in many cases, a construction as proposed in this project might evoke several frames. For example, one of the constructions proposed for the pattern V of n is the ‘complain of’ construction, exemplified by … both sides complained of ceasefire violations (BNC). The verbs listed as occurring in this construction are: BOAST, COMPLAIN, SPEAK, TALK, TELL, and WARN. In FrameNet, each of these is regarded as evoking a different frame. This makes sense, as ‘boasting of something’ and ‘complaining of something’, for example, clearly relate to different real-world situations. Table 4.3 shows six verbs that occur with the V of n pattern and the frames that these LUs evoke. In each case the table shows and example and lists the FEs appearing in the example.

Table 4.3Frames evoked by verbs in the ‘complain of’ construction (from framenet.icsi.berkeley.edu)

Table 4.3:Long description

Verb 1: boast. Frame: Bragging. Example: She has boasted of her tendency…. Frame elements: Speaker (she); Topic (her tendency). Verb 2: complain. Frame: Complaining. Example: Schools have complained of interference. Frame elements: Complainer (schools); Complaint (interference). Verb 3: speak. Frame: Chatting. Example: Let us speak of the present. Frame elements: Interlocutors (us); Topic (the present). Verb 4: talk. Frame: Statement. Example: He talks often of events at home. Frame elements: Speaker (he); Topic (events at home). Verb 5: tell. Frame: Telling. Example: One woman tells of being forced to give up her baby. Frame elements: Speaker (one woman); Message (being forced to give up her baby). Verb 6: warn. Frame: Warning. Example: Activists warn of political repercussions. Frame elements: Speaker (activists); Topic (political repercussions).

A conclusion that might be drawn from Table 4.3 is that the Elements SPEAKER, MESSAGE, and TOPIC might be found in several frames, with a note that the specific role of the speaker (‘BOASTER’, ‘COMPLAINER’, ‘WARN-ER’, etc.) and the nature of the message (‘BOAST’, ‘COMPLAINT’, and ‘WARNING’) depends on the frame. Further attempts to use FrameNet as the source for semantic labelling of construction elements made it apparent that applying this consistently would indeed require taking frames as the starting point for the study, as Perek and Patten (Reference Perek and Patten2019) do. Frame Elements cannot reliably be mapped on to the constructions that have been identified in this project. However, it can be argued that the current constructions are available for revision by other researchers who may wish to increase the compatibility of patterns, constructions, and frames.

4.6 Semantic Fields and Participant Roles in This Study

It is clear from these descriptions of approaches to form and meaning that the annotation of unit elements is an inspiring, if ambitious, goal. Several annotation schemes are available, in most cases linked to specific reference resources, such as FrameNet or the PDEV. The schemes are dependent on the aim of the project and also on its starting point. The PDEV, for example, starts from the analyses of thousands of verbs in English and has built up an ontology of semantic roles that are necessary to annotate the distinctive elements of each pattern. FrameNet, by contrast, starts with the concept of the Frame and allocates labels that are specific to each Frame. The current study starts with neither the verb nor the Frame but with the pattern. The constructions that are derived from the patterns are not single-verb specific and it would therefore be unfeasible to apply Hanks’ (Reference Hanks2013) ontology reliably. Similarly, the constructions do not necessarily map uniquely on to Frames, so applying the FrameNet semantic labels is similarly unfeasible.

The project has adopted from FrameNet, however, the principle that semantic labels have to relate to a semantic field, so that a comprehensive labelling would entail having an extensive and finite set of such fields. That is also outside the scope of this study, but as a trial, nine semantic fields have been selected for annotation. The fields are inspired by, though not entirely based on, SFG and the distinction between process types. This enabled the identification of fairly broad semantic areas. A finite set of participant or argument roles were assigned to the elements in the constructions related to those fields. Because there are only nine fields, by no means all the constructions identified in the study are coded in this way. The semantic fields chosen can be divided into the following SFG process types: material, relational, mental, and verbal. The list of semantic fields is:

• Related to material processes

  • ◦ Causation

  • ◦ Change

  • ◦ Creation

  • ◦ Location change

  • ◦ Possession transfer

• Related to relational processes

  • ◦ Equivalence

  • ◦ Logical relations

• Related to mental and verbal processes

  • ◦ Cognition

  • ◦ Communication

The relationship between semantic fields and constructions is the basis of Chapters 6–8. In this chapter, the participant roles assigned to each field are listed and briefly explained. The terminology used for the participant roles is again in part inspired by that used in SFG but there is no strict adherence to that schema and the labels are atheoretical. In particular, in SFG (e.g. Halliday and Matthiessen Reference Halliday and Matthiessen2014), fine distinctions of meaning are made between related participant roles, but the role labels used here are less finely distinguished. The principle is to have as few roles as is feasible, to avoid errors in coding. Many of the participant role labels are used across a large number of constructions, such as ‘Sayer’ or ‘Cognizer’. Some are used in more than one semantic field, such as ‘Cause’ or ‘Source’. A few are more restricted in their use, however. One example is the role ‘Time’, which is used when an element in a construction expresses a time period. It is used in the Cognition and Equivalence fields, in examples such as (5) and (6).

1. (5) The following day saw a remarkable change in the weather … (BNC)

2. (6) Dangerfield had started the day tired … (BNC)

In example (5) the following day is construed metaphorically as something that perceives. In example (6), the day is an additional element in the equivalence clause ‘Dangerfield was tired’.

One point that needs to be stressed is that the coding procedure allows for multiple labelling. That is, it is considered that a single construction, and a single example, can exemplify more than one semantic field. In other words, the semantic fields are not mutually exclusive. For example, the ‘call someone something’ construction, from the pattern V n n, is assigned to both the Communication field and the Equivalence field. In an example such as She called him a coward, there is a representation of an action of communication (‘she communicated a message’), with the participant roles ‘Sayer’ and ‘Message’, and also a statement of equivalence (‘he is a coward’), with the participant roles Carrier and Attribute.

A further point to be made is that, unlike in the FrameNet model, only the verb arguments as identified by the construction are analysed. Other parts of the clause are not coded, even though they may be an integral part of the situation or Frame. This is illustrated in example (7).

1. (7) He pushed it [the door] open with his foot. (BNC)

The relevant FrameNet entry (for the LU push evoking the ‘Cause_motion’ frame) suggests that the FEs are: Agent (he); Theme (the door); Instrument (his foot). However, the pattern identified here is V n adj (not ‘V n adj with n’) and the construction specifies ‘A person or entity moves something so that it is open or closed’. Because what is analysed here is the construction rather than the frame, the prepositional phrase with his foot is not part of the pattern or construction and is therefore not allocated a participant role.

The allocation of participant roles in relation to each set of semantic fields will now be discussed.

Fields Related to Material Processes

Table 4.4 summarises the Participant Roles in the fields associated with material processes. The fields are: Causation, Change, Creation, Location Change, and Possession Transfer. Although no distinction is made between ‘core’ and ‘peripheral’ roles, intuitively one might say that some roles are more basic to the processes than others. In the case of Change, Creation, and Location Change there are two basic or core roles:

• Change: Agent + Affected e.g. He lowered the volume;

• Creation: Creator + Outcome e.g. They published a book;

• Location Change: Cause-move + Mover e.g. She shifted the furniture.

Table 4.4Participant roles in the fields associated with material processes

Table 4.4:Long description

The Causation field has three roles: Cause, Affected, and Result. The cause is the person or entity that makes something happen. The affected is a person or entity that is affected by a cause. The result is the outcome of a caused event. In the example The party elected a businessman as its leader, the party is Cause, a businessman is Affected and its leader is Result. In the example The news surprised her, her is affected. In the example The play left a lasting impression, a lasting impression is Result. The Change field has five roles: Affected, Outcome, Agent, Parameter, and Means. The affected is the entity that undergoes change. The outcome is the consequence of the change. The agent is the entity that makes the change happen. The parameter is the parameter along which the change occurs. The means is the entity that is added to something to make a change. In the example Losing her hair has affected her deeply, her is Affected, losing her hair is Agent. In the example The iron deposits changed Corby into a red brick town, Corby is Affected, a red brick town is Outcome and the iron deposits is Agent. In the example He lowered the volume by 5 decibels, 5 decibels is the Parameter. In the example The rice increased in volume, volume is the Parameter. In the example He sweetened his coffee with sugar, sugar is the Means. The Creation field has five roles: Creator, Outcome, Beneficiary, Location, and Constituent. The creator is the person who creates something. The outcome is the entity that is created. The beneficiary is a person for whom something is created. The location is the place where something is created. The location is changed by the creation. The constituent is the material from which something is created. In the example They published a book, they is Creator and a book is Outcome. In the example He built her a castle, her is Beneficiary. In the example She carved the words in the tree bark, the tree bark is Location. In the example They make the Gollum out of clay, clay is the Constituent. The Location change field has four roles: Mover-physical, Mover-abstract, Location, and Cause-move. The mover-physical is a physical entity that moves from one place to another. The Mover-abstract is an abstract entity in a construction that construes abstract change as movement from one place to another. The location is the place where someone or something moves to or from. The cause-move is the person or entity that makes something or someone move. In the example The servants brought ladders, ladders is Mover-physical and the servants is Cause-move. In the example He was gathering information about the incident, information about the incident is Mover-abstract. In the example introducing environmental safeguards into its structural adjustment, environmental safeguards is Mover-abstract. In the example Her mother had dragged her into her room, her room is Location. The Possession transfer field has three roles: Donor, Recipient and Possession. The donor is the person who originally has the item. The donor may voluntarily give the item or may have it taken from them. The recipient is the person to whom possession is transferred. The person may be given the item or may take it. The possession is the item, physical or abstract, that is transferred. In the example ‘They fed the troops, they is Donor and the troops is Recipient. In the example He robbed the family next door, the family next door’ is Donor and he is Recipient. In the example John borrowed from Jim, Jim is Donor and John is Recipient. In the example They provided money, money is Possession. In the example He took the television, the television is Possession. In the example John deprived Jim of his birth right, his birth right is Possession.

In the case of Causation and Possession Transfer there are, arguably, three basic roles:

• Causation: Cause + Affected + Result e.g. The party elected a businessman as its leader;

• Possession Transfer: Donor + Possession + Recipient e.g. John borrowed money from Jim.

Other roles are identified where they occur as parts of a construction, which in turn means they occur as elements in a pattern. An obvious example is the Beneficiary role that occurs when the pattern V n n expresses the Creation field (e.g. He built her a castle). Less obviously perhaps, the pattern V n in n contributes to a construction in the Change field and the noun phrase following in is coded as Parameter (e.g. The rice increased in volume). The same pattern contributes to a construction in the Creation field, where the noun phrase in the same position is coded as Location (e.g. She carved the words in the tree bark).

One of the features of the Location Change semantic field is that abstract entities may be construed as moving, as well as physical ones. For example, the verb INTRODUCE, used in the pattern V n into n, may be used to construe a physical movement, as in example (8), or an abstract one, as in example (9).

1. (8) [They] do not recommend introducing pigs into existing woodland … (BNC)

2. (9) … to introduce adversarial elements into the criminal process. (BNC)

In example (8), pigs is coded ‘Mover-physical’, meaning ‘a physical entity that changes location’. Example (9) is considered to express metaphorical movement and adversarial elements is coded ‘Mover-abstract’. Both existing woodland and the criminal process are coded Location, although these could be distinguished as ‘physical’ and ‘abstract’ also.

Fields Related to Relational Processes

In SFG, relational processes express connections between two entities or actions, most usually processes of ‘being’ or of ‘having’. In this project, the concept of ‘relation’ is interpreted more widely. Two separate fields are proposed, though one of them is in fact a portmanteau for a range of relation types. Table 4.5 shows the two fields and the participant roles associated with them. The first field is Equivalence. The prototype for this field might be example (10).

1. (10) Until his death in 1768, the Duke was the most powerful man in England. (BNC)

Table 4.5Participant roles in the fields associated with relational processes

Table 4.5:Long description

The Equivalence field has eight roles: Carrier, Attribute, Cognizer, Sayer, Causer, Actor, Time and Source. The carrier is something that is assigned a quality, identity or role. Numbers are used to show the sequence of occupiers of a role or two Carriers that are connected. In the example John appears to be the wealthiest man in the world, John is Carrier. In the example John likes his coffee hot, his coffee is Carrier. In the example Moynihan had replaced Spencer as Professor of Painting, Moynihan is Carrier 2 and Spencer is Carrier 1. In the example the structure of your own language differs from that of the new language, the structure of your own language is Carrier 1 and the new language is Carrier 2. The attribute is the quality, form, identity or role assigned to the Carrier. This includes attributions of quantity, purpose and compositions, and denotes items added to the Carrier. In the example John is the wealthiest man in the world the wealthiest man in the world is Attribute. In the example John professes himself amazed, amazed is Attribute. In the example The dollar peaked at 0.86 pounds, 0.68 pounds is Attribute. In the example …plunging the passageway into semi-darkness, semi-darkness is Attribute. In the example His weight had compressed the carbon atoms into diamond crystals, diamond crystals is Attribute. In the example The party elected a businessman as its leader, its leader is Attribute. In the example …to designate an area for mistakes, mistakes is Attribute. In the example I decorated my bicycle with flags, flags is Attribute. The cognizer is a person who believes or wishes that a Carrier has an Attribute. In the example John likes his coffee hot, John is Cognizer. The sayer is a person who communicates that a Carrier has an Attribute. In the example John professes himself amazed, John is Sayer. The causer is a person or entity that causes a Carrier to possess an Attribute. In the example John pushed the door open, John is the Causer. The actor is a person or entity that acts upon a Carrier with an Attribute but does not cause it. In the example The attacker captured him alive, the attacher is Actor. The time is a time period or event during or at which a Carrier has an Attribute. In the example Stocks started the day higher, the day is Time. The source is an entity that provides evidence for a Carrier having an Attribute. In the example His designs revealed him as highly competent, his designs is Source. The Logical relations field has ten roles. The first two are Influencer and Influenced. The influencer is an entity, action, event or situation which influences another entity, action, event or situation. The influence may be cause, enablement or a lesser degree of influence. The influenced is an entity, action, event or situation which is influenced by an influencer. In the example Quinn’s background dictated that he could take care of himself, Quinn’s background is Influencer and that he could take care of himself is Influenced. In the example The durability of the coins has ensured the survival of these images, the durability of the coins is Influencer and the survival of these images is Influenced. The next two roles are Conditional and Conditioner. The conditional is an action or event whose occurrence is conditional upon another. The condition may be causative or only a necessary condition. The conditioner is an action or event which is a condition for a Conditional. In the example The opening of the service necessitated Blackpool buying trams…, the opening of the service is Conditional and Blackpool buying trams is Conditioner. The next two roles are Sequential and Sequencer. The sequential is an action or event which occurs in a time sequence with another. The sequencer is a person who does two or more things in a sequence. In the example The presentation was followed by a reception, the presentation is Sequential 1 and a reception is Sequential 2. In the examples Martin prefaces his lesson with a statement that…. his lesson is Sequential 1, a statement that is Sequential 2 and Martin is Sequencer. The next two roles are Correspondence and Correspondence-Cause. The correspondence is an action or entity which is connected to another in a relation of similarity, contrast, or necessary co-existence. The correspondence-cause is a person who construes a correspondence between two entities. In the example The arrows correspond to the decision path, the arrows and the decision path are Correspondence. In the example They based their ideas on ancient writing, their ideas and ancient writings are Correspondence. In the example They couched their teaching in story form, they is Correspondence-cause. The next two roles are Evidencer and Evidenced. The evidence is an entity which provides evidence for a situation. The evidenced is a situation for which evidence is provided. In the example Further drilling proved the presence of a Permo-Triassic basin, further drilling is Evidencer and the presence of a Permo-Triassic basin is Evidenced.

In this example, there is an equation or equivalence between the Duke and the most powerful man in England. Following Halliday (Halliday and Matthiessen Reference Halliday and Matthiessen2014: 267)), in this example the Duke is coded Carrier and the most powerful man in England is coded Attribute. The category of Attribute raises the possibility of a more fine-grained analysis; the Attribute can be a quality (‘She is clever’), or an identity (‘She is a Swiss national’), or a role (‘She is the President’). Moving beyond the simple ‘a = b’ Equivalence, the Attribute can relate to a state (‘ … translate a book into German’), the composition of the Carrier (‘The images are composed of pixels’), the purpose of the Carrier (‘ … designate an area for mistakes’), or an addition to the Carrier (‘I decorated my bicycle with flags’). These subdivisions add a useful degree of specificity, but in the end are difficult to apply consistently. For this reason, the superordinate term ‘Attribute’ only is used in the coding.

As previously noted, double coding is used in this project and the semantic field of Equivalence is coded when the construction belongs to other fields also. Examples (11) and (12) illustrate both Equivalence and Cognition, example (13) illustrates both Equivalence and Communication, while example (14) belongs to both Equivalence and Causation. As shown in these examples, the participant roles of Cognizer, Source, Sayer, and Causer are included even in the Equivalence semantic field coding.

1. (11)

   	John	likes	his coffee	hot. (BNC)
   Equivalence	Cognizer		Carrier	Attribute
   Cognition	Cognizer		Phenomenon	… … … … … …

1. (12)

   	His designs	revealed	him	as highly competent. (BNC)
   Equivalence	Source		Carrier	Attribute
   Cognition	Source		Phenomenon	… … … … … … ..

1. (13)

   	John	professes	himself	amazed. (BNC)
   Equivalence	Sayer		Carrier	Attribute
   Communication	Sayer		Message	… … … … … … …

1. (14)

   	The party	elected	a businessman	as its leader. (BNC)
   Equivalence	Causer		Carrier	Attribute
   Causation	Causer		Affected	Outcome

Table 4.5 shows two other participant roles used with the Equivalence semantic field. ‘Actor’ is used to code a participant that acts upon a Carrier without causing the Attribute (as in The attacker captured him alive). ‘Time’ is used in the construction illustrated by Stocks started the day higher.

The second field related to relational processes is given the name ‘Logical relations’. Prototypically, this is intended to provide a coding for constructions that express a relationship between two entities or events that is other than simple equivalence. The following relations are identified and the roles shown in Table 4.5:

• Influence. One entity or process influences another. The influence may be as strong as ‘causation’, but it may also be that one thing enables or facilitates another. The roles involved are Influencer and Influenced. An example is: The durability of the coins [=Influencer] has ensured the survival of these images [=Influenced].

• Condition. One event is conditional upon another. The roles are Conditional and Conditioner. An example is: The opening of the service [=Conditional] necessitated Blackpool buying trams [=Conditioner]. In this example, the service is able to open only because Blackpool buys trams.

• Sequence. Two or more events happen in sequence. There is only one role, Sequential, but the constructions each contain more than one of these elements, with the order of events shown by numbers. An example is: The presentation [=Sequential 1] was followed by a reception [=Sequential 2]. In some cases, a person is construed as placing two events in a sequence. The role Sequencer is used to indicate this. An example is: Martin [=Sequencer] prefaces his lesson [=Sequential 2] with a statement that … [=Sequential 1].

• Correspondence. Two entities or actions are similar to or different from each other, or are connected through coexistence. There is one role – Correspondence – but a construction will contain more than one. Unlike in the Sequence relation, there is no ordering. An example is: The arrows [=Correspondence] correspond to the decision path [=Correspondence]. In some cases, a person construes the correspondence. This role assigned is Correspondence-Causer. An example is: They [=Correspondence-Causer] couched their teaching[=Correspondence] in story form [=Correspondence]. This logical relation is very close to the Equivalence semantic field.

• Evidential. One entity, situation or event provides evidence for another. The roles are Evidencer and Evidenced. An example is: Further drilling [=Evidencer] proved the presence of a Permo-Triassic basin [=Evidenced].

Fields Related to Mental and Verbal Processes

The semantic fields related to mental and verbal processes are Cognition and Communication. Prototypical examples of each would be:

• Cognition (‘thinking’): We thought she was really stupid! (BNC)

• Communication (‘saying’): My brother said that I would make a good banker. (BNC)

Halliday specifies formal criteria that distinguish process types including mental and verbal (Halliday and Matthiessen Reference Matthiessen2014: 249); they are not simply ‘clauses that indicate someone is speaking or thinking’. For the identification of semantic fields rather than process types, however, a much broader definition is adopted. This means that whereas some constructions are similar to the ‘thought that’ and ‘said that’ examples, others are very different, such as those in examples (15) and (16).

1. (15) The images … stuck in his mind. (BNC)

2. (16) Louise chuckled at the recollection. (BNC)

Including examples such as these in the Cognition and Communication fields allows us to capture more data and give a broader sense of the resources available to express them. Similarly, while some terminology from SFG is used to provide participant role labels, the role labels used here are not limited to those used in SFG. Here, I explain the process towards the various decisions taken in specifying participant roles in these semantic fields. The full range of roles can be seen in Tables 4.6 and 4.7.

Table 4.6Participant roles associated with the Cognition semantic field

Table 4.6:Long description

There are eight thinker roles: Cognizer, Desirer, Perceiver, Emoter, Emoter-Action, Source, Inferer, and Time. The cognizer is an actor in a cognitive process. It may be a person or a part of a person. These are examples of Cognizer: I in I remember the teacher telling us…; his mind in The images stuck in his mind; those in Those who believe in theory…; me in The thought hit me that she was lying. The desirer is an actor in a desiderative process. These are examples of Desirer: he in He’s never hankered after the travelling life; Joe in Joe wanted to see the results; he in He still shrank from returning to England. The perceiver is an actor in a perceptive process. An example of Perceiver is she in She watched him cross the road. The emoter is an actor in an emotive process. These are examples of Emoter: she in She likes her coffee strong; I in I envy her her confidence; him in Guilt gnawed at him; oneself in It is easy to delude oneself that the results do not matter. The Emoter-Actor is an actor who brings about a mental state in themselves. An example of Emoter-Actor is he in He deluded himself that… The source is something represented as doing something which leads to an inference. An example of Source is similar tests in Similar tests showed it to have a well-defined field of view. The inferer is a person who draw an inference from a source. An example of Inferer is us in The discovery proved to us that our theory was correct. The time is a time period expressed as a perceiver. An example of Time is the third week in The third week saw him finish the task. There are four thought-about roles: Phenomenon, Phenomenon-topic, Phenomenon-thought and Inference. The phenomenon is a representation of what someone thinks or perceives. Examples are: He wondered what she wanted (what she wanted is Phenomenon), I saw some people I recognised (some people I recognised is Phenomenon), She thought him a fool (him a fool is Phenomenon), She saw him leave (him leave is Phenomenon). The phenomenon-topic is a representation of the entity a person thinks or has an emotion about. An example is the ceasefire in Both sides worried about the ceasefire. The phenomenon-thought is an idea or emotion or expression conceptualised as an active participant in an act of cognition. Examples are: The thought hit me that she was lying (the thought is Phenomenon-thought), Guilt gnawed at him (guilt is Phenomenon-thought), He was seething with anger (anger is Phenomenon-thought). The inference is the information that is derived from a source. An example is that our theory was correct in The discovery proved that our theory was correct. There are four other roles: Cognition-cause, Amount, Proxy, and Anticipatory. The Cognition-cause is an entity that causes someone to think or feel something. Examples are: Their actions decided him against pushing on… (their actions is Cognition-cause); …an investigation that shakes her out of her complacency (an investigation is Cognition-cause); Migrants gain happiness from moving to another country (moving to another country is Cognition-cause). The Amount is an element that indicates that a lot or a little is shown. An example is much in the study fails to reveal much about ordinary ideology. The proxy is a person on whose behalf someone experiences an emotion. An example is those players who played in I felt for those players who played. The anticipatory is it or what in a cleft or pseudo-cleft sentence that anticipates later content. An example is what in What I like about you is your innocence.

Table 4.7Participant roles associated with the Communication semantic field

Table 4.7:Long description

There are three speaker roles: Sayer, Source, and Behaver. The Sayer is the speaker, writer or signer in the representation of an act of communication. It may be a person, part of a person, or an institution. The examples are: Debbie asked them to pay her directly (Debbie is Sayer); Jeremy listened to the soothing voice until he fell asleep (the soothing voice is Sayer). The Source is an entity represented as the source of communication. An example is the results in The results told us that the rocks dated from before the Ice Age. The Behaver is a person who communicates through voice or gesture. Examples are he in He shot her an angry glance, He motioned to them and He chuckled at the memory. There are two hearer roles: Receiver and Inferer. The Receiver is the person to whom speech, writing, signing, or para-linguistic communication is directed. Also, a person who hears, reads or sees a communication. Examples are He apologised to the audience (the audience is Receiver); She listened intently to the broadcast (she is Receiver); He shot her an angry glance (her is Receiver); …beckoning the others to be quiet (the others is Receiver). The Inferer is a person who draws an inference from a source. An example is us in The results told us that the rocks dated from before the Ice Age. There are three speech content roles: Message, Topic, and Inference. The Message is a representation of the content of communication. Examples are: Piaget argued that children…lack the ability to explain… (that children lack the ability to explain is Message); The Council commanded the warden to cease these unlawful practices (to cease these unlawful practices is Message); He asked her how old she was (how old she was is Message); All the witnesses described him leaving the bank (him leaving the bank is Message); …beckoning the others to be quiet (to be quiet is Message). The Topic is a representation of the topic of communication but not the content itself. Examples are: Greenpeace has criticised I C I’s plans (I C I’s plans is Topic); He told us where the treasure was buried (where the treasure was buried is Topic); …both sides complained of ceasefire violations (ceasefire violations is Topic). The Inference is the information that is derived from a source. An example is The results told us that the rocks dated from before the Ice Age (that the rocks dated from before the Ice Age is Topic). There are four other roles: Scope, Amount, Stimulus, and Situation. The Scope is an element that is not acted on by the verb but that with the verb constitutes an act of communication. Examples are She told us a story (a story is Scope); He shot her an angry glance (an angry glance is Scope). The Amount is an element that indicates that a lot or a little is said. An example is a lot in He told us a lot about her. The Stimulus is something that causes a behavioural reaction. An example is the memory in He chuckled at the memory. The Situation is an element used only with talk your way to indicate a situation that is or is not the case as the result of communication. An example is a situation in lie your way out of a situation.

The first roles that might be identified are the ‘speaker’ or ‘thinker’ roles e.g. we and my brother in the previous Cognition and Communication examples. Within the Cognition field, Halliday’s distinction between thinking, perceiving and feeling is maintained (Halliday and Matthiessen Reference Halliday and Matthiessen2014: 258), and the following roles are proposed:

• Cognizer: an actor in a cognitive mental process. E.g. I remember the teacher telling us …

• Desirer: an actor in a desiderative mental process. E.g. Joe wanted to see the results.

• Perceiver: an actor in a perceptive mental process. E.g. She watched him cross the road.

• Emoter: an actor in an emotive mental process. E.g. I envy her her confidence.

In the Communication field there is one equivalent role:

• Sayer: the speaker, writer or signer in the representation of an act of communication. E.g. Debbie asked them to pay her directly.

However, in both the Cognitive and Communication fields, an inanimate entity may be the source of information, as in examples (17) and (18):

1. (17) Similar tests showed it to have a well-defined field of view. (BNC)

2. (18) The results told us that the rocks dated from before the Ice Age. (BNC)

As argued in Hunston (Reference Hunston2013), these examples are instances of interpersonal grammatical metaphor (Halliday and Matthiessen Reference Matthiessen2014: 659–731; Thompson Reference Thompson2014: 246–251). In both cases, the situation is construed as if the information arose unproblematically from inanimate entities: ‘tests’ and ‘results’. Plausible paraphrases, however, would be ‘Our reading of the tests/results led us to the conclusion that it had a well-defined field … / that the rocks dated … ’. Both examples, therefore, could be said to belong to the semantic field of Cognition. In addition, in example (Example 18), an inanimate entity is represented as speaking (told us), as if it was animate. This example, then, belongs also to the Communication field. In both the Cognitive and Communication fields, the role of ‘Source’ is identified to account for instances such as these:

• Source (in Cognition): Something represented as doing something which leads to an inference. E.g. Similar tests … showed it to have a well-defined field of view. (BNC)

• Source (in Communication): An entity represented as the source of communication. E.g. The results told us that the rocks dated from before the Ice Age.

The complementary role to Sayer in the Communication field is ‘Receiver’. This includes people who are aware of a communication, whether or not it is directed to them. Where the person receiving the message infers the message from a Source rather than from a Sayer, they are termed the Inferer. Unlike Receiver, the role of Inferer can be found in both the Communication and the Cognition fields:

• Receiver: The person to whom speech, writing, or signing is directed. Also, a person who listens to, hears, reads, or sees a communication. E.g. I … apologised to the audience. (BNC)

• Inferer: A person who draws an inference from a source. E.g. The results told us that the rocks dated from before the Ice Age.

The other main part of an act of thought or communication is what is variously called the Message (that which is said/written/signed) or the Phenomenon (that which is thought). The problem for coding is that Messages/Phenomena can be expressed in various forms, which can be combined or separated for coding. A common distinction (see, e.g. the previous discussion of FrameNet) is between a Message, which presents a form of words, or from which a form of words can be deduced, and a Topic, which does not. In the example The Council commanded the warden to cease these unlawful practices, it is a reasonable inference that the Council (is supposed to have) said something like ‘You must cease these practices which are against the law’. However, in the example They argued about the height of buildings it is not possible to infer actual words, and the height of buildings might be more reasonably coded as a Topic than as a Message.

It is possible to propose finer distinctions in the Message or Phenomenon category. For instance, in an example such as She called / thought him a fool, it could be argued that the wording ‘he is a fool’ can be deduced and that this Message or Phenomenon represents a non-dynamic Situation. In contrast, the Message or Phenomenon elements of examples such as She saw him leave could be seen to represent a dynamic Situation (‘he left’) and those of examples such as He asked / wanted to see the results could be seen to represent an Action that took place as a result of a desire or communication (‘he saw the results’). The difficulty with this level of granularity is that it is difficult to tell where to stop with making fine distinctions and with more closely related categories there is a greater danger of inconsistency. An alternative and simpler solution is to identify two roles in each of Communication and Cognition:

• Message: A representation of the content of communication. E.g. He asked what she wanted; Piaget argued that children … lack the ability to explain … ; She called him a fool.

• Topic: A representation of the topic of communication, but not the content itself. E.g. Both sides complained of the ceasefire; Greenpeace has criticised ICI’s plans.

• Phenomenon: A representation of what someone thinks or perceives. E.g. He wondered what she wanted; I saw some people I recognised; She saw him leave.

• Phenomenon-Topic: A representation of the physical or abstract entity a person thinks or has an emotion about. E.g. Both sides worried about the ceasefire.

While these roles account for the most central instances of the Communication and Cognition fields, accounting for the many constructions in these fields necessitates the proposal of many other roles, some of which have a very restricted use. For example, the role Proxy is used only with constructions such as ‘sympathise with someone’ or ‘feel for someone’ and the role Emoter-Actor is used only in constructions with a reflexive pronoun (e.g. He deluded himself that …). Tables 4.6 and 4.7 give a full account of the roles used.

Special mention should be made of constructions which sit on the periphery of the Communication or Cognition semantic fields and which are therefore annotated with roles that are less predictable in the context of Communication or Cognition than those previously discussed. Mostly these are connected with metaphoric representations of cognitive acts. In Cognition (Table 4.6) an example would be the role Time, which is used when a period of time is represented metaphorically as though it were a participant in an act of perception. Another example is the role Phenomenon-thought, where an idea or emotion is expressed as an active participant, with the Cognizer or Emoter having a more passive role. In the following example, the verb hit is a material process, though the construction as a whole construes an act of thinking.

• Time: A time period expressed as a perceiver. E.g. The third week saw him finish the task.

• Phenomenon-thought: An idea or emotion or expression conceptualised as an active participant in an act of cognition. E.g. The thought hit me that she was lying.

Specialised roles in Communication (Table 4.7) include ‘Scope’. This is a term used by Halliday (e.g. Halliday and Matthiessen Reference Halliday and Matthiessen2014: 496); it is used here to denote an element that is not acted upon by the verb; instead verb and noun together express an action e.g. ‘sing a song’ or ‘tell a story’. Another example is Behaver, which indicates someone who communicates paralinguistically, through a vocalisation such as laughter, a facial expression or a gesture. In SFG, these constructions would belong to the Behavioural process type rather than Verbal (Halliday and Matthiessen Reference Matthiessen2014: 301–302), whereas in this book they are treated as part of the Communication semantic field. Where the Behaver responds to something, that is termed the Stimulus.

• Scope: An element that is not acted on by the verb but that with the verb constitutes an act of communication. E.g. She told us a story.

• Behaver: A person who communicates through voice or gesture but without words. E.g. He shot her an angry glance.

• Stimulus: Something that causes a behavioural reaction. E.g. He chuckled at the memory.

Recording the Semantic Roles

The constructions that are interpretable in terms of one or more of these nine semantic fields have been annotated on the Transitivity-Net website. Table 4.8 gives an example. The pattern illustrated in that table is V n that (verb + noun phrase + that-clause) and the construction is the ‘persuade someone that something is the case’ construction. The description of this construction is: ‘A person causes someone to change their cognition’. This description distinguishes the construction from similar ones such as ‘inform someone that something is the case’, where a change in cognition may well take place but it is not entailed by the construction. The ‘persuade’ construction is recorded as using only two verbs: PERSUADE and CONVINCE. Example (19) illustrates this construction.

1. (19) Gould succeeded in persuading Darwin that his were the right decisions. (BNC)

Table 4.8A database entry showing participant role annotation

Table 4.8:Long description

The 2 columns are pattern and V n that. Construction name: the persuade someone that something is the case construction. Construction number: 4. Construction description: A person causes someone to change their cognition. Verbs: convince and persuade. Example: Gould succeeded in persuading Darwin that his were the right decisions. The first semantic field is Communication. The roles are: N P 1: Sayer; N P 2: Receiver; that-clause: Message. The second semantic field is Causation. The roles are: N P 1: Cause; N P 2: Affected; that-clause: Result.

The annotation schema treats the construction as having three relevant components: NP1 (Gould in example (19)); NP2 (Darwin); that-clause (that his were the right decisions). It is worth noting that it is the construction rather than the example that is being coded. If the example were in the passive (‘Darwin was persuaded that Gould was right’), Darwin would still have to be interpreted as NP2.

This construction belongs to both the Communication and the Causation semantic fields, so has a double coding: Sayer, Receiver, and Message; and Cause, Affected, and Result. Table 4.8 shows how this appears on the website.

4.7 Conclusion

This chapter has explained the choices, and in some cases the dilemmas, involved in mapping semantic roles onto the formal elements of constructions. This has been discussed in the context of other approaches to the same issue, from SFG, CPA, LG, and FrameNet. The point has been made that the selection of semantic role labels is dependent on the starting point of the research project and its purpose. In this project, the annotation has been carried out in relation to nine semantic fields only. For each field, a finite set of labels is used to annotate all the constructions identified as belonging to that field. The labels are not drawn from any one approach, but they are inspired mostly by the participant roles in SFG. This marks a shift in the focus of this book from an emphasis on constructions to an emphasis on SFG. Chapter 5 provides background on SFG.