On 9 September 2019, a small symposium was held at the British Academy in London in honour of Professor Richard Hudson’s eightieth birthday. The presenters included former students, colleagues and collaborators of Hudson’s as well other figures from the world of linguistics. The themes of the symposium were led by the ideas that Hudson has championed and developed in his own writing, in particular the notion that language is represented in a single cognitive network, seamlessly integrated with the rest of cognition.
Hudson has been a leading figure in linguistics for decades. He was awarded a PhD by the University of London in 1964, having studied at the School of Oriental and African Studies. For his PhD, he undertook field research on Beja, a Cushitic language spoken in Sudan. He was then employed at University College London for six years as a research assistant on a project led by Michael Halliday and subsequently as a member of the regular academic staff in the Department of Phonetics and Linguistics until his retirement in 2004. He was promoted to Reader in 1980 and to a Chair in 1989. He was elected Fellow of the British Academy in 1992. His books have been published by leading presses, and his articles have appeared in the major journals of the field. He is a pioneer in both dependency grammar and cognitive linguistics, with much of his life’s work having been devoted to developing a theory of language called Word Grammar (WG), which expresses Hudson’s theorising about language and cognition, and about dependency theory and syntax.
Word Grammar synthesises ideas from the two traditions we have just mentioned, dependency grammar and cognitive (or usage-based) linguistics. In Hudson’s work, language is understood as part of general cognition, and it includes all knowledge about language including social and affective knowledge. For Hudson, knowledge (including knowledge of language) is organised in a symbolic, classified network, with default inheritance as the logic that underwrites generalisations. Hudson was one of the pioneers of cognitive linguistics: Hudson (Reference Hudson1984) on WG was the first published book presenting a theory of grammar cast in the terms of cognitive linguistics. Many of the ideas that Hudson developed or espoused early in their history are now in the mainstream, particularly the idea that language is a cognitive network. However, Hudson is not only a theorist: he has made a substantial contribution to the study of syntax and semantics, particularly English grammar, with further contributions examining a range of phenomena from semitic interdigitation to the analysis of ‘fused’ forms such as the French pronoun/determiners au and du.
Word Grammar was first developed and proposed in Hudson (Reference Hudson1984), as we have noted; it was applied to English in a formalism that was intended to be computationally tractable in Hudson (Reference Hudson1990) and updated with a renewed focus on the psychological constraints underscoring language in Hudson (Reference Hudson2007). A subsequent textbook (Hudson Reference Hudson2010) explicitly addresses Hudson’s major research themes: Part I explains his theorising about how the mind works; Part II elaborates a theory of how language works; and Part III applies the theoretical ideas to an account of English. One of the key claims of WG is that language is socially embedded, with words being actions by their speakers: indeed, it is this claim that leads to the claims about cognitive structure. A number of Hudson’s ideas have developed and proliferated to different fields of linguistics, such as sociolinguistics (Hudson Reference Hudson1996), as well as education.
We introduce the key ideas of WG here. As we have noted, WG argues that language is represented in a fragment of a larger cognitive network which the language network is merely part of. The representation is a radical network, with atomic nodes, unlike the complex nodes of some versions of Construction Grammar (Diessel Reference Diessel2019). That is, it represents knowledge of language in a symbolic, classified network, which has classification in terms of default inheritance hierarchies that permit the overriding of information, on the one hand, and multiple classification (multiple inheritance) on the other. This basic premise of WG, that language is mentally represented as part of a wider cognitive network, is called the ‘Network Postulate’ by Hudson (Reference Hudson1984: 1): cognition is a network, and language is part of that network. And, given default and multiple inheritance, the network is classified. The dependency architecture of WG syntax follows from the Network Postulate. We take these claims and present their different facets in the next section, where we present WG.
I.1 Word Grammar
In order to describe the network, we start with WG’s theory of categories – default inheritance – and then we show the differences between relational and non-relational concepts. Next, we move on to discuss dependency syntax, and WG semantics and morphology. We end this section with a discussion of some of the theory’s research results. Formally, WG representations are built around a skeleton of five primitive relations, Isa (‘is an instance of’), Argument, Value, Identity and Quantity. We introduce these relations in the following subsections. The main introductory text to WG is Hudson (Reference Hudson2010), which presents the key ideas of WG in some depth.
I.1.1 Inheritance and Multiple Inheritance
Inheritance is a very simple idea: knowledge is stored in a taxonomic hierarchy (Hudson Reference Hudson2010: 12–14), and information is inherited from ‘higher’ (i.e. more general) nodes in the hierarchy when it needs to be drawn upon. To use an example from Hudson (Reference Hudson2007), the question of whether a bird has a heart is mostly not relevant to our interactions with birds; we draw on the information when it is needed, and we draw on it from the classification of ‘bird’ as a subtype of ‘animal’. Default inheritance is found in a number of theories, including Network Morphology (Brown & Hippisley Reference Brown and Hippisley2012) which exploits DATR (Evans & Gazdar Reference Evans and Gazdar1996), Cognitive Grammar, where it is called ‘schematicity’ (Langacker Reference Langacker1987, Reference Langacker1991), Head-driven Phrase Structure Grammar (Pollard & Sag Reference Pollard and Sag1994) and Construction Grammar (Goldberg Reference Goldberg1995; Sag Reference Sag, Boas and Sag2012). Default inheritance ultimately derives from Quillian (Reference Quillian and Minsky1968), who developed a model of inferential, analogical reasoning in a semantic network. It was mathematically formalised in Touretzky (Reference Touretzky1986). Hudson has worked out a particular model of default inheritance for WG, described in Hudson (Reference Hudson2007: 10–30). The predicate of default inheritance is Isa, which means ‘is an instance of’. The proposition, ‘Elmer isa Elephant’ means, ‘Elmer is an instance of the category Elephant’. Isa is one of the primitives of WG theory.
Most theories of inheritance allow for multiple inheritance (Hudson Reference Hudson2010: 22–4). This makes obvious sense: imagine your friend’s dog, Jumble. The concept ‘Jumble’ belongs in an inheritance hierarchy, with a number of properties being available at each level of abstraction: ‘Jumble isa border terrier’ tells us information about Jumble’s breed, which includes the typical properties of that breed, such as its usual size. Isa is transitive, so at a greater degree of abstraction ‘border terrier isa dog’ tells us that Jumble barks, whines, eats meat and does not have retractile claws. From ‘dog isa mammal’ we can infer that Jumble is warm-blooded, that his or her mother gave birth to a live animal and that he or she initially fed on maternal milk. And in turn ‘mammal isa animal’ offers up the inferences that Jumble has limbs, a heart, skin, a digestive tract and does not have roots or leaves.
But the concept ‘Jumble’ also inherits from ‘pet’. This tells us that he or she lives with your friend as a member of their family. Pets are domesticated animals – we don’t keep dingos as pets, for all that they are canids that can interbreed with domestic dogs – so ‘pet isa domesticated animal’. And different domesticated species have been domesticated for specific purposes: not all domesticated creatures are pets. (Sheep are domesticated, but we do not have pet sheep sleeping in our kitchens.) This kind of multiple inheritance (Hudson Reference Hudson2010: 22–4) works perfectly well because the categories ‘pet’ and ‘border terrier’ do not have conflicting properties. Note that because ‘pet’ inherits from ‘animal’, the concept ‘Jumble’ inherits from ‘animal’ by transitivity in both taxonomies.
However, the various theories that permit multiple inheritance have to address the problem of feature conflict in multiple inheritance (Touretzky Reference Touretzky1986: 8–11). This is the ‘Nixon Diamond’ (Hudson Reference Hudson2010: 24), where two contradictory inferences are inherited. President Nixon was a Quaker, and Quakers famously are opposed to fighting wars. But Nixon was also a Republican, and as such had no constraints against war. The properties ‘anti-war’ and ‘pro-war’ are (obviously enough) contradictions, so the theory needs a way of handling such contradictions. Different theories take different approaches. DATR and the theories that derive from it use a model of orthogonal default inheritance, which privileges one of the inheritance pathways. Word Grammar, on the other hand, works with ‘free’ default inheritance. This means that in the case of a contradiction, there is either a stipulated solution, as Nixon himself stipulated that he was pro-war, or the contradiction blocks a form. Hudson (Reference Hudson2000) used the fact that contradiction blocks a form as an explanation of why *I amn’t is blocked in many varieties of English.
One of the attractions of default inheritance from the point of view of WG and theories like it is that it makes sense of human inferential processes, and it is compatible with many theories, grounded in the research results of cognitive psychology, of how human reasoning works. This is evident in the research literature that begins with Quillian (Reference Quillian and Minsky1968) and Collins and Quillian (Reference Collins and Quillian1969). Quillian takes his model of default inheritance to be the basic link in a semantic network, and Hudson has seen language as a part of the larger cognitive network since the very beginnings of his theorising, with work on WG bringing that idea into very clear focus (Hudson Reference Hudson1984: 1). Default inheritance is a key feature of WG’s mentalism. There is a thorough presentation of the WG theory of inheritance in Hudson (Reference Hudson2010: 10–30).
I.1.2 Relational and Non-relational Concepts
Word Grammar draws a fundamental distinction between relational and non-relational concepts (Hudson Reference Hudson2010: 39–43). Non-relational concepts are straightforward: ‘table’, ‘dog’ and ‘grunt’ are all non-relational. It is a bit harder to identify relational concepts because relations are abstract. We can start with family relations: ‘mother(-of)’ identifies an individual, but it identifies an individual who is in a relationship with other individuals. Marge Simpson is the mother of Bart, Lisa and Maggie. If Bart, Lisa and Maggie did not exist, Marge would not be a mother. Nouns like mother involve two kinds of information: the non-relational entity and the relation associated with that entity. But there are other relational terms that are purely relational. The prepositions before and after just name positional relations; in the linear order of this sentence, before is before after. Likewise, if you look from Admiralty Arch in London down the Mall to Buckingham Palace, St James’s Park is before Green Park. There is also ordering in temporal systems: January comes before February.
In WG, there is a split at the top of the inheritance hierarchy into relational and non-relational concepts. Relational concepts are defined by a pair of primitive relations, ‘argument’ and ‘value’, making each relation a function such as before(Green Park)=St James’s Park. In that formula, Green Park is the argument and St James’s Park the value. Similar considerations work in language. In a sentence such as Jumble barked, the verb barked is the argument of the relation subject and Jumble is the value. This can be shown diagrammatically as in Figure I.1.
Figure I.1 Jumble barked.
As Figure I.1 shows, by having argument and value as primitive values (Hudson Reference Hudson2010: 39), it is possible for each relation to be a node in the inheritance hierarchy. The figure also shows multiple inheritance: the word barked is classified as an instance of the lexical entry bark and as an instance of the morphosyntactic category, past.
As we have noted, relational concepts are found as part of the meanings of words such as mother and as the meanings of words such as before and after. They are also part of the technical description of language. Word Grammar is a dependency grammar, and each dependency is a grammatical function: subject, direct object, indirect object, adjunct and so forth are all relational concepts. Word Grammar syntax consists of non-relational concepts (words) linked by relational concepts (dependencies) with both types of concept classified by isa relations.
There are also other relational concepts involved in the description of language. Semantics is a discrete level of the language network with its own descriptive vocabulary, and so are morphology and phonology. These levels of language have their own relations. For example, the semantics of event structure has a vocabulary including the relations ‘er’, ‘ee’ and ‘result’. ‘Er’ is a generalisation over the semantic relations that link to the subject of an active voice verb; ‘ee’ generalises over the relations that link to a direct object; and ‘result’ is a relation between states of affairs. The different domains of grammar are also linked by relations. The morphosyntactic feature category past is linked to the morphological element {-ed} by a realisation relation.
I.1.3 Dependency Syntax
The relational architecture requires a dependency syntax, and WG eschews phrase structure as the basis for its syntax, instead choosing to analyse sentence structure as involving pairwise relations between words (Hudson Reference Hudson2010: 146–62; see especially 146–8). Whereas in a Phrase Structure Grammar, grammatical relations can only be inferred from the phrase structure graph, in a Dependency Grammar, grammatical relations are basic, and it is phrases that are emergent, at least in languages with a sufficiently fixed word-order type. The syntax is maximally simple: it consists only of words and relations between them. Such a structure makes the syntax a network.
Theories of syntax have to capture a range of phenomena such as extraction and the island constraints, coordination, the distribution of reflexives, raising and control, and interface properties such as agreement and argument linking. The WG analysis of these phenomena is handled entirely with dependency structures, except in the case of coordination, which requires a limited use of word-strings, which need not be phrases (Hudson Reference Hudson1990: 404–21). The dependencies of WG are classified in a taxonomy (Hudson Reference Hudson2010: 152–4, 159), allowing for generalisations to be very fine-grained or highly general.
1.1.4 Semantics
There is a useful introductory presentation of WG’s theory of semantics in Hudson (Reference Hudson2010: 220–45). All of WG’s representations are intramental and the grammar classifies mental representations of utterances. The semantics includes not only the compositional semantics of the elements of the utterance but also an understanding of the utterance in its situational context. This allows for a simple understanding of deictic pronouns such as I and you because speaker and hearer are both included in the representation of the utterance. It also affords a straightforward approach to tense, which is inherently related to speech time. At the same time, it requires a particular, arguably radical, approach to the semantics–pragmatics divide; in WG there is no divide: the knowledge that we use you to address our interlocutor is integrated, in a language such as French, with the knowledge that tu is informal and vous is formal.
The phenomena that theories of semantics need to be able to describe include quantification, co-reference, the definite/indefinite contrast, event structure and modification, as well as more encyclopaedic facts to do with word meaning and the pragmatic facts of how utterances relate to the speech situation. Existential quantification in WG is just a matter of instantiation: i.e. it involves a link from a permanently stored semantic node associated with a lexical entry to a mental equivalent of a token, which is associated with the instantiated word. One way of seeing how this works is by looking at how singular count nouns work. The word dog cannot occur in an argument position without a determiner. The lexical entry of the noun dog includes a sense and in Hudson’s analysis when dog occurs with a determiner, the noun and the determiner share a referent which instantiates the sense of the noun (Hudson Reference Hudson2007: 226). This shows how WG has a semantic category ‘referent’ as well as ‘sense’, despite being entirely intramental: a WG referent is a mental representation, not an entity in the world. WG referents are therefore also available as part of the system of discourse reference, which means that the referent of the dog or a dog is not only an instance of the more abstract generalisation over your experiences of members of the category ‘dog’ but is also available as the antecedent of himself in the dog tripped and surprised himself. In this case, the referent of himself is identical with the referent of its antecedent; this gives us another of WG’s primitives, the ‘identity’ relation. The concepts we identify as the referents of the two words are identical.
Other kinds of quantification involve introducing sets into the theory’s ontology. We need these sets in order to be able to capture the semantics of plurals, as well as numbers, the quantifiers such as every, each and all, and sentences involving coordination such as Kim and Alex bought a house. The WG theory of quantification is introduced in Hudson (Reference Hudson2007: 31–6). In order to capture the set semantics, ‘quantity’ is identified as a primitive; a quantity of zero is how WG represents ‘not’ (Hudson Reference Hudson2007: 36). In terms of the formal representation, sets are defined as a particular kind of node in the network. One feature of sets is that they have to have a member (otherwise the set is the empty set), so each set is defined as having a typical member, which is a definitional property of the set and not part of its cardinality. Hudson (Reference Hudson2007: 228–32) extends the earlier analysis to include analyses of the joint and the distributive interpretations of Two researchers wrote three articles.
I.1.5 Relating Semantics to Word Grammar’s Wider Themes
We have seen how WG has a semantics that can capture the quantificational facts familiar from more mainstream theories entirely within the architecture of its network. But it is also worth noting that the nature of the network architecture is not just a formal tool that allows the formal description. It is also one of the claims of the theory: WG posits that human cognition is represented in a symbolic mental network – also known in the literature as a semantic network (Hudson Reference Hudson2010: 70–100). Because this network is classified and associative, there is a claim that the very network is based on ordinary human experience. The child acquires the building blocks of the network through its social experiences – the network is built up out of the building blocks of social cognition. The word mummy expresses a concept that relates directly to the child, with both a node and relation in its semantic structure. Word Grammar hypothesises that the very building blocks of language and cognition are calqued on these early social experiences.
This hypothesis leads us to Hudson’s ideas about language and society and the cognitive theory of sociolinguistics. Above, we began our discussion of relational concepts by talking about the word mother. One of WG’s claims is that semantics is inextricably bound up with social meaning. Part of this claim comes from how WG understands language learning in the infant: social experience gives the infant a vocabulary of relational and non-relational concepts, and language is built up from that starting point. But it also derives from the claim that the network classifies utterances which are embedded in the speech situation. Understanding the relationship between an utterance and its speech situation gives us some fundamental linguistic concepts – indexicality and deixis, for example. However, it also requires us to imagine, as analysts, that other aspects of the interpersonal situation will also affect an utterance’s interpretation. Take an utterance such as May we come in? (Fillmore Reference Fillmore1997: 8). In asking for permission, the speaker reverses the usual deictic orientation of come from being oriented towards the speaker to being oriented towards the addressee. This empathic re-orientation is part of placing the ‘control’ of the situation in the hands of the addressee, alongside the interrogative clause structure and the deontic modal. There is no understanding of the semantics of May we come in? without an awareness of these contextual facts, which are therefore part of WG semantics. And this is a short step from discussions of power, solidarity and identity, which are the stuff of sociolinguistics. Semantics and sociolinguistics are inextricably bound: for WG, meaning is social (Hudson Reference Hudson2007: 211–48).
I.1.6 Realisational Morphology
The morphological theory of WG is laid out in Hudson (Reference Hudson2007: 63–116); there is a very brief introduction to some key concepts in Hudson (Reference Hudson2010: 122–6) and Hudson (Reference Hudson2010: 131–6). As with all other parts of language, it is organised around the principle of the network. WG morphology is an autonomous domain of structure (Aronoff Reference Aronoff1994; Sadock Reference Sadock1991): the morphological structures are not directly meaningful, nor do they reduce to phonology. These choices place WG morphology at odds with morpheme-based theories and align it with realisational theories.
It is straightforward to see that morphology does not reduce to phonology: the word dancer has the morphological structure {danc-}+{er}, but in the phonology the /s/ is the onset of the second syllable, not part of the coda of the first. The mismatch shows that the phonology and morphology are distinct. Similar arguments apply to the relationship between the parts of morphological structure and morphosyntactic features. The cumulative exponence of {-es} in the Latin verb cantes realises second person, singular, subjunctive mood, active voice and present tense. Unlike in agglutinating languages, there is not a direct relationship between form and function.
Within this broad architecture, WG has a vocabulary of categories and relations which allows it to develop a rich set of analyses of morphological phenomena. These include the verbal morphology of the Bantu language, Swahili (Creider & Hudson Reference Creider and Hudson1999); clitics in Bosnian/Croatian/Serbian (Čamdžić & Hudson Reference Čamdžić and Hudson2007); the blocking of *amn’t (Hudson Reference Hudson2000); and phenomena such as the ‘fused’ realisation of preposition and article in examples such as French au (Hudson Reference Hudson2007: 100–4). Many of the details of fused realisation were also worked out in Rosta (Reference Rosta1997). Gisborne (Reference Gisborne, Audring and Mansini2018) presents an overview of WG morphology.
I.1.7 Results
Word Grammar has a number of research results. Hudson was among the first to develop the idea that in a determiner–noun construction, the determiner is the head (Hudson Reference Hudson1984: 90–2). In Hudson’s analysis, determiners are pronouns, rather than having the category–inclusion relationship the other way around as is more common since Postal (Reference Postal, Reibel and Schane1969). Hudson (Reference Hudson1995) was the first to explore dependency distance as a measure of syntactic complexity, an idea subsequently developed by Gibson (Reference Gibson1998), Hiranuma (Reference Hiranuma2002), Liu (Reference Liu2007), Gildea and Temperley (Reference Gildea and Temperley2010), Duran Eppler (Reference Duran Eppler, Gerdes, Hajičová and Wanner2014) and others. More recently, Hudson has explored multiple inheritance as an explanation of the *amn’t gap, as we have noted, and as an explanation of the distribution and behaviour of English gerunds (Hudson Reference Hudson2003). A more recent paper (Hudson Reference Hudson2018a) argues for a WG theory of pied-piping, which invokes the systems and structures of general cognition as well the formal model of WG. These results are embedded in a series of ongoing discussions about the nature of language and its mental representation, such as Hudson (Reference Hudson2018b).
I.2 Hudson’s Wider Ideas about Language
As we have seen, WG captures Hudson’s theorising about the relationship of language to the rest of cognition, and his theorising about syntax within a dependency framework. Although we have discussed social cognition, we have not explored Hudson’s contribution to sociolinguistic theorising or his contribution to having language represented in the school curriculum in the UK. By and large because of the view that language and society are inextricable, Hudson has made a significant contribution to language in education. He has argued that understanding how language works is an essential part of understanding ourselves and, therefore, that it ought to feature in the school curriculum. This has resulted in a lot of work with politicians; with organisations such as the LAGB, both as President and in his work for the Association’s Education Committee; and with the Committee for Linguistics in Education, as well as his work in organising the Linguistics Olympiad. We want to acknowledge this contribution of Hudson’s although there is not any work in the volume that addresses it.
In this book, some chapters relate Hudson’s ideas about language and cognition to a diverse set of linguistic problems which shows their general applicability and consistency across domains of grammar, including aspects of language change. Others look more closely at syntax. The first chapter, by Gisborne and Müller, locates WG in its intellectual context, talking about it as a dependency grammar, a constraint-based grammar and a cognitive linguistic theory. Some of the contributions explore Hudson’s claims about language in cognition and the argument that language is not a discrete system. This is true of Rosta’s chapter on the question of whether structures argued by Hudson (Reference Hudson and Roca1992) to occur both in syntax and in general cognition can also be found in phonology, as well as Gisborne’s and Trousdale’s chapters on language change, Imrényi’s on the conceptual metaphors underlying phrase structure grammars and dependency grammars, and Hollmann’s exploration of the linguistic embedding of aspects of social cognition.
Four chapters, Kahane’s, Osborne’s, Nordström’s and Aarts’s, investigate issues in syntax. Kahane is concerned with criteria for heads and dependents, and revisits the criteria for head-hood in Hudson’s earlier work. Osborne takes another look at the debate between Hudson and Dahl about whether syntax needs phrasal nodes or not (Dahl Reference Dahl1980; Hudson Reference Dahl1980), and presents additional new evidence that phrasal nodes are not required. Aarts is concerned with a problem in categorisation: the category of participles in a particular syntactic context in English. Although he does not engage directly with WG theory, the nature of categories and the classification of marginal cases is central in Hudson’s work. Nordström explores the direction of dependency between two clauses, one a complement of the other, when subordination is marked and when it isn’t. Moving beyond pure syntax, the chapter by Eppler, Luescher and Deuchar is an exercise in theory comparison, looking at the success with which three different theories predict code-mixed determiner–noun constructions. The volume ends with Niu and Liu’s chapter on dependency distance, which is a fitting place to end, as Hudson originally proposed dependency distance as a measure of syntactic complexity in 1995.
We finish this Introduction with summaries of the twelve chapters.
I.3 Summaries of the Chapters
Chapter 1: Word Grammar in Its Intellectual Context (Nikolas Gisborne & Stefan Müller)
In this chapter Gisborne and Müller relate WG to a number of different trends in linguistic theorising and explain the various traditions that the theory belongs to. Word Grammar belongs in three main theoretical traditions: Dependency Grammar, Constraint-based Grammar, and Cognitive Linguistics. The authors show how WG relates to these approaches, and explore how the network model of linguistic representation adopted by WG relates to each tradition as well as exploring aspects of how it has contributed to these research traditions.
Chapter 2: Raising in Phonology (And Rosta)
Rosta proposes that there is evidence of raising in phonology. Hudson (Reference Hudson and Roca1992) argued that the raising pattern familiar from syntax also occurs in semantics and in extra-linguistic general cognition. In Hudson’s view, this evidence that linguistic structures are of the same sort as the structures of general cognition supports the hypothesis that language is part of general cognition. It might therefore be expected that phonology will also involve structures found in general cognition, and Hudson (Reference Hudson and Roca1992) concludes by wondering whether raising is also to be found in phonology. As Tallerman (Reference Tallerman2006) notes, there is a long and unconvincing history of attempts to demonstrate that phenomena normally thought of as characteristically syntactic occur in phonology as well, but a truly convincing demonstration would show that movement, normally understood to be quintessentially syntactic, also occurs in phonology. Raising is a kind of movement. In this chapter, Rosta identifies two instances of raising in phonology. The first instance is z-suppression, where the genitive z ending is ‘suppressed’ when the base already has a z ending (child’s, children’s, kid’s, *kids’s, kids’). The second is a proposed phenomenon of ‘raising to onset’: the chapter presents a dependency structure for English phonology and argues that if it is to adequately account for positionally conditioned consonant allophony and if, as is plausible, weak (i.e. schwa) syllables contain no phonological nucleus, then there must be raising to onset analogous to syntactic raising to subject.
Chapter 3: Grammatical Change in the Network (Nikolas Gisborne)
In this chapter, Gisborne discusses the emergence of the have perfect in English, paying particular attention to the development of the perfect participle. He uses this case study as a vehicle for discussing what causes directionality in language change. The have perfect is one of the features that is a common property of Standard Average European, and an open research question asks why this change happens recurringly in related languages. Gisborne makes three main claims: that the change to a have perfect only involves one strictly syntactic change – the reanalysis of a complement as an adjunct; that there are semantic changes in the participle driven by the bleaching of have; and that the emergent new category of participle is driven by these semantic changes. The evolution of participles involves the creation of a new linguistic category, in a particular grammatical environment, which is analogous to an ecological niche in evolutionary change. The main argument is that the mental network of WG, together with its theory of speakers and its theory of categories, offers an explanatory account of speaker innovation and its role in language change.
Chapter 4: Morphological Change in Word Grammar (Graeme Trousdale)
Trousdale’s chapter provides a further contribution to work on WG and language change. It explores particular developments in English derivational morphology in order to look in more detail at what kinds of changes occur in the language network over time. This relates to discussions in other cognitive linguistic theories about diachronic variation in the language network, especially in terms of changes to nodes and changes to links between nodes. The main claims that are made in this chapter are, first, that much change in the network is very local and involves micro-steps but, second, that some changes can occur which involve more significant restructuring, for instance where language users have reanalysed a part of a word as a word in itself. Since the central goal of WG is to understand the grammar of words, such changes can be revealing in terms of the theoretical underpinnings of the framework.
Chapter 5: The Metaphorical Bases of Constituency and Dependency (András Imrényi)
In this chapter, Imrényi offers a cognitive linguistic analysis of metaphors informing the conceptualisation of sentence structure. In line with cognitive linguistics, Imrényi assumes that the construal of this highly abstract conceptual domain necessarily has a metaphorical basis. Accordingly, he argues that key differences between alternative syntactic theories can be linked to the metaphorical choices they make. The chapter begins with a critical comparison of constituency and dependency syntax. The key virtue of the latter is seen in its ability to focus on relation types rather than unit types and unit boundaries. Subsequently, it is argued that the metaphor the sentence is a building serves as the core metaphor underlying constituency analysis, whereas the sentence is a family may play a similar role in dependency grammar. The final part of the chapter discusses two instructive metaphors invented by the Transylvanian Hungarian polymath Sámuel Brassai, both supporting a dependency grammatical understanding of sentences, namely the sentence is a feudal society and the sentence is a solar system. It is demonstrated that each of Brassai’s metaphors has its share of advantages and may be of great service to dependency grammar in language pedagogy.
Chapter 6: From Social Psychology to Cognitive Sociolinguistics: The Self-Serving Bias in Causative Constructions (Willem B. Hollmann)
Hollmann explores the relationship between language and social cognition, taking the discussion beyond the claims made in Hudson (Reference Hudson2007). The main aim of his chapter is to show that the notion of the ‘self-serving bias’, well established in social psychological research, may have an impact on the way in which speakers verbalise certain experiences. Hollmann hypothesises that this perceptual bias interacts with other factors – specifically, gender stereotypes (as defined by psychologists and linguists) and modesty (as defined in linguistic pragmatics). He presents corpus evidence for the relevance of the self-serving bias and the complex interplay with gender stereotypes and modesty, based on variation between three different causative constructions (CAUSE, X MAKE Y happen and X BRING about Y) as well as the use of the adverbs cleverly and stupidly. His analysis focuses on the co-occurrence with personal pronoun subjects – specifically, differences in terms of person (first vs. third) and gender (masculine vs. feminine). In this chapter, he draws connections between social psychology and linguistic constructs, in a promising development of WG ideas about the social embedding of meaning.
Chapter 7: Hudson on Heads: About the Distributional Criteria (Sylvain Kahane)
In this chapter, Kahane discusses the criteria used to define dependency structure and to characterise the syntactic head of every syntactic unit. Hudson was the first, in the 1980s, to really try to justify his choices of analysis in dependency syntax by using both distributional criteria with and without removal (of subordinate elements), but he did not state them explicitly. This chapter is an attempt to propose distributional criteria for choosing a syntactic head and for proving the head status of words that can generally not stand alone, such as determiners, prepositions or auxiliaries. Three criteria are stated: a Positive and a Negative distributional criterion with removal and a Distributional criterion without removal, as well as a Distributional criterion for the head of a sentence. These criteria are compared to criteria used in Garde (Reference Garde1977), Hudson (Reference Hudson1984, Reference Hudson1987) and Mel’čuk (Reference Mel’čuk1988), showing that Mel’čuk circumvented them in practice, while in practice Hudson applied them quite systematically even without having stated them clearly.
Chapter 8: Ordinary French Houses: Revisiting the Dependency vs. Phrase Structure Debate (Timothy Osborne)
In this chapter, Osborne revisits the debate about dependency vs. phrase structure that occurred in 1980–1981 between Richard Hudson, on the one hand, and Östen Dahl and Pertti Hietaranta, on the other. The debate is taken up by looking at adjective scope. Dahl’s argument in favour of phrase structure based on adjective scope (e.g. ordinary French house) can be convincingly countered in terms of the component unit of dependency syntax. The component and two additional units of dependency syntax, the full component and the full catena, are presented and developed here. The claim is that the motivation for the layered trees of many phrase structure grammars disappears if the much flatter Dependency Grammar analyses acknowledge these units of dependency syntax. The overarching message, then, is that barring the analysis of coordinate structures, the theory of syntax does not need the higher nodes associated with phrase structure, in line with Hudson’s original message back in 1980.
Chapter 9: Dependency Grammar and Subordination (Jackie Nordström)
Nordström (Reference Nordström2014) argued for WG style syntactic representations, and in this chapter she develops these ideas, arguing that language discretely combines lexical items into dependency structures rather than recursively embedding syntactic objects into phrase structures. Nordström argues that the ubiquity of subordinators and other subordinating elements in the languages of the world, as well as the fact that these are obligatory in many contexts, clearly shows that language is not recursively embedding. If language were recursively embedding, the subordinators would be redundant, since subordinate clauses would automatically be embedded under their superordinate predicates. Other potential functions of universal subordinators such as Germanic that are discussed and addressed. The chapter illustrates its arguments with examples from English, Swedish and German where the universal subordinator is needed to signal that the clause is subordinated to the matrix predicate, and cases where it is not needed but where other subordination strategies are employed instead. Finally, Nordström demonstrates that in those cases where the subordinator is omitted and there is no other subordinating strategy, the clauses become desubordinated.
Chapter 10: Participles and So-Called Synthetic Compounds as Attributive Modifiers in English (Bas Aarts)
Aarts’s contribution addresses the careful focus of WG work on English grammar, as well as the theory of categories. Aarts is interested in the categorial status of English participles as modifiers of nouns. English allows participial forms of verbs to modify nouns, as in the following example:
The Rapids in 1834 was a straggling village whose 44 residents clustered mainly along the river on the east side of a single dirt path – the future Front Street.
He addresses the question of whether attributive V-ing premodifiers in noun phrases are adjectives or verbs. He discusses the evidence for treating (some of) these formatives as adjectives, e.g. deverbal adjectives such as interesting, satisfying, etc., and explores the evidence for regarding others, such as straggling in the example above, as verbs. He also discusses so-called synthetic compounds, such as cake-eating (bear), beer-swilling (neighbour) and wall-straggling (flower). These are analysed as verbal constructions, rather than as adjectives, with the evidence involving the semantics and combinatory properties of V-ing premodifiers in English noun phrases. Aarts shows that V-ing premodifiers can take a full range of dependents and that, with some restrictions, combinations of dependents, e.g. a complement and an adjunct, are also possible.
Chapter 11: Evaluating the Predictions of Three Syntactic Frameworks for Mixed Determiner–Noun Constructions (Eva Duran Eppler, Adrian Luescher and Margaret Deuchar)
This chapter shows that a WG analysis can accommodate code-mixed constructions. Eppler, Leuscher and Deuchar present a comparative evaluation of WG, the Minimalist Programme (MP), and the Matrix Language Frame Model (MLF) regarding their predictions of possible combinations in a corpus of German–English mixed determiner–noun constructions. Word Grammar achieved the highest accuracy score. The comparison furthermore reveals a difference in accuracy between the three models and a significant difference in the accuracy of the predictions between WG and MP. The analysis suggests that these differences depend on assumptions made by the theories and the mechanisms they employ. The difference in accuracy between the models, for example, can be attributed to the MLF being concerned with a broader notion of agreement in language membership between the verb and the subject DP/NP of the clause. The significant difference between WG and the MP can be attributed to the distinct roles features play in the two syntactic theories and how agreement is handled. Based on their results, the authors draw up a list of characteristics of feature accounts that are empirically most adequate for the mixed determiner–noun constructions investigated and conclude that the syntactic theory that incorporates most of them is WG (Hudson Reference Hudson2007, Reference Hudson2010).
Chapter 12: Factors Influencing Dependency Distance: An Account of the MDD Variation between Chinese and English (Ruochen Niu and Haitao Liu)
Niu and Liu explore dependency distance (DD), which is a measure of syntactic complexity proposed by Hudson (Reference Hudson1995). Based on Hudson’s pioneering work and dependency treebanks of twenty languages, Liu (Reference Liu2008) identified a universal principle later called ‘dependency distance minimisation’. The mean DD of natural languages tends to be minimised below four words, which is a working-memory constraint. However, there also seem to be considerable differences among languages, particularly between Chinese and English, as noted in Hudson (Reference Hudson and Liu2009). Niu and Liu investigate whether the DD variation between Chinese and English is induced by corpus-based factors (such as sentence length, genre and annotation scheme) or by deeper motivations (such as syntactic differences between languages). They find that while corpus-based factors may influence mean DDs, the DD variation between Chinese and English is more likely to be the result of different syntactic structures and processing mechanisms, such as the use of words rather than affixes to express tense in Chinese. They argue that these findings suggest that the syntactic patterns of human languages are shaped by cognition and grammar together.
