The grammar network: How linguistic structure is shaped by language use by Holger Diessel (review)

The network is one of the key concepts that describe our life experience. We communicate via social networks and use the World Wide Web. The masses of big data we produce, consciously or not, are analyzed by neural network algorithms. It is not surprising that networks are a popular concept in linguistics as well. Well-known examples are the construction in construction grammar, radial polysemy networks in cognitive semantics, and semantic maps in typology. The grammar networkby Holger Diessel takes the network approach to a new level, integrating different strands of usage-based linguistics and discussing a multitude of different phenomena of grammar and lexicon, from lexical polysemy to word order, and from parts of speech to morphological productivity. The aim of his book is to elaborate on and integrate two main ideas of usage-based linguistics: first, all aspects of linguistic structure are emergent and fluid, being shaped by domain-general processes in language use; second, the grammatical system is organized as a network.

The book consists of eleven chapters. The introductory chapter describes the aims of the book and summarizes the general principles of usage-based linguistics. It questions the usefulness of such distinctions as competence vs. performance, synchrony vs. diachrony, and words vs. rules. Usage-based linguistics sees grammar as an emergent system, which has evolved for the purposes of communication and processing on the basis of general cognitive principles. From this follows an important methodological conclusion: 'we cannot approach the study of grammar with a predefined set of primitive categories. On the contrary, what we need to explain is how linguistic categories evolve, stabilize and change' (6).

In the first part of the book D introduces the basic assumptions and concepts, describing the architecture of the grammar network (Ch. 2) and the domain-general principles of language use (Ch. 3). The most important elements of a grammar network are signs (constructions and lexemes). The links (edges) in a network can be symbolic (connecting form and meaning), sequential (connecting elements in sequences, similar to syntagmatic relationships), and taxonomic (connecting patterns at different levels of abstraction). Analogous to activation strength in neural networks, the links have different weights, which depend on different factors, such as frequency of occurrence in linguistic input and output, conceptualization, and pragmatic inference. A grammar network is nested, which means that nodes of a network can themselves be analyzed as networks.

Different from Goldberg's (1995) construction grammar, D makes a distinction between constructions and lexemes. Lexemes are monomorphemic words and other morphemes, which tap directly into world knowledge. In contrast, constructions are meaningful templates with slots for other linguistic expressions, which provide instructions for integration of lexical expressions into a coherent semantic representation. Based on this distinction, three further types of links are proposed: lexical (connecting lexemes with similar or contrastive forms and meanings), constructional (connecting constructions at the same level of abstraction), and filler-slot relationships (connecting particular lexemes or phrases with particular slots of constructional schemas).

The grammar network is shaped by domain-general cognitive processes (Ch. 3). The most important ones are social cognition, conceptualization, and memory, which compete in determining linguistic decisions—that is, the choice between different ways of conveying one's communicative intention for the speaker and between different interpretations for the hearer. There is evidence, for instance, that 'speaker-oriented processes', such as memory retrieval, priming, and automatization, can dominate 'hearer-oriented processes' of common ground and audience design. The domain-general processes leave long-term effects on language structure in diachronic change, and on individual language development in ontogeny. At the same time, D emphasizes that L1 acquisition and diachronic change differ substantially. In particular, children extract novel schemas from the input, while diachronic change involves the modification of already existing ones.

In the second part of the book, D focuses on the interpretation of signs as networks, beginning with taxonomic links between representations at...