Igor A. Mel'čuk's Semantics: From meaning to text (henceforth SMT) is a work in Benjamins' Studies in Language Companion Series and an essential component in Mel'čuk's Meaning–Text Models (MTMs). Before SMT, books on various levels in MTMs, including syntax, morphology, and communicative organization have been published (Mel'čuk & Pertsov Reference Mel'čuk and Pertsov1987, Mel'čuk Reference Mel'čuk1988, Reference Mel'čuk2001, Reference Mel'čuk2006). The long time coming of SMT manifests both its core position in MTMs and the difficulty in semantic research. Mel'čuk based the writing on his life-long study and answers in a unique way the fundamental question: How is meaning expressed in a language? In the opinion of the present reviewers, SMT presents an innovative and novel approach to the field, and is a paradigm-shifting contribution to the language sciences.

SMT is presented in five parts over three volumes. The present volume, or the first two parts, is composed of six chapters, which contain the problems of Semantic Representation (SemR), including the semantic structure of utterances, the semantics of causation in English, and communicative structure.

The general introduction provides an informal characterization of meaning–text semantics with its brief history, indicating its different focus of attention from other semantic approaches (e.g. Lappin Reference Lappin2001, Cruse Reference Cruse2004, Davis & Gillon Reference Davis and Gillon2004) and its nature of being purely linguistic, synchronic and static. Although psychological and neurological aspects of language are quite important, SMT leaves them out and concentrates on correspondence between linguistic meanings and the linguistic expressions that carry them, i.e. mapping from meaning to text, hence the names Meaning–Text Models and Meaning–Text Theory (MTT). While an MTM is a functional linguistic model or, in other words, a finite set of formal rules, MTT is a theory for building MTMs of actual languages. Mel'čuk distinguishes linguistic semantics as meaning–text oriented, cognitive semantics as meaning–reference oriented, and logical semantics as meaning-centered. SMT considers exclusively the first, namely the {Meanings} ⇔ {Texts} correspondence, where the braces indicate sets. According to Mel'čuk, the very essence of natural language is formalizing the transition between linguistic meanings and linguistic texts. Thus, MTT is synthesis-oriented rather than analysis-oriented, which makes it quite unique.

The first three chapters of SMT (Part I, ‘Meaning–Text Approach and Meaning–Text Models’) sketch out a general theoretical framework of MTT. Chapter 1 (21–44) is an inventory of basic linguistic notions used in SMT, like an ‘utterance (U)’ as a speech segment or a linguistic expression, the ‘linguistic meaning (here in this book, meaning)’ of U as the only invariant property of all possible paraphrases, a ‘grammeme’ as an inflectional value, and a ‘semanteme’ as the signified of a full lexical unit of language. It also introduces two major types of rules. Correspondence rules take the general form X ← corr → Y|C, where X and Y are configurations of linguistic units, ← corr → stands for correspondences like ⇔ (between two adjacent levels of representation), ≡ (between units in the same level), → (X to be replaced by Y of the same level), and ⇔ (X, Y being semantically equivalent). A filter rule *XY|C indicates that XY is ill-formed when condition (C) holds.

Chapter 2 (45–83) is dedicated to linguistic paraphrase, an essential notion denoting the relation of synonymy between sentences. It deserves the elaboration as the whole formal system of the MTT framework is based on the assumption that an important part of the linguistic competence of native speakers is to judge whether two given sentences are paraphrases.

Chapter 3 (84–159), entitled ‘MMT and MTMs’, starts with the basic tenet of MTT, admitting that strictly speaking, MTT is in fact rather a methodology or an approach than a theory. Then it states the three postulates formalizing the basic tenet and describing the MTM: language as a many-to-many correspondence between meanings and texts, linguistic description as a functional model and multiple levels of linguistic representation. In sequence, these levels go from SemR through the Syntactic Representation (SyntR) to the Morphological Representation (MorphR) and, ultimately, Phonetic Representation (PhonR). SemR, SyntR and MorphR are further divided into their surface (S) and deep (D) levels, hence the seven levels in the model, which are in turn connected by six modules, or sets of rules, namely semantics, deep syntax, surface syntax, deep morphology, surface morphology and deep phonology. Each level of representation is composed of one central structure and several peripheral structures, e.g. SemR=〈SemS; Sem-Comm(unicative)S; Rhet(orical)S; Ref(erential)S〉. The main properties of MTT are: its non-generative character (it is equative, or translative), the exclusion of linear order as a formal means of expression of its statements (all relations are denoted by appropriate symbols), and especially its meaning-to-text orientation (it is synthetic rather than analytical). Various representations of linguistic utterances are written each in their own artificial formal language. The semantemes are language-specific but the formalized semantic networks operate with the same syntax across languages.

If the first three chapters serve as the necessary background, laying the foundation of the basic linguistic notions on which SMT relies across all three volumes, then the ensuing three chapters are devoted to SemR, the starting point of linguistic synthesis, or the input to a MTM.

Three further chapters constitute Part II (‘Semantic representation in a meaning–text linguistic model’). Chapter 4 (167–241) concentrates on the notion and requirements of SemR, presenting a feature of semantics in SMT as distinct from other semantic approaches in that it is at the core of a complete theory MTT, and for a formalizable model MTM. The representation SemR=〈SemS; Sem-Comm(unicative)S; Rhet(orical)S; Ref(erential)S〉 indicates that SemR is composed of four structures, with SemS expressing the propositional meaning of U, Sem-CommS(U) expressing its communicative organization, RhetS(U) presenting its ‘artistic’ or expressive effects, and the RefS(U) linking semantemes in the SemS(U) to their referents. Structured in the same way in any language, the SemS is cross-linguistically universal. Its characterization is networks whose labels on the nodes are semantemes and whose arcs are numbers. Semantemes are classified into predicates and (semantic) names, which Mel'čuk calls two semantic ‘parts of speech’. A predicate has argument slots, which is the combinatorial property of genuine predicates. All semantic (and syntactic and morphological) links between the elements of utterances are expressed in terms of dependency relations. Using numeral labels as mere distinguishers rather than semantic roles constitutes a major difference from other semantic approaches.

Chapter 5 (243–288) is about semantemes of causation, with ‘cause₁’ (=‘be the cause of’) referring to non-volitional causation, and ‘cause₂’ (=‘be the causer of’) referring to volitional or agentive causation. Employed in a great proportion in semantic decompositions, causation maintains its crucial place in semantic description.

Chapter 6 (289–394) focuses on Sem-CommS, the first peripheral structure of SemR, but its significance is by no means peripheral. Being the ‘framing’ or ‘packaging’ of meaning, it concerns communicative organization of the meaning expressed by U, rather than the communicative organization of U itself. The latter is specified in the MTT framework by DSynt-CommS. This novelty of MTT sets it apart from other approaches. Unlike the SemS, which is conveniently and naturally represented by using genuine linguistic units, Sem-CommS's formal description must be based on an artificial metalanguage, which makes use of eight oppositions, namely Thematicity, Givenness, Focalization, Perspective, Emphasis, Assertivity, Unitariness, and Locutionality, each with two or three values. The most universal one, Thematicity, has three semantic values, Thematicity={Theme, Rheme, Specifier}, and can be illustrated by the sentence ‘[In 2005]_Sem-Spec [this student]_Sem-T [passed his final exam]_Sem-R’.

In volume 2 of SMT (Studies in Language Companion Series 135, published too late for inclusion in this review), the author introduces Deep-Syntactic Representation (DSyntR), or the endpoint of semantic operations (Part III), and describes the semantic module itself by detailing the major types of semantic rules that perform the {SemRi} ⇔ {DSyntRj} transition, i.e. semantic and deep-syntactic paraphrasing (Part IV). We look forward to the publication of volume 3 of SMT, which – in the final Part V, entitled ‘Linguistic excursuses’ – is to address several important semantic phenomena and discuss in depth some corresponding concepts including actants, Government Patterns, and phraseology (xxi).

As SMT came out after three decades since its conception, we can expect the whole of this multi-volume ‘crowning achievement of a long and illustrious career’ (xiii) to be insightful and well presented, but not necessarily abstruse. Judging by our reading of the first volume of the work, the vast and deep topic is in fact very clearly structured and presented, with the contents and structures for each part and each chapter clearly indicated at the outset. All the key ideas are highlighted and each key concept is well defined, exemplified and illustrated. These make it easy for readers to closely follow the logical thread. With some very basic ideas of semantics, one can start reading, move quickly and happily with the contents, and really enjoy the novel ideas presented in the highly readable text. Sharply dissimilar from other semantic approaches, this theoretically-complete formalizable model of human language is unique in many ways.

In providing an overview of MTT, we note first that it has deep roots in many of the schools of early 20th-century linguistics, such as transformational grammar, generative semantics, stratificational linguistics, case grammar and rational grammar, whose influences are evident in MTT. Secondly, as a theory of the fundamental question of the nature of language, MTT is a unified framework, whose models operate on the idea that language is a mapping (correspondence) from the meaning (semantics) of U to its form or text (phonetics), with intermediate levels of syntax and morphology. Thus, MTT has a stratificational and modular organization, providing a coherent theoretical framework for semantics, and making it an essential part of an integrated whole instead of being a stand-alone, distinct component of grammar. In most other semantic approaches, semantics is self-contained and is often independently studied with no connection with other levels. Thirdly, relations (rather than classes) are the main organizing factor in language and the concept of dependency (rather than constituency) is used extensively in MTT. The representations at the different levels are non-linearized, like the unordered network of SemR(U) indicating relations between the semantemes. In the international context, MTT has established a very successful model in studying dependency grammar. For instance, a SyntR of U is a dependency tree-structure. And then the relationships between representations are translations or mappings, rather than transformations. Sets of rules, or components ensure transitions between levels. While many of the current trends treat various levels of language in a more heterogeneous way, MTT uses different formal languages for different levels of representation. The fourth notable feature of MTT is that it attaches much importance to communication, as can be seen from the first peripheral structures for SemR, DSyntR, SSyntR, which are all communicative structures.

Finally and most importantly, because of its formal and theoretical character, MTT serves as a bridge between theoretical and applied linguistics. MTT actually started from early synthetically-oriented practice of machine translation, as Mel'čuk maintains his basic tenet that everything starts from meaning. MTT's current applications in machine translation, phraseology and lexicography are good cases in point, proving that the theory lends itself readily to applied linguistics. The key position of the lexicon in MTT is particularly worth mentioning: the lexicon is represented in the form of an Explanatory Combinatorial Dictionary (ECD) (Mel'čuk & Žolkovsky Reference Mel'čuk and Žolkovskij1984), a catalogue of the lexical units of the language and conditions governing their use. An ECD has its roots in dependency grammar and the ECDs for various languages have been a successful marriage between a modern linguistic theory and lexicography. Emphasizing lexicon and lexicography has not been practiced in other theoretical frameworks until recently, while since its very beginnings half a century ago, MTT has held its interest in linguistic synthesis (rather than analysis), paraphrase and the lexicon.

Mel'čuk is both creative and not afraid to take a non-mainstream stance. Though MTT remains relatively obscure to many researches for various reasons, we have good reasons to foresee a better understanding from a wider audience. A tremendous contribution in the field, this book will in due course prove itself to be well worth reading and will yield more significant and far-reaching theoretical and empirical results. Those unaccustomed to formalism will find it a great illustration of accurate description of a wide range of linguistic phenomena. And the formal schools will find in SMT an utterly novel design since current semantic theories are centered on truth-conditional logic. Semantics: From meaning to text is thus strongly recommendable to those interested in natural language processing, machine translation, lexicography, lexicon, semiotics, anthropology or semantics per se. Open-minded scholars in both theoretical and applied linguistics can cross the boundary and listen to Mel'čuk's unique voice.