2.1 Key notions

The first chapter, ‘Introduction’, by the editors of the volume (Kibort & Corbett)Footnote ^[2] provides a comprehensive overview of what is to follow and acknowledges a long tradition of language research and description which has appealed to features. Mention is made in the volume of the work of Saussure (Dresher: 37), the Prague School (where features were particularly prominent in describing the phoneme), followed by lexical semantics, Chomsky’s revolution in syntax, as well as morphological theory, down to formal models such as Generalised Phrase Structure Grammar (GPSG) and later Head-driven Phrase Structure Grammar (HPSG), which notably offer a coherent theory of features (see also Corbett Reference Corbett2012: 11–12, with references). Features are justified in that they enable linguists to capture generalisations by placing items in natural classes that share certain information or important similarities – English plural nouns, for instance, are characterised by the fact that they are compatible with the determiner these (Corbett: 17; Corbett Reference Corbett2012: xiii, 1ff.; Kibort: 64; Kibort & Corbett: 1). Making generalisations without features would be an extremely burdensome task, if not outright impossible (though see Kapatsinski Reference Kapatsinski2011 for an alternative). Despite potentially having different exponents, features are claimed to show ‘consistency across items, and to some extent across languages’ (Kibort & Corbett: 1). English plurals can be marked by -$s$, -en, vowel mutation, or zero, but they all behave in the same way with respect to plurality, regardless of their morphology. Moreover, English plurals are generally thought to be comparable to functional equivalents in other languages (Corbett Reference Corbett2012: xiii, 1–2). I will return to this non-trivial issue in the discussion below, where I will also present some dissenting views, most notably Haspelmath (Reference Haspelmath2010a, Reference Haspelmathb, Reference Haspelmath, Graf, Paperno, Szabolcsi and Tellings2012).

The book brings together scholars of various persuasions, which highlights the fact that, although features tend to be a currency that virtually all linguists share, they are often invested with different content or put to different uses, depending on who is using them (Kibort & Corbett: 2; see also Corbett Reference Corbett2012: 1).Footnote ^[3] This brings to the fore the importance of building a more or less generally accepted feature infrastructure. In his contribution, Corbett (34–36) outlines the first practical steps for this, noting some theoretical implications. Kibort is also especially concerned with this issue.Footnote ^[4]

Several threads run through the book, reflecting the situation in linguistics more broadly – the problem of identification (How do we establish the features and the values that make them up?), the meaning and function of features, their status, their structure and interactions, and the domains where they operate (Kibort & Corbett: 2–3). An important contrast is drawn between inherent and contextual features – a plural noun denotes a plurality of referents, and its number is therefore an inherent property, but a plural verb that agrees with its subject does not necessarily denote a plurality of events (see Corbett: 25; Corbett Reference Corbett2012: 66–68, with references, as well as the discussion of Kibort’s chapter below; see also Sag: 258). That is reminiscent of the distinction between interpretable and uninterpretable features in Minimalism (Chomsky Reference Chomsky1995; Adger: 187ff.; Corbett: 26; Kibort & Corbett: 3). It was noted above that researchers might invest the same term with different meanings, which is an unfortunate fact about our field. We now see another problem of a similar kind – linguists tend to come up with different terms for what is essentially the same concept (see esp. Kibort: 66–67). An apt illustration can be found in Table 4.1 in Kibort (66), where the synonymous terms ‘feature’, ‘Kategoriengefüge’ [complex of categories], and ‘(inflectional) category’, are listed. There are even more alternatives for the concept of ‘value’ (see Adger: 197, on similar issues within Minimalism). Corbett’s and Kibort’s contributions highlight both those setbacks in a laudable effort to introduce more uniformity across the field. An attempt is made in the volume to combine formal approaches to features with their diverse potential typological manifestations across various languages.

Corbett’s chapter, ‘Features: Essential notions’, looks at how features are integrated in grammatical description. He advocates an approach with sharp distinctions between semantics, morphology, and syntax; in his view, those should be strictly separate modules supplied with appropriate interfaces (see Corbett Reference Corbett2012: Chapter 3). Syntax is assumed to be phonology- and morphology-free, so syntactic rules cannot refer to purely phonological or morphological rules (see Corbett: 18; Corbett Reference Corbett2012: 42ff., with references, and the discussion of Popova’s chapter and Section 3.2 below; see also Sag: 259–260). In other words, purely morphological, phonological, semantic or morphosemantic information should not be visible to syntax. However, it is unclear how exactly morphosyntactic features will fit in – it is acknowledged that such features ‘have an effect across component boundaries’ (Corbett: 18; see also Corbett Reference Corbett2012: 48ff.). As Kibort (67) notes,

As regards the division between semantics and morphosyntax, Corbett (also Corbett Reference Corbett2012: 48) notes that the morphosyntactic genders in many languages correspond in part to semantic distinctions, though the correspondences are only partial. It remains to be worked out how to integrate and appropriately interface semantic features such as animacy, which can be proven to be different from morphosyntactic animacy, and yet influence morphosyntactic patterns. I return to this in Section 3.2.1 below.

One of the major achievements of Corbett’s chapter is the work the author has done on the internal structure of features (see Corbett Reference Corbett2012: Chapter 2). Corbett (20–23) convincingly demonstrates that features can indeed have a hierarchical internal structure similar to the structure we find in many other components of language; however, this need not be the case and a linguist should argue for, rather than assume, the analysis he or she adopts (see Danon – an illustration of how this can be implemented). A three-person system can just be made up of a list of values (first, second, and third), or it can be decomposed into [±participant] and [±speaker], thus producing a binary-branching hierarchy (Corbett: 19). Crucially, unstructured (flat) lists of values are acceptable in the model proposed by Corbett, which sets it apart from the views advocated in much previous work. Thus, neither layering nor binarity of feature values should be seen as universal (see Adger: 192ff.; Pullum & Tiede: 273ff.). It is additionally shown how syncretism can be accounted for by appealing to a flat list of values in combination with default settings. Languages could also operate with facultative values, i.e. values which need not be selected. For example, there are languages with dual number which allow the use of the plural instead of the dual with reference to two items (see Corbett: 20; also Kapatsinski Reference Kapatsinski2011). The latter situation would justify positing internal structure. Either type of analysis is fine for Corbett, as long as it is motivated.

Although his focus is primarily typological, Corbett (27–31; Corbett Reference Corbett2012: 39–40) outlines the problems agreement phenomena could pose for various syntactic theories. Such problems may come from hybrid controllers like English committee or from conjoined phrases. Both these types of constituent admit variable agreement which is not always easy to capture formally and which seems to be constrained by Corbett’s Agreement Hierarchy (Corbett Reference Corbett1979 and elsewhere).Footnote ^[5]

Attempts to deal with variable agreement in a constraint-based setting can be found in Dalrymple & Hristov (Reference Dalrymple, Hristov, Butt and Holloway King2010) and Hristov (Reference Hristov2012, Reference Hristov, Butt and Holloway King2013b, with references). Kibort (Section 4.3.3.1) takes up the dichotomy between syntactic and semantic agreement too – e.g. the committee has/havedecided. At first glance, she seems to conveniently gloss over an important problem: Where exactly do the properties of these hybrids belong and how are they best classified and described? Will agreement involving a semantically justified feature value make that feature syntactico-semantic, semantico-syntactic or morpho-syntactico-semantic? In the end, Kibort does offer a reasonable solution. An inherent value, for example the number or gender of a noun, can be assigned on the basis of semantic or formal criteria, or both, in the case of so-called hybrid nouns like committee. In turn, the feature values thus assigned may be visible to agreement targets and will therefore qualify as contextual values on agreeing verbs or adjectives. Therefore, this will still count as a morphosyntactic agreement phenomenon, albeit less canonical if it appeals to the more semantically justified value (Kibort: 79). This, I believe, moves the field forward quite perceptibly.

The lucid style and cogent explanations are hallmarks of Corbett’s exposition throughout. Even the discussion of some of the relatively complex cases is quite accessible and impeccably argued, making the chapter a pleasure to read. Corbett often revisits material from his earlier work, though now it is sometimes presented in a slightly different light. Probably due to limitations of space, some issues are only skated over in passing and the reader is left to chase up the references provided in the body of the text. The project under discussion is the culmination of Corbett’s long involvement with gender (Corbett Reference Corbett1991), number (Corbett Reference Corbett2000), and agreement (Corbett Reference Corbett2006), which originally started with typological outlines of Slavonic (Corbett Reference Corbett1979, Reference Corbett1983). Along with Blake (Reference Blake1994), those are highly respected field-defining texts. This volume also paves the way for Corbett’s (Reference Corbett2012) book entirely dedicated to features. The importance and the merits of this immense body of work can hardly be overstated.

Dresher’s chapter, entitled ‘Feature hierarchies and contrast in phonology’, draws a distinction between contrastive and redundant phonological features (see Pullum & Tiede for the philosophical underpinnings of feature redundancy). Dresher argues that the traditional minimal pair in phonology, which involves a pair of fully specified segments, should be abandoned as inadequate and replaced by a contrastive hierarchy dividing the set of phonemes in a reductive manner by appealing to a hierarchically ordered set of features. Dresher acknowledges the work in the same spirit of his predecessors, most notably Trubetzkoy (Reference Trubetzkoy1939) and Jakobson & Halle (Reference Jakobson and Halle1956), as well as more recent work in descriptive linguistics and Optimality Theory (see Dresher: 48–49). His ideas have parallels in morphosyntax (see Corbett: Section 2.3.1; Kibort & Corbett: 5). The obvious similarities are between distinctive phonemes, which differentiate words, and inflectional contrasts, which signal different semantics and/or syntactic functions. The locus of the phonological contrast has now been shifted from the phonetic properties of segments to their functions. There is some rather intriguing discussion of the interaction between phonetics and phonology (see esp. Dresher: 53–55, 60–62), which has important theoretical implications. As with phonemes, it is proposed elsewhere in the volume that grammatical feature values should be distinguished on the basis of function rather than form (see Kibort; Kibort & Corbett: 5). What phonological and morphological segments do will then be deemed more important than what they actually look like. One of the undisputable merits of the volume is that, diverse as it is, it gets us a step closer to a unified approach to phonology and morphosyntax. Like Corbett’s, Dresher’s chapter makes explicit what lies behind central assumptions which are normally taken for granted but rarely justified. This is healthy for the field in general.

However, as with most analyses with a relatively high degree of abstraction, this one also raises some questions which are not at all easy to answer in a theory-neutral fashion. Dresher (51–52) has to assume that the vowel [u] in his example (16b) is derived from an underlying /ʊ/, which has phonetically merged with underlying /u/. This assumption is needed to maintain vowel harmony in Classical Manchu. It would be nice, however, to see more independent non-circular evidence supporting this analysis, rather than simply conceding that the examples in (16b) do not contain vowel harmony. Without additional background information, the justification currently seems to be little more than a prop for an elegant model. It is suggested at the end of the chapter that this is further explored in Dresher (Reference Dresher2009), where interested readers can find out more about the topics touched upon here.Footnote ^[6]Footnote ^[7]

Kibort’s contribution, ‘Towards a typology of grammatical features’, works out a typology informed by the descriptions of a representative sample of languages. In the introductory sections, she considers data which are mostly well known in the literature, such as the problematic gender systems of Polish and Romanian, and adjectival declension in German. Along with the other contributors, the author upholds excellent standards of terminological precision (see esp. Kibort: Section 4.2). Kibort is very careful about her use of terminology and frequently makes explicit the assumptions behind the terms she does or does not adopt, as well as the reasons for (not) adopting them. As mentioned above, an important distinction is recognised between contextual features required by syntactic rules, e.g. number on verbs agreeing with their subjects, and inherent ones, which are independent of the syntactic context, though they might have consequences for syntax, e.g. noun number or verb tense (see Kibort: Section 4.3.1). For Kibort, the ultimate criterion for establishing the existence of a feature is the presence of morphological exponence. Here form is given priority over function. This is in stark contrast to the insistence on function we found with Dresher.

Corbett and Kibort concur that a feature’s relevance to syntax is determined by its participation in agreement and/or government. Such a feature will qualify as morphosyntactic. This is contrasted with a morphosemantic feature, which will have no relevance for syntax; morphosemantic feature values can only be inherent and are not involved in agreement or government. Verbal tense is often taken to illustrate this category (though see below). In turn, a purely morphological feature will be irrelevant to both syntax and semantics – such values can only be inherent, cannot participate in agreement or government, and cannot be selected to indicate function or meaning contrast. A typical example is declension or conjugation class. The conclusion following a complex case study is that number, gender, person, respect, case, and definiteness do participate in syntactic agreement and/or government in the languages of the world, and hence qualify as morphosyntactic features, though tense, aspect, mood, and polarity (among others) appear to be restricted to the interface of morphology and semantics due to lacking the above defining properties.Footnote ^[8] These issues will be revisited in Section 3 below, where I argue that there can be agreement in tense. Likewise, at first glance, the subjunctive mood in many languages appears to be governed, selected or at least licensed by certain verbs but not others, where the verbs governing the subjunctive do not possess the subjunctive value themselves (this is briefly touched upon in Sag: 241, with references). Therefore, I think more work is needed to ascertain if this is merely a morphosemantic phenomenon which is due to meaning alone.

Kibort (77) asserts that English singularia tantum nouns such as happiness and poverty are assigned their inherent number on the basis of semantics, unlike pluralia tantum ones such as scissors and dregs, whose fixed inherent number value is based on form. It is often noted in the literature, however, that pluralia tantum nouns like scissors and trousers have at least some semantic justification because they are made up of two parts (see Corbett Reference Corbett2000; Reference Corbett2012: 26; Payne & Huddleston Reference Payne and Huddleston2002: 340–342; see also Copestake & Flickinger: Section 9.4, 233, for a sophisticated discussion of the interpretation of number, including so-called pair or binary nouns). Finally, Kibort (82–83) points out that her definitions are concerned with canonical morphosyntactic, morphosemantic, and morphological features. Various deviations from canonicity are expected in this canonical approach. We return to it in Section 3.1 below.

2.2 Perspectives from syntactic description and theory

Part II of the volume explores how features fit in syntactic description and syntactic theory. Plaster & Polinsky’s chapter, ‘Features in categorization, or a new look at an old problem’, looks into how languages determine the values of the gender feature. The validity of relying on cross-linguistically comparable formal and semantic cues is confirmed, which the authors put down to limitations on language acquisition and the perceptual salience of central concepts such as male, female, animate or edible. In order to illustrate their point, they reassess the system of gender in Dyirbal, which was famously said to conceptually group together women, fire and dangerous things (Dixon Reference Dixon1972, Lakoff Reference Lakoff1987, and the references cited in the chapter). The authors’ reassessment, which invokes diachronic reanalysis that was sometimes based on formal analogy, removes the need for a synchronic conceptual association between disparate items which now happen to belong in the same gender class. As a result, Dyirbal no longer seems as exotic as it was previously made out to be – it turns out that Dyirbal too relies on a combination of relatively straightforward formal and semantic factors.

Like earlier chapters, this one also moves the debate forward by reappraising assumptions that have held sway for decades. We once again see terminological precision (Plaster & Polinsky: 109 fn. 2, 123), as well as an attempt to draw parallels between different components, such as phonological and gender systems (Plaster & Polinsky: 109). Like other chapters, this one too is concerned with cross-linguistic generalisations, more specifically regarding ‘the content of gender features possible in human language’ (Plaster & Polinsky: 109; see Corbett Reference Corbett2012: 111–119, with references). Though I am by no means an expert on Dyirbal, I cannot help but praise the fact that the diachronic dimension has been brought in. In spite of the mixed feelings many eminent 20th-century linguists had about language history and its place in linguistic explanation, the picture without the diachronic dimension in it cannot be truly complete (see Haspelmath Reference Haspelmath, Carnie, Sato and Siddiqi2014 on diachronic explanations in typology). This chapter beautifully demonstrates that it is often impossible to make sense of synchrony without appealing to diachrony, truistic as this may sound (see Plaster & Polinsky: 125). In line with prevalent assumptions about the origin of a great deal of language change (and hence structure), the centrality of young learners acquiring their native language is also duly acknowledged (see Ringe & Eska Reference Ringe and Eska2013, among others, as well as Chomsky’s generative tradition).

However, a diachronic account without much corroborating historical evidence is of necessity highly speculative (see Plaster & Polinsky: Sections 5.5.4, 5.5.5). It is always a danger to invoke analogy as a last resort, the way the Neogrammarians did when the results of regular pre-historic sound changes they proposed didn’t turn out as expected. When formal parallels are invoked, it is naturally imperative to find sources for any putative analogy, and the authors do try to identify exactly such instances of formal similarity (see esp. Plaster & Polinsky: 118–121, 128ff.). They mostly look quite plausible, though analogy, notoriously irregular by nature, is sometimes a matter of rather subjective interpretation and point of view (see Plaster & Polinsky: 121 fn. 9, on alligators, as well as Plaster & Polinsky: 129ff., 134, for other problematic cases and exceptions, including frequency effects on analogical change).Footnote ^[9] Nor is synchrony neglected at the expense of diachrony (see Plaster & Polinsky: Section 5.5), and it is nice to see the plausibility of the proposed system confirmed by evidence from loanword adaptation (Plaster & Polinsky: 124). The situation reminds one of the celebrated debunking of the myth surrounding Eskimo snow terminology (Pullum Reference Pullum1989).

Danon’s chapter, ‘The definiteness feature at the syntax-semantics interface’, uses material from Hebrew. It looks into complex genitive constructions where the definiteness value is said to be shared or spread. Danon (143) argues for analysing this spreading as a syntactic phenomenon in Hebrew, which works by sharing a morphosyntactic feature of definiteness, whereas in other languages it is just a byproduct of the semantic interpretation of noun phrases, without any reference to a grammatical definiteness feature. Danon demonstrates (Section 6.2) that definiteness does participate in agreement (or government) in Hebrew, in which case its occurrence can be semantically vacuous on the target (see Danon: 144–145; Corbett Reference Corbett2012: 134–137, with references; see also the discussion of English backshift in Section 3.2.2 below). Since the phenomenon of spreading depends on syntactic structure, Danon claims that definiteness should be treated as a morphosyntactic feature, highlighting mismatches between morphosyntactic marking and semantically definite interpretations. For instance, noun phrases which are semantically definite do not have to be marked for the morphosyntactic definiteness feature (see Danon: 146–147, for examples and discussion). Conversely, constructions whose embedded nominal is marked as morphosyntactically definite are not necessarily interpreted as semantically definite (see Danon: 151–152, 152ff., for all the possible permutations). This nicely fits in with the criteria and the methodology developed by Corbett and Kibort in earlier chapters, since an important distinction is drawn between morphosyntactic definiteness and its semantic counterpart.

However, Danon (149) disagrees with Kibort on the morphosyntactic status of tense and aspect. He notes that, in many languages, features such as number, person and gender are all involved in overt agreement. He then says:

Without assuming any abstract agreement whatsoever, I will demonstrate in Section 3.2.2 below that English does exhibit overt agreement in tense.

A few passing thoughts are perhaps in order regarding the discussion in Danon on pages 157–158. Although [[a man’s]_Det:NP dog]_NP does not allow definite determiners, it should be pointed out that it will not allow any determiner at all, since there is one already – the genitive noun phrase. Contra Siloni (Reference Siloni and Lecarme2003), Alexiadou (Reference Alexiadou2005) and Danon, and siding somewhat with Woisetschlaeger (Reference Woisetschlaeger1983), I would like to stress that a man’s dog is at least paraphrasable with the definite article, i.e. ‘the dog of a man’, in addition to having an indefinite interpretation: ‘a dog of a man’. Similarly, a king’s sons can conceivably stand for ‘the sons of a king’ and probably even ‘(some) sons of a king’, where the latter is not definite and does not have to refer to the entire set of male siblings (see Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 265ff.; Danon: 157–158). Barring these minor objections, Danon’s arguments seem sound and the whole analysis looks plausible.

Popova’s chapter, entitled ‘Features in periphrastic constructions’, follows up claims in the literature that the morphosyntactic features of periphrastic constructions cannot be easily derived from the properties of their component parts. The English perfect, for example, comes about as a combination of the auxiliary have and a past participle, but neither of those components can express the perfect on its own (Spencer Reference Spencer, Junghanns and Szucsich2003; Kibort & Corbett: 8). Popova’s contribution explores the interaction of features in such periphrastic environments. The focus is on meaning mismatches in the clausal domain, in particular the values associated with ‘analytic tense forms’. Popova sets out to demonstrate that the interactions of features in periphrastic constructions are more complex than has hitherto been assumed. She makes a case for a top–down approach, whereby a periphrastic construction is associated with morphosyntactic features directly as a multi-part unit (Popova: 167). Given the alternative, bottom–up view (which is ultimately rejected), the features associated with the construction as a whole should come from some or all of the elements making up the bigger unit (see Popova: 172–173). This chapter aims to shed light on how language creates meaning by relying on a division of labour between syntax and morphology, again taking up and developing a thread that runs through the book.

Popova treats periphrases as forms of a lexeme, a view which has a long history in traditional grammatical description and which will be challenged in Section 3.1.1 below. In order to capture this formally, Popova adopts a realisational approach to morphology; in this approach, lexemes are associated with the features appropriate for them independently of the manner in which their word forms are constructed (Popova: 169, with references). Crucially, the realisational framework adopted by the author allows word forms to be made up of synthetic or analytic constructions.

Finally, the chapter contains some controversial assertions (though they are largely orthogonal to the main line of the argument). Bulgarian verbs are said to inflect for aspect, and nouns for gender (Popova: 167–168). It is a moot point whether morphologically marked aspectual pairs are instances of inflection or derivation (see Nicolova Reference Nicolova2008: 247ff.; Corbett Reference Corbett2012: 71). Nouns, on the other hand, normally have a lexically specified, inherent gender value and they certainly do not inflect for gender in the same way that adjectives do, or in the same way that nouns themselves inflect for number (see Corbett Reference Corbett2012: 67 fn. 23, and the discussion of inherent and contextual features above). In other words, an adjective can be masculine, feminine or neuter depending on the syntactic environment, but a noun is exclusively and fairly immutably assigned to one gender class only. Bulgarian masa ‘table(f)’ can hardly be masculine or feminine in the same way that it can be singular or plural.Footnote ^[10]

Adger’s chapter sketches out ‘A Minimalist theory of feature structure’. The author (185) is interested in the architecture of features and the syntax of how features combine to build lexical items and larger structures. In a typically Minimalist fashion, all of structure embedding is said to stem from a single operation – that of Merge (though Adger: 199, 201, 204 does talk about variants of Merge). This is a sweeping generalisation which entails that structure embedding in human language can only be syntactic. As a JL referee suspects, the idea is that there can only be an X inside an X in the syntax, and hence there may be no features inside the values of other features in the lexicon (HPSG-style). I am not at all certain whether this will stand up to close scrutiny – it would therefore be good to see some more empirical argumentation offered in its favour in order to convince the agnostic reader.

This chapter is a textbook illustration of the differences between transformational and lexicalist approaches to syntax (the latter are also represented elsewhere in the volume).Footnote ^[11] In the former, espoused by Adger, the syntactic component is all-powerful and this is where most of the relevant information resides, including complexity of feature structures. By contrast, lexicalist approaches like Lexical Functional Grammar (LFG) and HPSG will have a much richer lexicon with rich featural structures for lexemes. In the spirit of Minimalism, Adger sees the grammar as a lexicon of items, combined with a set of operations on lexical items (and on outputs of operations); the operations are themselves constrained by principles of admissibility and triggered by the need to check or provide a value for valueless features. The structures built by the syntax are subsequently interpreted by the modules that map such structures to sounds and meanings.

A problem for feature checking or valuing operations comes from phenomena like pro-drop, where there is no controller against which to check (or value) the features of the target (see Adger: 189, 192, 207; Corbett Reference Corbett2012: 37; Hristov Reference Hristov2012: 24–25, with references).Footnote ^[12] Subsuming agreement and (case-)government under the same Agree operation obscures important differences between them, outlined earlier by Kibort. Crucially, English verbal governors, for instance, never have overt case, the way agreeing subjects and verbs both show number and person (see Adger: 210–211; see also Sag: 258, for a constraint-based take on this). In general, Adger’s (198ff.) notation can at first be a little intimidating for anyone without a solid formal background, which might make it difficult for such a reader to effortlessly follow the points the author is making (see Kapatsinski Reference Kapatsinski2011).Footnote ^[13] The elaborate hierarchical structures intended to cope with the order of the many men/*many the men (Adger: 199–200) would have to be adjusted or enriched in order to account for many a man, for instance, though the latter could potentially be discarded as an exceptional idiom. There are also strings like as much a man, as good a man, how good a man, so good a man, although a minimalist would presumably see these as a product of movement (Bob Borsley, p.c.). In either case, observations following a thorough examination of natural language material should be the decisive factor in determining the theoretical model of choice (see Adger: Section 8.4 for some intriguing, albeit empirically questionable, discussion of data contrasting Minimalism to lexicalist approaches).Footnote ^[14] It would therefore be nice to see how Adger’s sophisticated system fares with a wider range of data, though there is obviously limited space in a chapter.

2.3 Formal perspectives

Following Adger’s Minimalist chapter, the rest of the volume is dedicated to formal perspectives on features. Copestake & Flickinger’s contribution deals with ‘Features and computational semantics’. This chapter elaborates on how the findings of linguistics can be implemented in other spheres of human activity – semantic representations are used in computer applications and text processing, including information retrieval and extraction, and machine translation and summarisation, for instance. Copestake & Flickinger address the issues faced by the developers of such programs when it comes to encoding semantic information. These issues include the representation of features that come from the morphology and the syntax without much access to extensive underlying background knowledge or large enough lexicons containing detailed lexical information. Additional matters include the need to make provisions for anaphora and discourse structure, the need to be able to disambiguate and underspecify, and the need to achieve expressive adequacy and compatibility with grammatical information, among others.

This model is implemented with HPSG-style feature structures. Unlike what we saw in the preceding chapter, the HPSG approach relies heavily on the lexicon, demonstrating the diversity of theoretical affiliations showcased in the volume, while still keeping the focus on the interplay between syntax, morphology and semantics. The authors also employ Minimal Recursion Semantics (Copestake et al. Reference Copestake, Flickinger, Sag and Pollard2005), coupled with an augmented version of standard predicate logic.

Although, as noted above, HPSG relies heavily on the lexicon, in a sense, this is arguably even more true of Minimalism. Bob Borsley (p.c.) points out that ‘in HPSG lexical elements interact with complex hierarchies of phrase types to determine the grammatical expressions of the language’. In Minimalism, by contrast, we get only a handful of very general mechanisms with which lexical items interact (Merge, Agree, Move/Copy); so whatever is possible arises mostly as a consequence of the properties of lexical items. (I am grateful to Bob Borsley for bringing this up.)

Sag’s chapter, ‘Feature geometry and predictions of locality’, explicitly tackles issues of locality of selection which have to do with idiomatic expressions, control of pronominals, and filler–gap dependencies. Locality of selection is defined as ‘the problem of delimiting what syntactic and semantic information lexical items select’ (Sag: 236). Another fundamental question concerns the locality of construction, i.e. how one can determine what syntactic and semantic information is accessible to grammar rules. Given that constructions are limited to mother–daughter relationships, non-local phenomena are amenable to an explanation which uses local encoding and the inheritance of grammatical information. It is the geometry of features that constrains the grammatical information available for lexical selection or syntactic structure building.

Sag’s proposal is developed in the framework of Sign-Based Construction Grammar, an offshoot of HPSG cross-pollinated with Construction Grammar (see Sag: Section 10.4). The awareness of these issues ultimately goes back to the Chomskyan tradition, as outlined in Sag (Section 10.2). Sag’s model is brought to bear on phenomena as diverse as non-local case assignment in English and in case-stacking languages, as well as various filler–gap dependencies and idiomatic expressions.Footnote ^[15] The coindexation of dummy subjects in English is of particular interest. Though this may seem obvious, it might still be worth mentioning that no coindexation will be needed for subjects like those in Itlooks like there’s going to be a storm/Kim’s going to win, which sets them apart from Therelooks like there’s going to be a storm/*Kim’s going to win (see example (39) in Sag: 266, and the constraint in (42) in Sag: 268). This raises the more fundamental issue of the possible abuse or misapplication of Sag’sxarg feature percolation mechanism, including questions of when and how it is motivated. Essentially, the xarg feature allows an argument of a head to be visible outside the maximal projection of that head (Bob Borsley, p.c.). The relevance of this to English grammar is most clearly seen in tag questions or in idioms like Helost his/*hertemper.

The final chapter of the volume, ‘Inessential features and expressive power of descriptive metalanguages’, by Pullum & Tiede, is based on model-theoretic syntax. It seeks to establish criteria for distinguishing genuine from spurious linguistic features. Another important matter brought up in the chapter is the amount of strength a metalanguage should have in order to be suitable for syntactic description. This is the most formal in a series of increasingly formal chapters, so it should come with a health warning for those not versed in formal logic. Model-theoretic syntax is a basic view of the nature of syntax assumed in a number of frameworks, e.g. LFG and HPSG, but not in Minimalism (Bob Borsley, p.c.). Structures in natural language are related to model-theoretic structures. This sort of grammar is then ‘a finite set of statements, formulated in a logical metalanguage adapted to the statement of descriptions of natural language structural properties’ (Pullum & Tiede: 276). Essentially, this approach recognises a structure as grammatical if and only if the structure satisfies (or is a model of) the statements in the grammatical metalanguage (see Kibort & Corbett: 11). The authors apply logics of varying strength to syntactic trees, and relate their model-theoretic approach to the generative paradigm via the notion of definability.

Pullum & Tiede point out that a feature can be deemed unnecessary if it can be eliminated without causing any loss of descriptive power. In purely structural terms, a feature is inessential ‘if its distribution is fixed by the other features’ (Pullum & Tiede: 289; quoting Kracht Reference Kracht and Retoré1997). Unsurprisingly, GPSG or X-bar features such as slash and bar turn out to be superfluous in this sense, but so does the familiar morphosyntactic feature of case, which comes as a bit of a shock (on GPSG slash and bar, see Adger: 195–196; Sag: 242–244; Corbett Reference Corbett2012: 45–47, with references).

Pullum & Tiede conclude that, at least for now, telling apart features as technical artefacts and features as genuinely motivated distinguishing properties of linguistic expressions cannot be reduced to any formal criterion (see Kibort & Corbett: 12). This has wider implications, meaning that it is worthwhile to capture, albeit informally, the distinctions intuitively relevant to natural language descriptions – a comforting conclusion, showing that typologists’ and descriptive linguists’ efforts are not in vain (see Kibort & Corbett: 12).

Such optimism, coupled with a cautionary note, is an appropriate way to close this volume, rich in enlightening theoretical perspectives and insightful analyses from several points of view. In many ways, it reads like a Modernist novel with multiple narrators retelling the same story in different styles and focusing on different aspects of it – as promised in the subtitle. In the discussion below, I return to some of the central recurrent themes.