Phi-features play a crucial role in syntax, semantics and morphology, and nowadays form the backbone of many grammatical theories. As obvious and crucial the role of phi-features may seem, the central role that phi-features currently occupy in linguistic theory has not always been that prominent. In fact, it is only a recent development that phi-features are taken to be at the core of grammar. Consequently, remarkably little work has appeared that presents both data and insights from morphological, semantic and syntactic studies of person, number and gender features. Daniel Harbour, David Adger & Susana Béjar's Phi theory fills this lacuna by giving an up-to-date overview of the state of the art on this topic. The book contains eleven chapters, written by well-established scholars in the field, providing a springboard for further inter- and intra-disciplinary research into what appears to be one of the few windows on how morphology, syntax, semantics, pragmatics and phonology interact.

In their introduction, ‘Why phi?’, the editors set the stage by acknowledging the recent prominence of phi-features and provide a brief history of the study of phi-features in syntax, semantics and morphology. They conclude by listing the most important themes within phi-theory, devoting a special section to one of the central topics in phi-theory, namely markedness.

The volume continues with two papers that deal with the semantics and pragmatics of phi-features, addressing the question of how pronouns and phi-agreement markers should be interpreted. In ‘Features on bound pronouns’, Irene Heim identifies a number of problems concerning the interpretation of phi-features, most notably on pronouns. She follows Cooper (Reference Cooper1983) and Heim & Kratzer (Reference Heim and Kratzer1998) by assigning phi-features a presuppositional semantics, and goes on to discuss several problems for the presuppositional treatment of phi-features as well as their potential solutions. One such problem concerns first person singular bound pronouns, as in (1).

1. (1) Only I did my homework.(43, ex. (24))

The sentence in (1) is ambiguous between a reading where nobody other than the speaker did the speaker's homework, and a more salient reading where nobody other than the speaker did their own homework. This ambiguity cannot be explained if my is construed as either free or bound.

This problem is not new and has been discussed by, for example, Partee (Reference Partee1989) and Kratzer (Reference Kratzer1998), who argue that my lacks semantic phi-features and behaves as a bound variable, with the additional property that it exhibits the morpho-phonology of a free pronoun. Heim argues that, once it is assumed that at the level of P(honological)F(orm) all features of a D(eterminer)P(hrase) must be copied onto all variables that it binds, problems concerning bound first person pronouns and others – for example, split-bound pronouns – can be overcome.

In ‘On the semantic markedness of phi-features’, Uli Sauerland builds on Cooper (Reference Cooper1983) and Heim's (1998) insights into the presuppositional nature of semantic phi-features and tries to derive markedness relations on the basis of phi-feature geometry: a feature F is marked with respect to feature G if the presupposition of F entails the presupposition of G, but not the other way round. Sauerland restricts himself to semantic markedness, distinguishing this type of markedness from morpho-syntactic markedness, which suggests that categories can be semantically marked but morpho-syntactically unmarked (cf. Sauerland, Anderssen & Yatsushiro Reference Sauerland, Anderssen, Yatsushiro, Kepser and Reis2005).

Sauerland presents four diagnostics for semantic markedness: dominance, quantification, epistemic status and emergence after blocking. Dominance guarantees that it is always unmarked features that determine verbal agreement in cases involving conjoined noun phrases: the unmarked status of masculine is the reason why the verb in the Czech sentence in (2) does not express feminine agreementFootnote ¹

1. (2) JanaVĕrašl-idobiografu.(Czech; 63, ex. (5))JanandVerawent-masc.pltothe.movies‘Jan and vera went to the movies.’

Unsurprisingly, Sauerland demonstrates that third person is less marked than second person, which, in turn, is less marked than first person. Surprisingly, though, he argues that plural is more marked than singular and, perhaps even more surprisingly, that dual is less marked than singular. This is due to the fact that a singular DP presupposes that the cardinality of its reference is one, whereas plurals do not presuppose that the DP denotes that its cardinality is more than one, as illustrated by the example in (3).

1. (3)

   • Q:Do you have children?

   • A:Yes, one.

Sauerland's paper contains a wealth of interesting ideas, although some of them need further elaboration. Moreover, his use of dominance as a diagnostic for semantic markedness is problematic, as it is far from clear as to why dominance does not reflect morpho-syntactic markedness.

Apart from the contributions by Heim and Jonathan David Bobaljik (see below), few chapters consider the relation between the syntax and the semantics of phi-features. Instead, the focus is largely on apparent mismatches between syntactic structures and morphological outputs. In a thorough and complex paper (‘Phi-agree and theta-related case’), Milan Řezáč addresses the notion of case opacity, i.e. the constraint that bans a DP with theta-related case from valuing a phi-probe. Řezáč argues that case opacity is due to the fact that theta-related case is expressed by prepositional phrases (PPs), which generally form an opaque domain (i.e. they constitute phases). An Agree relation in the Chomskyan sense (Chomsky Reference Chomsky and Kenstowicz2001) between probe α and DP β in (4) is thus forbidden.

1. (4)

   

However, an opaque domain can be ‘unlocked’ by specific properties of its head. In particular, Řezáč argues that if the phase-head P phi-agrees with a DP in its c-command domain, the entire PP is no longer opaque and the DP (inside the PP) may act as a goal that values a higher phi-probe (e.g. v selecting PP). This explains why in some Basque dialects the phi-features of dative experiencers are reflected in the verbal morphology (assuming a PP-internal Agree relation within the experiencer), as shown in (5).

1. (5) Geuriemong-a-it-u.(Basque; 87, ex. (5))us.datgiven1pl-thm.pl-√2v‘He has given it to us.’

In the remainder of the paper, Řezáč discusses the parametric variation that restricts the occurrence of this type of construction, and extends his analysis to cases of quirky case.

Susana Béjar's ‘Conditions on phi-Agree’ deals with partial agreement in domains where it is not obligatory. To illustrate this, she presents data from the Uralic language Erzya Mordvinian:

1. (6) Soda-s-am-iź.(Erzya Mordvinian; 131, ex. (1c))know-tns-1-plOA: ‘He knows us.’PA: ‘They know me.’

The sentence in (6) is structurally ambiguous between two readings: one where the verb exhibits object agreement (OA) only, and one where the verb agrees with the object in person and with the subject in number. This latter phenomenon is dubbed Partial Agreement (PA). Two questions arise: (i) why is PA allowed, given the notion of feature defectivity (Chomsky Reference Chomsky and Kenstowicz2001), and (ii) why does PA arise only in certain configurations?

Béjar argues that the Mordvinian data constitute a serious problem for the notion of feature defectivity in its present form and proposes an all-or-nothing condition on Agree, according to which a complete sub-phi set cannot Agree incompletely (where sub-phi sets are defined as sets of features where all features stand in an entailment relation). For instance, [plural] and [number] constitute one sub-phi set, as the presence of [plural] entails the presence of [number]. On the other hand, [speaker] does not entail [plural] or vice versa, so these features are in different sub-phi sets.

It then follows that Agree is restricted only when the probe contains more features in the same sub-phi set than the goal. If probe–goal agreement fails, the probe will have to seek a new goal (see Béjar & Řezáč 2009). This explains why OA sometimes fails and subject agreement acts as a rescue operation to value the probe. Moreover, Agree may be successful in one sub-phi domain (e.g. person agreement in (6)), but not in another domain (e.g. number agreement), resulting in PA. Béjar thus takes the position that the standard version of Agree is essentially correct but needs further refinement along the lines sketched.

In ‘Phi-feature competition in morphology and syntax’, Martha McGinnis also discusses cases of dependent agreement. Dependent agreement involves cases of agreement like (6) above, where verbal morphology is not necessarily restricted to one fixed syntactic category. Rather, it depends on the exact feature inventory of the object whether the verb manifests OA or subject agreement. McGinnis considers two other cases of dependent agreement, namely position-based and feature-based dependent agreement. Examples are given in (7) and (8), respectively.

1. (7)

   1. (a) t-aw-il-ah-en(Yucatec Mayan; 155, ex. (1a))pf.tr-2sgpart-see-cpl-1sg‘You (sg) saw me.’

   2. (b) hmeyahn-ah-en(155, ex. (1b))pf.intrwork.cpl-1sg‘I worked.’

1. (8)

   1. (a) g-biin-i(Ojibwa; 173, ex. (29a))‘You bring me.’

   2. (b) g-biin-ini(173, ex. (29b))‘I bring you.’

The difference between (7) and (8) is that, in (7), agreement shows a preference for the internal argument; only in the absence of an internal argument does the verb agree with the external argument. By contrast, in (8), the preference is based on a particular feature, namely [2SG]; regardless whether [2SG] is manifested on the object or the subject, the verb will always prefer [2SG] agreement over [1SG] agreement (which, in turn, is favoured over [3SG] agreement).

McGinnis provides a series of arguments for viewing position-based and feature-based dependent agreement as distinct phenomena. Position-based dependent agreement results from syntactic competition (Agree with the highest DP below v). Feature-based dependent agreement, on the other hand, is a morphological phenomenon, for which McGinnis formulates an analysis within Distributed Morphology (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993): if two different features occupy the same v node, the ranking of vocabulary items will determine which feature will be realized phonologically.

Unexpected agreement phenomena also feature in Daniel Harbour's paper, ‘Discontinuous agreement and the syntax–morphology interface’. Discontinuous agreement is attested in Georgian, where agreement with a first person plural pronoun results in the presence of two inflection morphemes on the verb, one reflecting person and the other reflecting number agreement, see (9).

1. (9) v-c'er-t(Georgian; 185, ex. (1a))1-write-pl‘We write.’

Moreover, Harbour notes that when a verb displays discontinuous agreement with two arguments, the linear order is always of the form person1–person2–number2–number1, as shown in (10) below.

1. (10) ta-pu-nan-ŋa-r-ŋkan-um(Yimas; 186, ex. (2a))neg-3-2pl-give-pauc.pl‘You few didn't give (it) to them.’

Harbour raises three questions: (i) why is agreement sometimes discontinuous; (ii) why does person always occur to the left of number in discontinuous agreement (an observation due to Trommer Reference Trommer, Booij and Marle2002); and (iii) why does multiple discontinuous agreement ‘flank’ (resulting in the ordering of person1–person2–number2–number1)? Harbour argues that discontinuous agreement facts can be accommodated by modifying Distributed Morphology in two ways: (i) by assigning internal structure to phi-features (person dominates number); and (ii) by assuming that vocabularization takes place cyclically (following Bobaljik Reference Bobaljik2000). If a particular phi-feature has two realizations for its parts, say, a particular person and number feature, principles of spell-out will require adjacency of the number feature to the previously formed verb. Since person dominates number, the person feature must be spelled out adjacently to the verbal stem as well, so that the only possible realization is person–verb–number. Applying this process cyclically derives the linear order person1–person2–number2–number1. In the remainder of this paper, Harbour applies this analysis to data from Yimas and Walmatjari. While Harbour's analysis appears to be simple, elegant and successful, the assumed internal feature hierarchy is not clearly motivated and does not comply with, for instance, what Béjar assumes to account for the Mordvinian data. Moreover, the paper relies on theoretical assumptions that are no longer uncontroversial, such as Kayne's (1994) Lexical Correspondence Axiom.

Jochen Trommer's paper, ‘Third person marking in Menominee’, presents highly interesting data from the Algonquian language Menominee, illustrating feature-based dependent agreement. In Menominee, pronominal verbal suffixes favour third person agreement over first and second person; only if all arguments are first or second person are third person markers absent, see (11).

1. (11)

   1. (a) Ne-na·n-ek-w.(Menominee; 222, ex. (2b))1-fetch.dir-[+3]‘He fetches me.’

   2. (b) Ne-po·se-m.(222, ex. (2a))1-embark-[−3]‘I embark.’

Trommer explores the consequences of these facts for current theories of phi-features and presents an O(ptimality)T(heory)-based analysis to account for the data. Although it is standardly assumed that third person results from the absence of first or second person features and that it does not constitute a third person feature of its own (see e.g. Harley & Ritter Reference Harley and Ritter2002), Trommer argues that the Menominee data show that these conclusions are too strong (a conclusion independently reached by Nevins Reference Nevins2005). He rejects a feature-geometrical model and suggests a rich feature inventory, where all features are binary and first, second and third person features have distinct semantic denotations.

The most important claim made in this paper concerns the existence of [±3]. It remains unclear what explains the tendency across third person markedness that otherwise would result from a feature hierarchy. One possibility may be that [±3] does fit in a feature hierarchy, but that it is ranked very low and hardly ever realized in languages (comparable to, say, quadral features), but this option is not explored in the paper.

In ‘When is a syncretism more than a syncretism?’, Heidi Harley is concerned with the presence of a single phonological form for a multitude of features (like English was, which realizes both first and third person past).

In classical Distributed Morphology analyses (e.g. Halle Reference Halle1997), syncretisms are the results of competition between a set of underspecified forms that appear in a particular order. Harley examines the case/number paradigms of pronominal and nominal suffixes in Russian, which lack gender agreement in the plural. Accordingly, third person nominative singular pronouns exhibit gender forms (on ‘he’, ona ‘she’, ono ‘it’), whereas their plural counterparts do not (the only available form is oni ‘they’). One way to derive this pattern is by assuming four competing Vocabulary Items (VIs).

1. (12)

   • /-i/⇔[plural](256, ex. (5))

   • /-a/⇔[feminine]

   • /-o/⇔[neuter]

   • -Ø⇔elsewhere

However, for such an analysis to work successfully, these VIs need to be ordered with respect to each other. Suppose that a node [plural, feminine] needs to be realized. Then, two candidates are in principle available: -i and -a. Only by ordering -i before -a can *ona be ruled out as a possible plural feminine pronoun.

At first sight, this is not a very attractive account, as it requires pure stipulation to derive the required ordering. However, third person nominative pronouns do not constitute the only paradigm where the plural exhibits syncretisms – this is attested for many other domains, such as, for instance, adjectival suffixes. Crucially, no paradigm shows gender marking only in the plural. Harley makes use of the concept of meta-syncretism (Williams Reference Williams1994) to solve this problem, suggesting that Russian has the rule in (13).

1. (13) [+plural, +gender]→[+plural]

If (13) applies before vocabularization takes place, the syncretism facts can be accounted for without having to adopt ordered VIs. Thus, for the example above, [feminine] is erased in a node with [plural] and [feminine], yielding *ona an impossible candidate for VI insertion.

However, one may even go further and assume that plural gender features are absent form the start of the derivation, which should manifest both morpho-phonological and semantic effects (i.e. semantic plurality should lack gender presupposition effects in Russian). While Harley raises this suggestion, she does not adopt it, and her overall conclusion is that regardless of any further theoretical implications, it is important to acknowledge the presence and theoretical status of meta-syncretism in phi-theory.

An important article in this volume is the one by Jonathan David Bobaljik, ‘Where's Phi? Agreement as postsyntactic operation’, which deals with the question as to what determines which DP controls for phi-agreement and in which part of the grammatical model phi-agreement takes place. Bobaljik first argues that it is morphological case (m-case) rather than grammatical function that controls Agree. M-case is the case that a DP gets assigned by the morphological output of the grammar. In Icelandic, nominative case is assigned to the highest DP if that DP lacks lexical case by itself. However, if the subject already carries dative case, it is the object that will receive (unmarked) nominative case, see (14).

1. (14) Jónilíkuðuþessirsokkar.(Icelandic; 298, ex. (4a))John.datlike.plthese.plsocks.pl‘John likes these socks.’

The type of case assigned to the object is thus not dependent on its grammatical function, but rather on its position in the clausal structure. For Bobaljik, this is a strong argument in favour of the idea that case assignment takes place after syntax proper (see Marantz Reference Marantz1991). If case assignment were to take place within syntax, it should be reflected in the semantics as well as in terms of the grammatical functions of the case-bearing DP, contrary to fact.

Furthermore, Bobaljik shows that it is m-case and not the grammatical function that controls agreement. In (14), it is the object that is responsible for the plural marking on the verb, not the subject, only because the object bears nominative case. Had the subject been nominative, then the verb would have reflected the number feature of the subject instead of that of the object.

The final conclusion that Bobaljik draws from these data is that, if case assignment is a post-syntactic operation and m-case is the source for phi-agreement, then phi-agreement is a post-syntactic operation as well. He thereby departs from classical theories of phi-agreement that put phi-features at the heart of syntax proper (e.g. Chomsky Reference Chomsky1995). If Bobaljik's approach is correct, it necessitates a radical revision of standard probe–goal agreement.

As mentioned in the final chapter, by Andrew Nevins (‘Cross-modular parallels in the study of Phon and Phi’), all previous papers in this volume discuss phi-features from either a pragma-semantic or a morpho-syntactic angle. In contrast, Nevins's contribution considers two striking parallels in the domains of phi-theory and phonological theory.

First, he argues that Person Case Constraint (PCC) effects, i.e. bans on various combinations of clitics or agreement markers, are remarkably similar to the way that vowel harmony is constrained in phonology, in the sense that both phenomena result from restrictions on agreement between features. Such restrictions are (i) based on markedness and contrastiveness and (ii) subject to cross-linguistic variation.

Second, Nevins argues that certain impoverishment rules reflect standard patterns in phonology. Take for instance the minimal pair in (15).

1. (15)

   1. (a) Aren't I?

   2. (b) *Amn't I?

What appears to be the case in (15) is that two marked features ([+author, +singular] and [+neg]) are to be spelled out on one and the same node, which is banned. As a repair strategy, the first feature changes into its unmarked counterpart ([−author, +singular]). This is similar, Nevins argues, to metaphony effects known from several Italian dialects (cf. Calabrese Reference Calabrese1998), where a phonological marked feature [+high, −A(dvanced)T(ongue)R(oot)] is banned if it appears under a [–consonant] node, which then leads to the deletion of the [−ATR] feature.

Nevins concludes that the language component in the human mind consists of a number of economy mechanisms that may apply in various domains and that such principles are not restricted to sub-domains of the grammar.

In this review, I have tried to reflect on what kind of problems are currently addressed in the study of phi-theory, what kinds of insights have been drawn and which questions are still open. It can be safely concluded that the volume under review collects in one place a number of high-quality papers on various issues in phi-theory. In my opinion, Heim's and Bobaljik's chapters stand out among the rest, but all papers are of solid quality and provide insights that will shape the research agenda for future years.

I would like to conclude by commenting on the goals and objectives for this agenda, as I see them. It strikes me that while much work is devoted to issues concerning the semantics–pragmatics and the syntax–morphology interface, more work needs to be done on the syntax–semantics interface. The interpretational effects of most phi-agreement configurations are still poorly understood, even though they are at the core of both syntax and semantics. This affects not only the way that agreement is driven by uninterpretability, but also the notion of entailment, which plays an important role in Béjar's and Harbour's papers. If verbal agreement features are not directly interpretable, it appears to be problematic to capture feature geometry in terms of entailment, at least not without further refinements.

Clearly, what is called for is much more interdisciplinary research on phi-theory and the way that phi-features interact in different domains of the grammar. The editors of this volume have made a remarkable contribution in bringing together researchers from different perspectives, which hopefully will result in a fruitful and flourishing research program in the years to come.