AGREEment and ellipsis

1 A window into agreement

Merge is generally considered a fundamental, “non-negotiable” property of human language that hierarchically organizes linguistic objects (making language structure-dependent). In contrast, the question of whether we need computational operations other than Merge for the construction of syntactic objects requires greater discussion. Looking at the pervasiveness of agreement phenomena in natural languages, many defend that one such operation should be posited: Agree, which relates features of syntactic objects (Chomsky 2000, 2001). These feature-relations find their expression in morphological inflection in highly variable, language-specific ways.

Regarding case, after Vergnaud’s famous 1977 letter to Chomsky and Lasnik, Lasnik (2008: 18) succinctly stated the following (the so-called Vergnaud’s conjecture):

Vergnaud’s now very familiar basic idea was that even languages like English with very little case^[1] morphology pattern with richly inflected languages in providing characteristic positions in which NPs with particular cases occur.

Thus, much current work assumes the need for an operation Agree even for languages without rich case morphology.^[2] This operation is considered to be asymmetric, as it relates initially unvalued ϕ-features of a probe to matching inherent (or valued) ϕ-features of a goal. And this procedure has been argued to take place under feature identity in a local domain. More formally, Baker (2008) summarizes Chomsky’s ideas as follows:

Even though many authors follow a formulation in these lines, others point towards the difficulties and questions that this proposal still leaves open. For example, in the above formulation Agree is assumed to only function downwards, since the probe has to c-command the goal. Nevertheless, recent work has also proposed that Agree may be bidirectional, working upwards when the goal must c-command the probe (see Baker 2008; Béjar and Rezac 2009; Toosarvandani and van Urk 2014, for example).

Besides, Chomsky (2000, 2001) argues that when a probe and a goal establish a relation, uninterpretable features get deleted, and that a probe can only agree with a ϕ-complete goal. In contrast, authors such as Danon (2011) consider these assumptions problematic and adopt a feature sharing approach to Agree (see also Frampton and Gutmann 2000; Pesetsky and Torrego 2007). According to this later account, when a link between a probe and a goal is created, unvalued features can establish an Agree relation with other unvalued features.

From a more global perspective, in their recent overview on agreement within the Minimalist Programme, Smith et al. (2020a) identify the following prominent debates surrounding various aspects of the formulation of Agree:

1. Should agreement be handled by a dedicated operation of Agree, which is a primitive operation of the syntactic component like Merge?

2. If so, what is the direction of the Agree operation?

3. Are agreement relations restricted to certain feature types?

4. What is the relevant locality domain of Agree?

5. What phenomena should be handled by Agree?

6. Is agreement parasitic on other factors, or can it apply freely?

7. What is the interaction of Agree with other operations (e.g. Labelling, Merge)?

8. Is Agree fully syntactic, fully post-syntactic, or spread across both domains?

Regarding the last item in the list, for the past decades there has been an important debate regarding an overarching question on the general organization of the human language system: does the narrow syntactic computation really need an operation Agree in addition to Merge? Or, looking at a connected issue from a different perspective, where in the derivation and how is case determined?

On the one hand, defendants of a syntactic approach to case and agreement such as Chomsky et al. (2019) argue that the syntactic approach to agreement is based on the assumption that Agree mediates assignment of structural Case and serves to eliminate semantically redundant ϕ-features from a given syntactic object (Chomsky 2000 et seq., building on Vergnaud 1977 [2006] and George and Kornfilt 1981). In this account, case-assignment or feature-checking would happen in the syntactic component. On the other hand, externalist approaches to these phenomena consider case to be a purely morphological phenomenon (Marantz 1991; McFadden 2004, for instance), where uninterpretable features can be neglected at the conceptual-intentional interface (in line with Sportiche 2015). Following this later externalization approach, case would be determined post-syntactically, when syntactic objects are transferred to their phonetic representation. Hence, rather than assuming a fully syntactic operation like Merge, another possibility would be relegating featural interactions to externalization. Following this second line of thought, agreement relations would be established in the mapping from narrow syntax to the phonetic representation, accessed by sensorimotor systems and providing instructions to the vocal or gestural articulators, were agreement and case assignment would take place (for further information on the interaction of case and agreement, the c-command factors and the timing of case assignment, the categories involved in case interactions, or on the theory of dependent case, see Baker 2015).

Moving to a different relevant issue regarding case and agreement, it is worth noting that natural languages display two possible case configurations: ergative-absolutive systems, or nominal-accusative systems, as shown in (2) from Coon et al. (2017):

The term ergativity has been most commonly used to refer to systems with one or both of the following properties: (i) transitive subjects (A arguments in Figure 1) pattern differently from intransitive subjects (S arguments) and transitive objects (P arguments); while (ii) transitive objects and intransitive subjects pattern alike. This type of system contrasts with nominative–accusative configurations, schematically illustrated on the right in Figure 1.

Figure 1:

On the left, ergative-absolutive case configuration and, on the right, nominative-accusative case configuration.

In the last decades, authors such as Longenbaugh and Polinsky (2017) – which review recent experimental work testing ergative-specific questions involving alignment, long distance relations, and agreement in a range of languages including Hindi, Basque, Niuean, and Ava – have stressed the importance of also doing work on ergative languages to resolve some confounds that are found in the existing literature on nominative-accusative languages and on the general theories on agreement. This is due to the fact that the alignment of grammatical case with grammatical function can be teased apart in ergative languages. For overviews on ergativity and related matters in the generative tradition, see Oyharçabal (1992), Bobaljik (1993), Johns (2000), Aldridge (2008), Rezac et al. (2014), Deal (2015), or Coon et al. (2017) and references therein.

Besides, another well-studied phenomenon with relevant consequences for the general theory of agreement is Differential Object Marking (DOM): a syntactic phenomenon where certain direct objects are overtly marked to (arguably) signal specific semantic or pragmatic features, such as definiteness, specificity, or animacy (for a dive in the relationship between DOM, case and agreement, see Baranji and Kalin 2020). This marking often involves the use of prepositions or morphological markers. For instance, in Spanish, animate and specific direct objects are often preceded by the preposition a, as in Veo a la profesora (‘I see the teacher’), whereas inanimate objects lack this marker, as in Veo la casa (‘I see the house’) (Ormazabal and Romero 2007, 2010, 2013a, 2013b; Torrego 1998; Uriagereka 1988, among many others). Understanding DOM is particularly relevant for a general theory of case and agreement, since authors who have examined DOM, such as Fábregas (2013), for instance, have tried to provide an answer to the following highly debated questions in the literature: What is the nature of case marking? DOM differentiates between at least two classes of objects both correlating with arguments traditionally described as accusative, but which are distinctly marked. If both markings are ‘accusative’, what really differentiates them, and why does grammar differentiate between them? And, in turn, what does this fact tell us about what case really is?

The presence of DOM varies across languages and is influenced by factors like the prominence of the object in discourse, the syntactic structure in which a DOM constituent appears and its inherent properties. Some studies, such as Lazard (1984), Bossong (1985) or Aissen (2003), provide in-depth analyses of DOM across different languages.

Finally, be it from an experimental perspective – such as Hagstrom (2016), which looks at children’s acquisition and use of case and agreement morphology – , or be it from a more theoretical perspective as previously discussed, most influential work on case and agreement is focused on offering a general theory on agreement within a Minimalist framework (Baker 2013a, 2013b; Boeckx 2008a, 2008b; Bošković 2007; Chomsky 1998, 2000; Pesetsky 2007; Polinsky 2016; or Smith et al. 2020b; Sigurðsson 2012, among many others).

2 Significatio ex nihilo: ellipsis

A traditional characterization of language, going back to Aristotle, defines it as “sound with meaning”. In contrast, ellipsis generally refers to the presence of a form/meaning mismatch, where there is significatio ex nihilo (that is, ‘meaning out of nothing’).

From a descriptive perspective, elliptical expressions are ubiquitous in natural language and include a wide range of phenomena, such as those in (2)–(4), where certain aspects of their meaning are left unexpressed.

Two questions that may come to mind by looking at the surface differences among these examples are the following:

1. To what extent does the term ellipsis adequately describe the entire set of phenomena illustrated above?

2. Under what general conditions are the clausal, predicative or nominal gaps in (2)–(4) allowed in natural languages?

The common property shared by these examples is that they feel incomplete in some sense. This discrepancy between what is overtly expressed and what is intended poses great challenges for generative theories of language and of form/meaning correspondence. From a theoretical perspective, competing accounts try to offer an answer to the essential issue of how to map sound and gestures to their corresponding meanings when words appear to be missing. Early work on this form/meaning conundrum raises general and fundamental questions about the relationship between grammar and cognition (Hankamer 1979; or Sag 1976; Ross 1967; Wasow 1972, among many others), which cannot be answered without developing fine-grained analyses of the inter- and intra-linguistic variation regarding diverse elliptical constructions in different languages.

Work on ellipsis in the last decades has revisited previous proposals, such as focus-based approaches to antecedent identity (Barros and Kotek 2019; Fox 2000; cf. Merchant 2001; Romero 1998; Rooth 1992; Winkler 2011) or pragmatic accounts on ellipsis based on discourse equivalence (AnderBois 2011, 2014; Ginzburg and Sag 2000; Poppels 2022; Weir 2014).

Also, recently there has been a rise in experimental approaches to ellipsis, where theories on various elliptical phenomena (VP Ellipsis, Sluicing, Gapping, RNR, Fragment answers) and in various languages (French, English, Spanish, Saudi Arabic) are informed on results based on empirical data obtained through experimentation (Alshaalan and Abels 2020; Bîlbîie and de la Fuente 2019; Ronai and Stigliano 2022; Xiang et al. 2019; among many others), as in the case of voice or antecedent mismatches (Grant et al. 2012; Parker 2018; Poppels and Kehler 2019).

Besides, within a generative enterprise, most authors follow a minimalist approach to elliptical phenomena and try to reduce its effects to independent restrictions on the interfaces. An example would be prosodic theories of ellipsis, that is, accounts which locate the computation of ellipsis in the syntax-prosody mapping, sometimes connected to Match Theory (Selkirk 2011), for example Colley and Bassi (2022) (see also Wagner 2010).

Following a desire for economy as well as the idea that entia non sunt multiplicanda praeter necessitatem –‘entities are not to be multiplied unnecessarily’ (William of Ockham, c.1285–1349)–, recent work has also disposed of construction-specific phenomena such as Comparative Ellipsis and Comparative Deletion, by showing that those phenomena can be reduced to other more general cases of ellipsis in coordinate or dependent/subordinate constructions (see Lechner 2004; or Vela-Plo 2023, 2025 for different proposals on several languages).

Recent developments on ellipsis in natural languages include influential work such as Merchant (2001) on sluicing, islands, and proposals for a general theory of ellipsis; Winkler (2011) on ellipsis and focus; Johnson (2014) on gapping; or the collective work in van Craenenbroeck and Temmerman (2018), which presents an overview of some of the main theories of ellipsis, introduces ellipsis as a diagnostic tool, and collects a taxonomy of elliptical constructions and several case studies exploring the elliptical inventory of certain languages.

Moreover, latest work on ellipsis has not only developed informed theories on particular phenomena (such as Valmala 2013 on RNR; Aelbrecht and Harwood (2015) on VP ellipsis; Johnson 2017 on gapping; or Ott 2016 on appositives), but also on instances of ellipsis in less-studied languages with this respect (van Craenenbroeck 2010; Landau 2018; or Bîlbîie and Faghiri 2022, for instance), and have pointed towards open questions for further research from an empirical as well as a theoretical perspective.

Nevertheless, we still find a need for comprehensive studies in language-by-language and construction-by-construction basis that enrich the empirical foundation for general theories on ellipsis, for example, of ambiguous configurations in particular less-studied languages, which a priori are candidates for more than one elliptical construction; or on the directionality of ellipsis from a typological perspective, among other lesser-studied topics.

3 Contributions of this collection

Bearing these concerns in mind, this special issue proposal stems from the stimulating works presented at the 32nd Colloquium on Generative Grammar and at the Workshop on Case, Agreement and Ellipsis immediately following the 32CGG (https://sites.google.com/view/cgg32/), organized by the Basque Research Group of Theoretical Linguistics HiTT in April 2023 (https://www.hittlinguistics.eus/). The volume contains six articles, based partially on some of the talks previously selected through a peer review process for the conference, as well as a preceding introductory article by one of the editors and two articles which do not emanate directly from the conference, but are directly related to the topics of agreement and ellipsis discussed there.

Three of the papers focus on agreement and related notions, as they discuss feature specification (or underspecification) on heads based on inflected infinitives in European and Brazilian Portuguese; the nature of φ-feature valuation and the agreement mechanism itself, challenging the traditional probe-goal model of agreement; or a predicate ellipsis alternation in contemporary Spanish where clitic lo appears to participate in agreement licensing.

Specifically, Martins and Nunes’s article examines adverbial infinitival clauses in Portuguese such as Os meninos saíram antes de jantar(em). ‘The boys left before having dinner(infl).’, where inflection is typically optional, but can become obligatory for certain speakers. Building on previous work by Nunes and Raposo (1998), the authors identify three contexts that enforce inflection: with adjectival predicates, passives, and reflexive clitics. The authors propose an analysis within Chomsky’s (2001) Agree-based framework, arguing that variation in inflection arises from phi-feature underspecification and minimality constraints. Their findings suggest that Portuguese infinitival structures are more complex than the traditional binary distinction between inflected and uninflected forms, leading to the identification of two distinct subtypes of inflected infinitives.

Ormazabal and Romero’s article explores agreement processes in syntax, arguing that φ-feature valuation is not the trigger but rather the result of a syntactic relation initiated by an independent operation. They challenge the traditional probe-goal model of agreement, proposing that some agreement effects are induced morphologically or in processing rather than being strictly syntactic. The authors illustrate their argument with cases such as regular subject agreement, the Agent-Focus construction in K’ichee’ (analyzed as agreement displacement in morphology), or existential constructions in English, Spanish, and French (treated as processing-based agreement). Ultimately, they emphasize that while some agreement processes occur outside syntax, they are still indirectly determined by syntactic derivations where the φ-feature matrix remains underspecified.

Next, acting as a bridge between de main topics of this volume – agreement, and ellipsis – , Alcaraz’s article explores a predicate ellipsis alternation in contemporary Spanish where clitic lo appears to participate in agreement licensing, challenging existing theories of ellipsis licensing within the Minimalist Program. Spanish predicate ellipsis appears in two forms: lo-insertion, where the neuter clitic lo precedes a silent gap, and lo-omission, where ellipsis is licensed under polarity focus. Rather than assuming two distinct [E]-features, the author argues that a single [E]-feature underlies both constructions, and that the presence or absence of lo reflects different ways of satisfying semantic requirements, rather than syntactic encoding of ellipsis. The study also extends this analysis to Germanic languages, highlighting parallels in predicate anaphora strategies despite differences in distribution.

The three remaining papers share a common theme of discussion, as they deal with various constructions that have been or may be analysable as cases of ellipsis. Besides, they offer insightful proposals into topics previously discussed in the ellipsis literature.

Ott´s article looks at an old problem of syntactic theory through the lenses of discourse: the phenomenon of Right-node Raising (as in Alice bought __, and Bill read __, Chomsky’s new book). Concretely, the author argues that Right-node Raising configurations consist of an unbounded sequence of potentially incomplete expressions whose prosodic realization provokes the expectation of a continuation, which is furnished by an either sentential or fragmentary sequence-final expression. Accordingly, Ott defends the idea that RNR is not a bona fide syntactic phenomenon at all, but rather a paratactic configuration composed in discourse.

In their article, Schwarzer, Weir and Fábregas examine a specific pattern of ellipsis in Spanish, where ellipsis in the main clause is licensed by an adverbial clause. They argue that this phenomenon supports the LF-copying approach to ellipsis, which explains why antecedents cannot be base-merged within ellipsis sites. The authors analyse examples like Cuando Juan un artículo, María escribe una monografía (‘Whereas Juan writes an article, María writes a monograph’) to illustrate that certain adjunct clauses (i.e. peripheral but not central adjuncts) can provide the antecedent for ellipsis in the matrix clause. Assuming an LF-identity condition between the antecedent and the ellipsis site, and that remnants move at LF, they derive the data appropriately, provided that the antecedent isn’t itself contained (i.e. interpreted) in the ellipsis site. Their study contributes to the broader understanding of ellipsis restrictions and syntactic structures in Spanish and other languages.

Finally, van Craenenbroeck and Johnson’s article examines the behaviour of symmetric predicates under VP-ellipsis. Symmetric predicates, such as marry and fight with, exhibit a unique property where their arguments can be switched while maintaining logical equivalence (e.g., John married Mary ⇔ Mary married John). The authors build on Stockwell (2020), who observed that symmetric predicates allow participant switching and transitivity switching under ellipsis, suggesting a semantic identity requirement. However, they challenge this view by showing that focus-related cases are not correctly predicted under Stockwell’s analysis. Instead, they propose an alternative syntactic approach, arguing that symmetric predicates should be reanalysed as unaccusatives, with both arguments initially merged as internal arguments before being raised to the subject position.

We consider that all contributions provide interesting data and impressive ideas which will enrich our knowledge of the way in which the language faculty works regarding agreement and ellipsis phenomena.