Linguistic Analysis

A research journal devoted to the publication of high quality articles in formal phonology, morphology, syntax and semantics. In continuous publication since 1976. ISSN: 0098-9053

Introduction

On Some Formal Approaches to the Syntax-Lexicon Interface

Introductory Remarks

This volume constitutes a collection of approaches to the interaction between the lexicon and the syntactic component that uses its items in order to build linguistic expressions. In the following sections we review some of these approaches in order to understand the intricacies of the lexicon-syntax interface, starting from the structuralist notion of lexical items (henceforth, LIs) and surveying the most relevant landmarks in the development of generative syntactic theory with the aim of showing how the notions of the lexicon and the role it plays in syntax have evolved over the years.

There are three questions that guide and form the backbone of the first sections of this Introduction:
Q1: What is the lexicon and how is it formed?
Q2: Where and when is it situated with respect to the syntactic component?
Q3: Does it behave in a syntactic way, that is, are there computational operations of a syntactic nature involved in its structure and functioning?

These three issues are naturally inter-related. Q1 can be partially stated as a matter of localizing the formation of the lexicon in the process of language acquisition. However, if we understand “formation of the lexicon” as referring just to whether LIs are formed in a certain way (by clustering features together, or any other method) before syntax begins taking elements from it to construct expressions, then that first question can be rephrased in terms of the other two.

The other two questions boil down to asking ourselves and the syntactic models we use just how atomic are the elements of the lexicon. The notion of atomic needs some clarification here since it will be a central concept for any approach to the lexicon-syntax interface across different linguistic models. An intuitive idea of atomicity would mean that each element constituting a separate entity in the mental lexicon (i.e., each LI) works like a “black box,” which either possesses internal separable content or not, but at any rate that content is invisible to all syntactic computations.

In physics and chemistry, atoms are known to be made up of smaller subatomic particles, but it looks like most natural processes at levels above that of the atom target atoms as a whole, and not the particles within. Extending the parallelism into syntactic theory, the question would be whether LIs are like atoms or subatomic particles.

With respect to Q2, it is obvious that purely atomic LIs–the contents of which never interact with syntax–must necessarily precede syntactic computations. Nevertheless, a conception of the lexicon in which parts of items (which we know cannot appear on their own as a general rule, if we think about, for example, things like past morphemes in a language like English) enter the derivation, may involve a more complex relationship in which syntax and the lexicon are interwoven. That is, if LIs contain (and in fact, are) bundles of features formed by some operation of merger, which we take to be the basic syntactic operation, then we have syntax before LIs, at least to a certain extent, since the former is necessary to build the latter.

Q3 underlies much of this discussion, since it involves deciding if a theory is lexicalist or not. In order to proceed, we must say what we mean by lexicalism. We may define a theory as lexicalist when the internal components of LIs (phonological, syntactic and semantic features) do not enter syntactic computations on their own, although they may affect their operations and outcome; that is, LIs are fully arranged in their spelled-out form before syntax tampers with them.

This introduction is organised as follows: section 2 reviews the approaches to the lexicon-syntax interface within the generative tradition, following the most important landmarks of that proposal; in section 3, we provide a summary of approaches to the matter from two different contemporary perspectives (which we could label endo- and exo-skeletal); finally, section 4 contains a survey of loose ends from different approaches, which are explained in relation to the papers that constitute the core part of this volume.

2. The generative tradition

The next sub-sections briefly review the status of the lexicon in the generative tradition starting from Chomsky (1965), but we will first provide a short overview of the concept of LI in the structuralist approach.

2.1. The notion of “lexical item” in a structuralist approach to language

Since Saussure (1916), LIs have been considered as arbitrary pairings of sound and meaning, a foundation of the structuralist tradition. This means that, simple as this notion may be, a concept of LIs containing a certain degree of internal composition in terms of phonetic and semantic sub-parts was present from the very beginning.

It should also be noted that the relationship between the lexicon and the syntactic component cannot be assessed within this framework, insofar as structuralist theories did not contain any elaborate theory of syntax that made explicit the relationship existing between these two core components of grammar.

2.2. Introduction to a review of generative approaches

As the first remark in our approach to generative insights into the lexicon-syntax interface, we shall mention how generativism has mostly been taken to be the staple of lexicalism and lexicalist approaches to syntax. It is generally understood that Chomskyan approaches to the lexicon assume that no internal computations play a role in the formation of LIs, and that words, so to speak, are taken as such from the mental list we call the lexicon and entered into the syntactic component as whole indivisible units.

As we will see below, this was not the case for the greater history of the development of generative approaches to the lexicon-syntax interface. As a matter of fact, we could claim that generative theory did not start out as a lexicalist approach to grammar, and only became so in its later developments.

2.3. The approach in Aspects of the Theory of Syntax (Chomsky, 1965)

Chomsky (1965:3) defines “formatives” as the “minimal syntactically functioning units,” which could lead us to assume that the approach is essentially lexicalist in nature, if all syntactic processes start from fully formed formatives, and if we consider such formatives to be synonymous with LIs. However, we are not given a precise definition of what amounts to the minimal syntactic unit. We will immediately see that it cannot mean a fully formed word, as they appear only after lexical insertion to form the Deep Structure of sentences.

A crucial aspect at this stage of the evolution of generative approaches to syntax is that the component preceding syntactic computations is not only the lexicon, but a more complex entity called the Base. The Base includes two separate components: the lexicon, subdivided into lexical and grammatical formatives, and a so-called categorical component that knows how categories are syntactically organized and manipulates elements of the lexicon accordingly.
The precise internal content of lexical entries is defined as follows (Chomsky, 1965:84&ff): “each lexical entry [is] a pair (D,C), where D is a phonological distinctive feature matrix “spelling” a certain lexical formative and C is a collection of specified syntactic features.” Those syntactic features include both the subcategorization frame (which specifies the selectional properties of items) and all idiosyncratic morphological properties that are not part of the computation. It is also important to note that syntactic features are only relevant for the interpretative components, not for syntactic computations.

A very important trait of the approach taken in Aspects to the question of the lexicon and its interface with syntax is the rewriting rules that regulate the insertion of terminal nodes (i.e., the phonological features of LIs) in syntactic derivations. Once we have built a syntactic tree in which all rewriting rules of the sort in (1) have been inserted, the terminal nodes of the tree will consist of a bundle Q of syntactic features:

(1) a. S → NP VP
b. VP → V NP
c. NP → D N
d. D → Q
e. N → Q'
f. V → Q''

That is the moment when rewriting rules apply, by comparing the set of features Q inserted by the categorical subcomponent with the specific set of syntactic features C contained in every lexical entry.

(2) Rewriting rule
When Q = C, then rewrite Q as D
D being the phonetic specification of the LI

These rewriting rules almost amount to a very early implementation of Distributed Morphology (DM) (Halle & Maranzt 1993, 1994, and subsequent work), and are therefore far from what we would consider strict lexicalism. The actual morpho-phonological content of items is introduced only after a whole structure has been built by the categorical component of the base, together with the selectional requirements of particular LIs. Therefore, if we consider those operations by the categorical subcomponent of the base to be syntactic in any way, we do have syntax before lexical insertion.

In the Aspects framework, however, those processes are not considered to be strictly syntactic, since they precede the transformational component of grammar that will turn the basic Deep Structures (formed by means of rewriting rules) into the final Surface Structures (with a fully fledged morpho-phonological content). Therefore, a difference from DM-based approaches (which will be reviewed later in this introduction) is that the insertion of lexical material is done prior to transformations, whereas DM allows (and sometimes demands) post-syntactic lexical insertions which are not subject to further computations.

Recall that the categorical subcomponent builds a deep structure that exists prior to any computation and that syntactic features are visible only to the interpretative components, not to syntactic computations. Besides, the fact that there are different points of access at which the lexicon can be interpreted means that the different subparts that make up every LI remain visible throughout the computation, even if the computation itself never interacts with them in any meaningful way.

A question that remains open up to now is how phonologically determined LIs are, that is, whether the set of phonetic specifications each lexical entry is endowed with allows for modifications or additions after the Base has completed its lexical insertion and transformational rules start their computation. From a traditional perspective, such morphological alternatives were divided under inflectional (related to agreement processes involving person, number, tense, and case) and derivational (related to category-changing morphemes or affixes involving a semantic change of the LI) morphology. If such alternations in the morpho-phonological component of formatives are to be explained by syntactic theory, we have an interesting case to see how many of the lexical features are introduced by a syntax-like computation device involved in lexicon formation.

The approach in Aspects, which would be roughly maintained in later formulations, is that each type of morphology receives a different treatment. Regarding inflectional morphology, Chomsky (1965:173-4) writes that:

[T]he often suppletive character of inflectional systems as well as the fact that […] the effect of the inflectional categories may be partially or even totally internal, causes cumbersome and inelegant formulation of rules. […] It follows that we must allow the transformational component to contain rules that alter and expand the matrix of features constituting a lexical item.

At any rate, features introduced during a derivation seem to be of a fundamentally different nature from those present from the start in the specification of the LI. If we compare this idea with the operation of Agree in the Minimalist Program (Chomsky 1995), we might see in this a seed of the operations dealing with uninterpretable features.

Concerning derivational morphology, the initial approach to the idea in Chomsky (1965) is inconclusive. On the one hand, Chomsky acknowledges that when there are no rules for correctly predicting the derivation of one word from another, those different items should indeed be listed in the lexicon, but at the cost of losing generalizations we could make about their internal structure. Nominalizations in particular are mentioned due to their apparent regular character (for instance in keeping argument structure), saying that the lexicon should not include one entry for destroy and another one for destruction, but rather a rule for turning one into the other. A crucial remark is made in Chomsky (1965:188-9):

In the light of such examples we may have to relax the requirement […] that branching is not tolerated within the dominance scope of a complex symbol. This restriction seems to hold *only above the level of the word.* [our emphasis]

The inclusion of a difference in the syntactic conditions that work above and below word-level means, on the one hand, the inclusion of a rudimentary distinction between what would become l-syntax and s-syntax, while on the other it further undermines the attributed lexicalist character of the proposal.

2.4. The approach in Remarks on Nominalization (Chomsky, 1970)

The approach taken in Chomsky (1970) regarding the format of the lexicon and its interaction with syntax does not differ to a great degree from the one in Aspects. Lexical entries continue to be specification matrixes including bundles of syntactic and phonetic features taken from an unordered mental list.

Chomsky (1970) nonetheless introduces as an important advance in his model of syntax in which syntactic features are part of the specification of LIs the fact that the model now allows defining lexicalist positions. Whereas the previous instantiation of transformative grammar in which LIs are totally atomic and impossible to alter by syntactic devices, is now considered a worse alternative than one in which syntactic features within LIs intervene in the computation that, together with general rules of combination (namely, X-bar Theory), captures the same generalizations as transformational operations, and at the same time avoids the problems that would occur if we adopted a purely transformational approach for derivational morphology.

Departing from the inconclusive stance assumed in Aspects, Chomsky adopts in Remarks a clear position against the syntactic treatment of derivational morphology. In spite of the loss of generalizations that it may imply, the lexicalist position on this issue is based in part on two arguments.

The first is that derived nominals do not exhibit a behaviour that differs in any sense for that of regular nominals, and therefore they would not need to appeal to transformational mechanisms to explain anything about their syntactic status.

The second is that nouns possessing a superficial morphology identical to that of nominals derived from verbs would in principle need a similar account of how they were formed. Therefore, nouns such as eagerness would entail the existence of abstract verbs that are never realized, like **to eager.* The existence of such verbs would then be a stipulation and no other motivation for their presence would exist anywhere in the system. Chomsky provides an example of how a lexicalist approach to nominalizations in which syntactic features in LIs play a role can be used to predict the behaviour of nominalized nouns without resorting to the transformational component to capture said behaviour by means of computational rules. The contrasts used to exemplify this are the following (among others):

(3) a. John is eager to please.
b. John’s eagerness to please.
c. John is easy to please.
d. *John’s easiness to please.

(4) a. John is certain to win the prize.
b. John’s being certain to win the prize.
c. *John’s certainty to win the prize.

Consider the examples in (3) first. The differences in grammaticality there are taken to follow from the different subcategorization frames allowed by the categorial component. Since eager admits a sentential complement, as in John is eager (for us) to please of the kind any noun would admit (as in A car for us to buy), it follows that a derived noun such as eagerness will also be able to fill that position. However, structures like John is easy (for us) to please cannot be generated by the Base: rather, they are the product of transformational rules applying to the structure For us to please John is easy, where easy is not in a position available to nouns. Therefore, a derived noun like easiness will not be able to occupy either the base or the derived position.

As for the examples in (4), there is a similar argumentation. John is certain to win the prize is considered to derive via transformational rules from a more basic structure in which certain is predicated of the proposition John to win the prize (Chomsky 1970:191). Therefore, at the level of the categorial component, certain cannot take a propositional complement, and neither can certainty.

As a result of this analysis, Chomsky managed to preserve his terms in describing both sides of the issue. In the first place, no transformational rules are applied to turn certain into certainty, each LI constituting its own entry in the lexicon. In the second place, the set of syntactic features that makes up the specification matrix of LIs allows predicting the behaviour of items in different syntactic contexts.

2.5. The approach in Lectures on Government and Binding (Chomsky, 1981)

Lectures on Government and Binding is probably the point at which previous generative perspectives on the lexicon-syntax interface started to change towards a view that fits more closely with the lexicalist approach that is usually attributed to generative models. This is achieved by the virtual elimination of the categorial component of the Base, leaving a more atomic lexicon as the only subcomponent of grammar preceding syntax.

The lexicon specifies the abstract morpho-phonological structure of each lexical item and its syntactic features, including its categorical features and its contextual features. The rules of the categorical component meet some variety of X-bar theory. [from Chomsky 1981:5]

Thus, the definition and format of the lexicon remained mostly unchanged from the model adopted in Aspects, except that categorial features are now included for each lexical entry, which strips the categorial component of much of its functions.

The perception of the lexicon and the role of the categorial component of the Base progressively change in the course of Chomsky (1981). Originally the categorial sub-component still generated D-structures where LIs were inserted (which is again interesting because this presupposes the existence of syntactic operations prior to lexical insertion). However, as we will see in the following section, the elimination of the categorial component that is undertaken in Lectures entails a more lexicalist approach to the lexicon-syntax interface and a more atomic view of the lexicon itself.

Two crucial steps are taken in Chomsky (1981) towards the removal of the categorial subcomponent: one is the generalization of X-bar theory (which had been introduced in Chomsky (1970)), the other is the application of the Projection Principle, which establishes a new way of interpreting the lexicon-syntax interface by proposing that all syntactic computations are projected from the lexicon. It is defined as follows (Chomsky, 1981:29):

(5) Projection Principle

Representations at each syntactic level (i.e., LF and D- and S-structure) are projected from the lexicon, in that they observe the subcategorization properties of lexical items.

Since subcategorization properties of lexical entries provided the particularities in the selectional frame of every item, and X-bar Theory provided a universal format for the building of phrases that were projected from a lexical head following the Projection principle, “the categorial component for a particular grammar will be quite meagre” (Chomsky, 1981:32). This was done for the sake of avoiding redundancies in the system:

[I]nformation concerning the class of subcategorization frames is in effect given twice in the grammar: once—implicitly—in the lexicon, as a property of the class of lexical items in its totality; and once—this time directly—by the rules of the categorial component. [from Chomsky 1981:31]

Therefore, if a LI is inserted into a frame different from the one it can project, the LF will be ill-formed and the derivation will be rejected.

In the “Principles and Parameters” framework introduced in Chomsky (1981), syntactic variation among languages was all that was left of the categorial component of the Base: “[T]he role of the categorial component for a particular grammar is reduced to the specification of values of such parameters as the order of constituents” (Chomsky, 1981:34). For instance, the different relative order of constituents that we find between a head-initial and a head-final language will be specified by the categorial component, but the kind of rules that used to establish D-structures in previous frameworks are now derived from more general principles, such as X-bar Theory.

As a result of all of these considerations, we can conclude that a shift towards a more lexicalist perspective was taking place; now there are no syntactic operations before lexical insertion, since the elimination of the categorial component makes sure that a syntactic object can only be built after LIs have been inserted, given that they are necessary ingredients for the Projection Principle to work.

2.6. The approach taken in The Minimalist Program (Chomsky, 1995)

The Minimalist Program presents the latest evolution in the conception of the lexicon within the generative perspective. Although the lexicon is seen under a more atomic perspective now in which LIs enter derivations fully formed, their internal features are central to syntactic computations, which are in fact driven by them. Furthermore, the existence of a Numeration of the LIs that enter the derivation provides a more complex interface between these two components of the grammar.

The Minimalist Program does not fundamentally change the format of the lexicon established in Chomsky (1981), but it proposes significant importance to syntactic features as part of LIs. In Chomsky (1995), syntactic features are taken from the lexicon, bundled together (or added from the same pool during the derivation), to form LIs, which are then introduced into a derivation. Crucially, these features are assumed to drive syntactic computations.

The definition provided in Chomsky (1995), while not being absolutely lexicalist, moves another step toward lexicalism from the previous instantiation in Chomsky (1981):

The lexical elements are sometimes called *atomic* from the point of view of the computational operations. Taking the metaphor literally, we would conclude that no feature of a lexical item can be modified or even addressed (say, for checking against another matching element) in a computational operation, and no features can be added to a lexical element. The condition as stated is too strong. [from Chomsky 1995:163]

By going back to the old discussion between inflectional and derivational morphology, The Minimalist Program adopts a position close to the one pursued since Remarks on Nominalization.

The Minimalist Program introduced a new perspective on the study of language when it examined not only the analyses behind linguistic data and phenomena, but also why those phenomena are the way they are and not different. For this reason, we could legitimately ask whether Chomsky (1995) addressed the issue of the lexicon in a similar fashion. In fact he does, although in a very brief manner.

The concept of Numeration is closely related to that of the Inclusiveness Condition; that is, all elements taking part in a syntactic derivation must have formed a closed list before the derivation itself starts. This list is formed by elements taken from the lexicon. Syntactic computations do not have the capability of adding elements that were not present from the start. Additionally, the notion of Numeration is related to that of phase (Chomsky 2000 and subsequent work). In its earliest incarnation, a phase was a subset of elements of a given Numeration that were known from the beginning to end up forming a part of the final syntactic structure.

By making use of the notion of phase as something integrated in the numeration prior to the whole computation process, we can establish a rich network of relationships within the lexicon than can be subsequently exploited by syntax. This enriches the interfaces but poses new questions as to how these intra-lexicon relationships can be formalized.

3. Some current perspectives

3.1 Endo-skeletal and exo-skeletal models

In this section we will introduce a very-well known classification that differentiates models dealing with the nature of the lexicon and its relations with the theory of grammar: namely, the division between endo-skeletal and exo-skeletal proposals. In addition, we will briefly review some of the most important current proposals that examine this topic.

One of the major debates around the nature of the lexicon and its relationship with the architecture of grammar is concerned with the difference between grammatical or structural meaning—as expressed by morphemes conveying e.g., [dual] or [feminine]—and the lexical or conceptual meaning—as expressed by cat or table. Whereas everyone agrees that natural languages convey both meanings, the disagreement emerges regarding, at least, the following issues: where are grammatical and conceptual meanings expressed, and how are grammatical and conceptual meanings related to the minimal units that syntax operates with and to the computational system itself.

Although the answers to these questions are very heterogeneous and diverse, it is possible to recognize two main lines of research. On the one hand, a lexicalist line considers that both grammatical and conceptual meanings are expressed in the atomic units used by syntax, and that the computational system manipulates them following self-contained instructions, like their category or their subcategorization frame. On the other hand, a structuralist line of research argues for a proposal where lexical units contain only conceptual meaning and all grammatical meaning emerges from the structure built by the computational system.

These two lines of research have received several nomenclatures. In this introduction we will follow Borer’s (2003) terminology in calling the first “endo-skeletal” and the second “exo-skeletal.” Although some proposals cannot neatly be ascribed to one or the other of the two groups, and different studies within the same group differ both theoretically and technically, this classification can serve as a useful guide to differentiate some of their main features.

Endo-skeletal theories provide listed items in the lexicon with more information than simply their conceptual specification.1 This is particularly clear in the case of verbs: the listed item eat would consist not only in its lexical meaning, but also in its syntactic and morphological properties.2

(6) EAT
Grammatical category: [v eat]
Lexical meaning: to ingest
Subcategorization frame: Agent/NP eat Theme/NP
Morphological properties: irregular verb > past: ate, participle: eaten

When listed items enter into the computation their properties are projected in the structure, for example the “agent” and the “theme” of the verb eat. In this sense, one could say that endo-skeletal theories propose a deterministic syntax where items of the lexicon drive the computation.

The following approaches (among others) can be considered endo-skeletal: Kaplan & Bresnan (1982), Jackendoff (1990), Grimshaw (1990), and Levin & Rappaport Hovav (1995), to the latter of which we will return.

On the other hand, exo-skeletal models assume that properties typically associated with listed items are really emergent properties from the syntactic structure that the items enter in to. Following this view, both the grammatical category and the subcategorization frame of a given item are not self-contained properties of it, but they are read off from the concrete structure in which the item appears.

If we take the same example as before, in the exo-skeletal models, the item eat would enter into the derivation as a non-categorized root √EAT. The root would be interpreted as a verb if merged with a verbalizing functional head (v), whereas the specifier and the complement would be interpreted from the configuration as the agent and the theme, respectively:3

(7) [vP Specifier [v’ v [VP eat Complement ] ] ]

As Acedo-Matellán (2010) observes, in attributing all but conceptual meaning to listed items, exo-skeletal models do not propose any relevant lexicon-syntax interface:

In attributing all not purely conceptual semantic aspects of linguistic expressions to the syntactic structure, paradoxically, exo-skeletal theories turn out not to be theories of the lexicon-syntax interface any more, as they do not envision any such interface. They attempt to explain problems of the relationship between lexical semantics and syntax, distributing what has traditionally been packed together as lexical semantics into compositional semantics and conceptual semantics, and rethinking the former as an emergence of syntactic structure. [from Acedo-Matellán 2010:23]

The following models (among many others) are considered exo-skeletal: Borer (1994, 1998), Harley (1995), Marantz (1996), van Hout (1996) or Mateu (2002). In light of this classification, some concrete proposals on the nature of the lexicon and its relation with grammar will be briefly reviewed: Levin & Rappaport Hovav’s (1995, 1998, etc.) model, Hale & Keyser’s (1992, 1993, etc.) model, the Distributed Morphology framework, the Nanosyntax framework, and Ramchand’s (1997, 1998, etc.) model. Our goal is not to provide an exhaustive characterization of each, but to highlight some of their main insights about the topic. Some of these frameworks are represented by the contributions contained in the present volume, while others are not. Nevertheless we consider it is interesting and useful to introduce all of them in order to have a more general picture of the current research scenario.

3.2 Levin & Rappaport Hovav’s model

The model developed by Levin & Rappaport Hovav (1995, 1998, and much subsequent work) can be considered endo-skeletal insofar as listed items in the lexicon are assumed to carry not only encyclopedic content but also grammatical meaning. Focusing on verbal items, Levin & Rappaport Hovav assume that listed items are associated with two lexical representations: the lexical conceptual structure, which corresponds to the conceptual structure (Jackendoff 1983), and the lexical syntactic representation, which specifies argument structure. Therefore, the authors claim that some syntactic properties of verbs are derived for their semantic specifications before they enter into the derivation: “Verb meanings include certain common elements that tie together into semantically defined classes” (Levin & Rappaport Hovav 1995, 23).

Regarding the lexical semantic side, Levin & Rappaport Hovav use representations that contain a) primitive predicates, like CAUSE or BECOME, provided by a universal set, and b) constants, which encode the idiosyncratic meaning of verbs, like BUTTER. See the semantic representation of this denominal verb:

(8) Butter: [x CAUSE [BUTTER BECOME Ploc z]
[from Levin & Rappaport Hovav 1995:24]

In (9) the lexical semantic representations of three verbs of change of state are expressed, which are all instances of the same lexical semantic template (10):

(9) a. Dry [y BECOME DRY]

b.  Empty   [y BECOME EMPTY]

c.  Warm    [y BECOME WARM]

(10) Noncausative verb of change of state [y BECOME STATE]
[from Levin & Rappaport Hovav 1995:27]

According to this model, all these templates are part of the speaker’s lexicon, that is, they are not generated by the computational system in the derivation.

As for the lexical syntactic representation, the authors assume that verbs are associated with their argument structure, which encodes the syntactical relevant properties. The argument structure they propose for put is the following:

(11) e, x <y, Ploc z>

where e is an event position, x is the external argument, and the variables inside the brackets, the internal arguments (Levin & Rappaport Hovav 1995, 21). Given the lexical syntactic representations of the argument structure of verbs stored in the lexicon, the Deep-Structure of syntactic derivations is automatically derived by syntax, which only has to follow the instructions in the templates.

3.3 Hale and Keyser’s model

The model by Hale and Keyser (1993, 1998 and much subsequent work) on argument structure and its relationship with syntax and the lexicon has generated a vast range of research on this topic, and has introduced into the linguistic literature one of the most influential distinctions on the theory of grammar and the lexicon of the last several decades: the division between l(exical)-syntax and s(yntactic)-syntax.

Hale and Keyser argue that, whereas s-syntax is the traditional module for the computational system of natural languages (plain “syntax” in other approaches), there is another component, called l-syntax, that is in charge of building LIs following general syntactic principles.

Actually, the authors follow one of the insights developed in Chomsky (1981): the idea that syntax is projected from the lexicon. Given this thesis, syntax obeys instructions contained in LIs, as the projection of an agent and a theme when a transitive verb is inserted into the computation. As Hale and Keyser explain: “each lexical head [V, N, P, A] projects its category to a phrasal level and determines within that projection an unambiguous system of structural relations holding between the head, its categorial projections, and its arguments (specifier, if present, and complement)” (Hale & Keyser, 1993: 53). The authors call these projections, which are stored as listed items, lexical relational structures (LRS).

Focusing on a concrete verb, Hale and Keyser propose that the lexical entry for shelve includes the following LRS, which has been built in l-syntax according to general syntactic principles, like binary branching (Hale & Keyser 1993:95):

(12) [VP V [VP NP [V' V [PP P [NP [N Shelf] ] ] ] ] ]

Therefore, lexical entries are learned and stored as syntactic structures that, once selected, enter into the syntactical derivation (or s-syntax). The entry of the verb shelf in (12), for example, would be inserted into the s-syntactic computation as a phrasal category and would be subject to traditional operations taking place in syntax.

As Acedo-Matellán (2010) points out, the distinction between l-syntax and s-syntax and the fact that Hale and Keyser assume Chomsky’s idea on projectionism in syntax, among some other features, makes this theory closer to endo-skeletal approaches.

Besides the distinction between l-syntax and s-syntax, another important feature of Hale & Keyser’s model is their position regarding θ-roles. They argue that θ-roles are not primitives and that, instead, there are just configurational relations determined by categories and their projections. Given this idea, notions as agent, theme or patient are epiphenomena, and have to be read off of the structural positions that arguments occupy.

3.4 The Distributed Morphology framework

The Distributed Morphology (DM) framework is a theory of grammar first proposed in the nineties by Morris Halle, Alec Marantz and their collaborators (Halle & Marantz 1993, 1994, and much subsequent work); since then it has become very influential.

Within DM, leaving aside many internal differences in concrete proposals, it is argued that syntactical and morphological constituents enter into the same types of structures, such as binary branching trees, which they call “Syntactic Hierarchical Structure All the Way Down.” Another important tenet of this framework is “Late Insertion,” which states that syntactic categories in syntax are completely abstract and do not have phonological content until Spell-Out.

Consequently, DM rejects any lexicalist approach to the lexicon and, indeed, it rejects the very idea of the lexicon in a traditional sense. As argued by Marantz (1997), DM distinguishes three different lists:

(13) List 1, Narrow Lexicon: morphosyntactic features available in the Universal Grammar repertoire, as [+plural] [+2nd person, +plural].

*List 2, Vocabulary*: correspondence rules between phonological exponents and morphosyntactic features, as /i/  <-->  [___, +plural], a Russian affix (Halle 1997).

*List 3, Encyclopedia*: correspondence rules between phonological exponents and sets of world-knowledge properties, as Cat > Fuzzy animal.

The abstract elements of the Narrow Lexicon are the atoms of the generative components, but not their phonological expressions, which are inserted later by means of the vocabulary items stored in the second list following the Subset Principle. Finally, conceptual or encyclopedic elements are inserted.

Regarding morphemes, DM distinguishes two types: f(unctional)-morphemes and l(exical)-morphemes. F-morphemes are deterministic in the sense that their specification suffices to determine a unique phonological expression, that is, there is no choice as for vocabulary insertion. However, l-morphemes are defined as the morphemes for which there are several options in the spell-out stage, as they correspond to lexical pieces as cat or dog.

DM can be classified as a purely exo-skeletal model insofar as it is a framework that considers that all grammatical meaning is derived from the syntactic computation and that conceptual or idiosyncratic meaning is, by definition, idiomatic and opaque for syntactic derivations. Actually, l-morphemes or conceptual elements do not have either inherent category. A root would become a “noun” or a “nominalized” root if its nearest c-commanding f-morpheme is a Determiner, or a “verbalized” root if its nearest c-commanding f-morpheme is v or some other functional morpheme such as Aspect or Tense. Thus the same root would correspond to different grammatical categories depending in the syntactic contexts they appear in.

Regarding argument structure, DM proposals assume Hale & Keyser’s idea: θ-roles are interpreted configurationally, that is, they are not primitives and are not specified anywhere in the verbal entry.

3.5 The Nanosyntax framework

The Nanosyntax framework (Starke 2002, 2005, Svenonius, Ramchand, Starke and Taraldsen (eds.) 2009, and much subsequent work) is an approach to the architecture of grammar that emphasizes a new conception of the lexicon and its relationship to the syntactic component.

One of the starting points of Nanosyntax is the assumption that formal features/submorphemic elements—that is, units smaller than classical morphemes—are the atoms of syntactic computations. It is clear, therefore, that Nanosyntax does not conceive LIs in the traditional sense of “words.” Within this framework, listed items are chunks of syntactic structure built by the computational component with related phonological and semantic expressions. Let us consider an example, in Modern Greek:

(14) Lexical entry: -u



[from Caha 2009:54]

A consequence of this approach is that there is no lexicon that feeds the syntax, the whole system works the other way around: syntax operates with submorphemic units that, once spelled-out, match lexical entries. In Starke’s words “Syntax is an entirely pre-lexical system and the lexicon is a way of interpreting syntax (and mapping it onto other representations—such as conceptual representations and gestural (phonological) representations)” (Starke, 2009:2). Consequently, different LIs have different sizes, that is, they correspond to different amounts of syntactic structure. Once syntactic structures are spelled-out and matched with lexical units, these lexical units can enter again into the derivation.

As Starke suggests (2009, 2011), categorial differences between LIs (and cross-linguistic variation among languages) can be derived under this view from the different sizes of lexical units in particular languages. Nanosyntax is considered, consequently, an exo-skeletal model: LIs correspond to syntactic (sub)trees of submorphemic elements that do not specify either their category or their subcategorization frame.

3.6 Ramchand’s model

The model of the lexicon and grammar by Ramchand (1997, 1998, 2008) has some exo-skeletal and also some endo-skeletal features. On the one hand, Ramchand assumes the basic tenets proposed by Hale & Keyser regarding argument structure and event participants inasmuch as these relations are claimed to emerge from the syntactic configurations in which the elements appear. In that sense, some relevant grammatical meaning is located in syntactic structures, not in the lexicon itself. On the other hand, Ramchand argues in favour of encapsulating some grammatical information in LIs: their selectional/categorial features, for instance, which makes this model divergent from purely exo-skeletal approaches.

The assumption is that, whereas some grammatical notions, such as argument structure, are derived in syntax, the grammatical information that is stored in lexicon sanctions their Merge in specific syntactic positions; that is, the lexicon constrains the way in which listed items can be associated with particular syntactic structures that are built by the computation system.

Another important feature of Ramchand’s model, also present in the Nanosyntax framework, is the assumption that LIs can lexicalize chunks of a tree.
For instance, in this model, the lexical entry for the verb run would contain the following conceptual, grammatical and phonological information:

(15) Run
Conceptual meaning: Continuous directed motion by rapid movement of legs.
Grammatical meaning: [initiation, process]
Phonological information: /r ∧ n/
[from Ramchand 2008: 15 and 74]

The grammatical information would be seen by the syntactic-semantic module in order for the computation to locate this verb in the correct syntactic position, giving as a result a felicitous sentence:4

(16) [initP [ [init run] [procP [ [proc ] ] ] ] ]

From this very general and schematic picture of some current theories dealing with the nature of the lexicon and its relationship with the architecture of grammar, it can be easily observed that differences among approaches are substantial and include fundamental issues about what the structure of the Faculty of Language is.

  1. Some loose ends and how they are addressed in this volume

We hope the previous sections have made clear what the most pressing matter about the lexicon-syntax interactions probably is: How are LIs formed? We hasten to point out that there is a catch in the question, for the wh-word how here is actually a cover term for a cluster of murky issues. Let us briefly review them, emphasizing those that are particularly puzzling from a minimalist perspective.

Consider, to begin with, the point at which LIs are assembled (i.e., before or after syntax), a matter that has received much attention in the recent literature (cf. Halle & Marantz 1993, Starke 2010, and related Distributed Morphology and Nanosyntax work). The traditional view, endorsed by some current approaches, takes it that LIs are complexes of information of the sort in (17) for the words tree, love, and after, with phonologic, semantic, and syntactic chunks of information attributed to each.

(17) a. tree: b. love: c. after:

   /tri:/                /lʌv/                 aftəʳ
      [+N, -V]              [-N, +V]              [-N, -V]
      [+count]              [+transitive]                 [+NP / CP]

The properties represented in (17) are easy to figure out, and as noted, they point to a word-internal complexity. The problem here is simple: these approaches typically do not explain how complexity gets off the ground. The interest in finding out the internal stuff of words goes back at least to European structuralism, which invoked distinctive features in order to capture phonological and morphological oppositions. The idea of decomposing linguistic units into smaller parts was quickly adopted to analyze functional vocabulary, and soon enough proposals emerged that took pronouns, affective operators, wh-words, etc., to be chunked down into parts for which there was no phonological correspondence (cf. Kayne 2005 for related discussion on invisible units).

Departing from the Saussure-Bloomfield-based picture in (17), more recent perspectives regard the notion of LI as an epiphenomenon—an artifact that arises from the conjunction of information from different sources. According to some, such sources are post-syntactic (Starke 2010), according to others, they are both pre- and post-syntactic (Halle & Marantz 1993), a decision with far-reaching consequences for the Y Model advocated since Chomsky & Lasnik (1977).

(18) Lexicalist architecture Non-lexicalist architecture

    Lexicon                 Narrow 
                        Syntax

????                Features

    Narrow                  Lexicon
    Syntax

The well-established idea that LIs have internal structure is also problematic if these objects are regarded as “atoms of computation” (Chomsky 2007, 2013), that is, units whose integral components are opaque to regular syntax. The “complex-but-at-the-same-time-atomic” status of LIs has in fact been at the crossroads of different linguistic debates, most of which are very much alive today. Let us refer to three of them here.

First and foremost, the internal composition of LIs was a key issue of the Generative Semantics debates, whose proposals were the source of much current work on the lexicon (cf. Hale & Keyser 1993, 2002, Kayne 2005, Marantz 1997, 2001, Mateu 2002, Ramchand 2008, Starke 2010, among others), at the bottom of which two questions were particularly pressing: Can every aspect of meaning be syntactiziced? Can paradigmatic relations be reduced to syntagmatic ones? (cf. Uriagereka 2008).

A primary goal of the Generative Semantics approaches to the lexicon was to determine the most basic units of interpretation—undecomposable linguistic primitives, be these distinctive features (e.g., [±N], [±animate], [±count], [±human]), or abstract conceptual predicates (e.g., CAUSE, BECOME, HAVE, IN)—and how these may help us formulate a theory of “possible LI.” Several authors have devoted their work to accounting for the fact that certain LIs are not possible with certain meanings, a situation that is usually attributed to the idea that syntactic principles apply to both syntactic objects and words. This is the idea that comes out of Hale & Keyser’s (1993) work:

We have maintained that […] certain gaps in the lexicon can, we think, be explained on the assumption that the formation of the lexical items in question is subject to principles known to be operative in syntax. If this is true, it follows that the structures over which lexical derivations are defined are true syntactic structures, over which syntactic relations are defined.  [from Hale & Keyser, 1993:64]

A second focus of inquiry on the lexicon-syntax interface is linguistic variation. More specifically, the nuances that derive from lexicon-syntax interactions are deeply embedded in micro-parameters and the so-called Borer-Chomsky Conjecture (BCC), namely the idea that variation is restricted to the way functional features are bundled in language-particular LIs. In Chomsky (2000, 2001), the lexicon is thought of as a repository of units built by using the features that UG provides. In particular, Chomsky (2000, 2001) assumes that the creation of a language-particular lexicon consists of three steps.

(19) a. UG provides a set of features ({F}) and operations (Merge, Agree, etc.)

b. A language L selects [F] from the universal set {F}

c. L assembles the members of {F} to create a lexicon  
                         [from Chomsky, 2000:100-101]

The first step in (19) concerns UG itself, whereas the second and third steps have to do with the specifics of the language-acquisition period (the children’s exposure to data). Although things are intuitively clear, what the members of {F} are (recall the discussion in section 2), and how they are bundled to form LIs has always been obscure. It still is. The list of distinctive features and their role in syntax is a matter of much ongoing debate, for unlike the situation in the domain of phonology, current morpho-syntactic theories have not told us what the list of distinctive features is—nor even what a “possible UG feature” should look like (putting aside roots). Proposals for and against features abound in the literature, with different perspectives on the matter (cf. Adger & Svenonius 2011, Corbett 2010, Rizzi 2004). Despite such an abundance of proposals, things do not seem to have moved much beyond the classification of Aspects.

This brings us back to one other ill-understood point: the manner in which features are put together. Two reasonable candidates come to mind: either “Merge” is responsible for giving rise to the complexity of LIs, or a variant of it is—which we could call “Bundle” (following Chomsky 2000). The tricky part of this new operation, of course, is that it would operate outside of the syntax (whether before or after is irrelevant to our purposes now), which amounts to having two sources of generative power. A way to overcome this duplicity would be to assume that Merge operates on LIs, whereas Bundle operates on features, but this would not work either, since we would then have two combinatory devices. Interestingly enough, a proposal close to this one was actually put forward in the context of Chomsky’s (1995) Move F approach to feature-checking, whereby features (particularly, agreement features) could move to the relevant functional heads, prior to pied-piping:

(20) a. [ Tφ [ . . . XP-tφ . . . ] ] Move F(eature)
↑________↑
b. [ XP [ Tφ [ . . . tXP-tφ . . . ] ] ] Pied Piping

However worth considering (and useful for cases with agreement without overt displacement), Chomsky (1995) ended up dismissing the Move F option, in part because of the complexity it would imply. Once equipped with Move F, the computational system would not only deal with LIs, but also with features, to which the Xmin/max status seems not to apply.

Let us now investigate what happens if we replace the operation Move α by the more fundamental operation Move F, F a feature […] *We now extend the class of syntactic objects available to the computational systems* […] In earlier discussion we kept to the case where α, β are lexical items or larger phrases constructed from them, but we have now been considering a more general case, with variables allowed to range over features as well. Specifically, we allow an object L = {γ, {F, K}}, F a feature, formed by raising F to target K without pied-piping of a category α. Several questions arise, including the following: (a) Can the operation be substitution? (b) Must the target project? (c) Can K be a feature rather than a category? The answers depend on how we interpret such notions as “Xmax” and “head,” which so far have been defined only for phrases constructed from lexical items, not for features. But these notions have no clear sense for features.[from Chomsky, 1995:262- 270, our emphasis]

One final issue about the lexicon-syntax interactions is important for the morpho-phonological component: the insertion of vocabulary items. Traditionally, this process could take place before or after the computation is activated, and it always targeted a terminal position (a “head,” in standard X-bar terminology). In more recent frameworks, lexical insertion can also take place in a phrasal fashion, occupying more space than that a head encodes.

(21) a. Lexical insertion b. Phrasal Spell Out

        the
                                    -ing
        α                           α

            nice                                

    β       γ               β       γ

                   book                   √GROW

        λ       δ               λ   δ
    the nice book                       growing

The options in (21) may look similar, but they are not. As the reader can see, the Phrasal Spell Out option actually hides a different conception of morphology and cyclicity, for heads can be submorphemic and the Spell Out points (the cycles) can vary from language to language. Whatever the details of Phrasal Spell Out, it is certainly a good option to recast both head movement (an operation that has been involved in different parameters and whose status has recently been seen as problematic; cf. Chomsky 2001) and bundling (the feature compiling process we have referred to above).

The goal behind any of these processes (Phrasal Spell Out, head movement, bundling) is to be able to capture the fact that one and the very same feature F may remain as such in a language L (that is, buried within word-internal margins) or as a regular LI in a language L', being able to project X-bar structure:

(22) a. Language L: [CP CF [TP T . . . ] ]

b. Language L': [CP C [FP F [TP T . . . ] ] ]

Many parameters, both micro and macro, have been approached from this perspective (cf. Biberauer 2008, Biberauer et al. 2010, and references therein), but researchers are still unclear about the effect of this approach for the conception of the lexicon, and its bearing on syntax.

Although it may not be enough to leave the discussion here, it seems to us that we have quickly overflown the most pressing issues surrounding the interactions between the lexicon and the computational system. To be sure, such issues concern old linguistic chestnuts—the feature inventory, the sources of generative power, and linguistic variation—and they all have to do with the question how (and where, with what, etc.) LIs are formed. The papers in this volume discuss some of these matters in detail, providing us with new tools and arguments to reach answers.

Richard S. Kayne’s contribution focuses on linguistic variation by considering how closely related the structures I want for John to arrive earlier and John was to arrive earlier are, and why they are not possible in other (Romance) languages. As Kayne observes, descriptive adequacy in the case of comparative syntax involves discovering generalizations over cross-linguistic differences and similarities, which can in turn provide us with new kinds of evidence bearing on questions concerning the general character of the language faculty. In this particular case, Kayne argues for the hypothesis that languages that deploy prepositional complementizers, like that in I want for John to arrive earlier, can resort to silent functional material (i.e., without a corresponding vocabulary item) with the import of MEANT / SUPPOSED in examples like John was to arrive earlier.

In the preceding pages, we have seen that the idea of syntactic terminals being subject to a process of lexical insertion is fairly standard. In lexicalist approaches such insertion occurs very soon, while in distributed ones it is delayed (although see Artemis Alexiadou’s review of Matushansky & Marantz’ 2013 collection of papers for a more comprehensive view). A significant aspect of the late insertion approaches is whether vocabulary items are inserted in syntactic terminals (Xº positions) or whether they can span over various syntactic nodes (XPs) in so-called Phrasal Spell Out.

Antonio Fábregas’ paper provides evidence in favor of Phrasal Spell Out taken from morphological haplology phenomena in Spanish. As Fábregas argues, some nominal affixes in Spanish, like -idad ‘-ity’ in solidaridad (Eng. solidarity), are not morphemic, but phrasal in nature; they spell out an [ FP [ NP ] ] structure. Fábregas argues that deadjectival versions involve a configuration where the AP has to be sandwiched between FP and NP, which has the effect of disrupting FP-NP adjacency: [ FP [ AP [ NP ] ] ]. In such a scenario, another exponent, -ist(a) (Eng. ist), has to be used. The surface result is that -idad is haplologised when -ist(a) is introduced, even though the morphosyntactic features of -idad are still present in the structure.

Given that LI complexity is conceptually unavoidable, what must be decided is how such complexity is built and what its ingredients are. Although much remains to be done in this area, a much agreed upon idea stemming from Pesetsky (1995) is that LIs are the outcome of combining two types of elements: lexical morphemes (roots) and functional morphemes. We have just seen how Kayne and Fábregas deal with some instances of functional item insertion, but the details of root insertion are also subject to debate in this volume, and two papers are devoted to them. Víctor Acedo-Matellán and Cristina Real-Puigdollers put forward a new approach to root insertion based on two tenets: first, root insertion is strictly late, where the syntax manipulates exclusively functional material; second, roots can be inserted directly into syntactic terminal nodes, by the very same mechanism as other Vocabulary Items. These authors discuss how root insertion takes place, either at the first nodes merged in the derivation, the categorial heads (v, n, a), or higher up. They further explore the consequences of those options, which matter for the nature of categorizers in natural language and well-known case of cross-linguistic variation at the lexicalisation level (Talmy 1991, 2000).

A second contribution dealing with the functioning of roots and insertion is Jeroen van Craenenbroeck and Marijke De Belder’s, where a unified insertion mechanism for functional and lexical Vocabulary Items is defended building on a previous approach to Merge put forward by the same authors. Assuming the existence of two different types of terminal nodes (roots and functional terminal nodes), van Craenenbroeck and De Belder argue that the lexicon contains two types of VIs: lexical VIs and functional VIs. In their paper, they explore whether it is possible for VIs to be matched with terminal nodes chiastically, that is, in a way that is the opposite of the way that they are generally taken to be inserted. On the one hand, van Craenenbroeck and De Belder show that, against the majority position in which functional VIs can be inserted across the board, but lexical VIs are restricted to root terminal nodes, the latter can occupy root terminal nodes, too. On the other hand, they also examine whether it is possible for lexical VIs to realize functional terminal nodes, their answer being negative. The overall consequences of these findings is that the traditional distributed approach to VI insertion is too conservative, for it prevents functional VIs from realizing root terminal nodes. In order to tackle such a scenario, the authors propose a unified insertion mechanism where all VIs (lexical and functional alike) are inserted via competition, through a revised version of the Subset Principle.

One final loose end with respect to roots—their purely conceptual nature—is addressed by Gillian Ramchand. Ramchand puts forward the idea that structural meanings are correlates of a hierarchically structured representation of abstract actional factors that give rise to linguistic generalizations concerning the realization of argument structure in the syntax. As for conceptual meanings, Ramchand suggests that they consist of the encyclopedic and conceptually rich information that provides detailed expression to highly specific named events. By doing this, Ramchand’s article seeks to provide specific answers to the following two questions: (I) The Separation Problem: which meaning components are part of the structural meaning skeleton and which are not? and (II) The Lexical Specification Problem: does a lexical root carry only information of the conceptual type, or does it contain information from both domains?

To conclude the volume, Michal Starke’s short but thought-provoking contribution is intended to show that the problems associated to traditional lexicalist views can be solved by adopting a nanosyntactic approach to the lexicon, which can be summarized as follows:

(23) The (syntactic) lexicon contains nothing but well formed syntactic expressions

The formulation of (23) entails adopting an entirely post-syntactic view of the lexicon whereby the syntactic engine is fed by a pre-syntactic inventory of features that assemble feature structures which are then subject to Phrasal Spell Out. Starke’s model pressuposes the existence of a lexicon where VIs are stored. As the syntactic derivation unfolds (creating trees of different size), the lexicon is checked in order to determine whether a relevant feature structure can be associated with a specific VI—and if so, lexicalization (via Spell Out) occurs. Starke further shows how this new conception of the lexicon can be exploited in order to account for subcategorization, irregularities, and other defining properties exhibited by the lexicon.

Works Cited

  1. Acedo-Matellán, V. 2010. Argument Structure and the syntax-morphology interface. A case study in Latin and other languages. PhD Thesis, Universitat de Barcelona.
  2. Adger, D. and P. Svenonius. 2011. Features in Minimalist Syntax. In C. Boeckx (ed), The Handbook of Linguistic Minimalism, 27-51. Oxford: Blackwell.
  3. Biberauer, T. (ed) 2008. The Limits of Syntactic Variation. Amsterdam: John Benjamins.
  4. Biberauer, T. et al. (eds) 2010. Parametric Variation: Null Subjects in Minimalist Theory. Cambridge: CUP.
  5. Borer, H. 1994. The Projection of Arguments. In E. Benedicto and J. Runner (eds), University of Massachussetts Occasional Papers in Linguistics 17. Amherst, MA: University of Massachusetts.
  6. ––. 1998. Passive without θ-Grid. In S. Lapointe et al (eds), Morphological and Its Relations to Phonology and Syntax, Stanford University: CSLI.
  7. ––. 2003. Exo-skeletal vs. Endo-skeletal Explanations: Syntactic Projections and the Lexicon. In J. Moore and M. Polinsky (eds), The Nature of Explanation in Linguistic Theory. CSLI Publications.
  8. Caha, P. 2009. The Nanosyntax of Case. PhD Thesis, University of Tromsø.
  9. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
  10. ––. 1970. Remarks on Nominalization. In Readings in English Transformational Grammar. Waltham, MA: Ginn and Company.
  11. ––. 1981. Lectures on Government and Binding: The Pisa Lectures. Dordrecht: Foris.
  12. ––. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
  13. ––. 2000. Minimalist Inquiries: The Framework. In R. Martin et al (eds), Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, 89-155. Cambridge, MA: MIT Press.
  14. ––. 2001. Derivation by Phase. In M. Kenstowicz (ed), Ken Hale: A Life in Language, 1-52. Cambridge, MA: MIT Press.
  15. Chomsky, N. and H. Lasnik. 1977. Filters and control. Linguistic Inquiry 8: 425-504.
  16. Corbett, G. 2010. Features: essential notions. In A. Kibort and G. Corbett (eds), Features: Perspectives on a Key Notion in Linguistics, 17-36. Oxford: Oxford University Press.
  17. Grimshaw, J. (1990). Argument Structure. Cambridge, MA: MIT Press.
  18. Hale, K. and S. J. Keyser 1992. The syntactic character of thematic structure. In I. Roca (ed), Thematic Structure: Its Role in Grammar. Dordrecht: Foris.
  19. Hale, K. and S. J. Keyser 1993. On argument structure and the lexical expression of syntactic relations. In K. Hale and J. Keyser (eds), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press.
  20. Halle, M. and A. Marantz 1993. Distributed Morphology and the pieces of inflection. In K. Hale and J. Keyser (eds), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press.
  21. Halle, M. and A. Marantz. 1994. Some key features of Distributed Morphology. In A. Carnie and H. Harley (eds), MIT Working Papers in Linguistics 21. Papers in Phonology and Morphology. MIT Working Papers in Linguistics.
  22. Harley, H. 1995. Subjects, Events, and Licensing. PhD Thesis, MIT.
  23. van Hout, A. 1996. Event Semantic of Verb Frame Alternations: A case study of Dutch and Its Acquisition. PhD Thesis, Tilburg University.
  24. Jackendoff, R. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.
  25. ––. 1990. Semantic Structures. Cambridge, MA: MIT Press.
  26. Kaplan, R. and J. Bresnan. 1982. Lexical-Functional Grammar: A formal system for grammatical representation. In J. Bresnan (ed), The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press.
  27. Kayne, R.S. 2005. Movement and Silence. Oxford: Oxford University Press.
  28. Levin, B. and M. Rappaport Hovav. 1995. Unaccusativity: At the Syntax-Lexical Semantic Interface. Cambridge, MA: MIT Press.
  29. Marantz, A. 1996. Cat as a Phrasal Category. Unpublished Ms., MIT.
  30. Marantz, A. 1997. No Escape from Syntax. In A. Dimitriadis et al (eds), University of Pennsylvania Working Papers in Linguistics. Philadelphia: UPenn Press.
  31. Mateu, J. 2002. Argument Structure. Relational Construal at the Syntax-Semantics Interface. PhD Thesis, Universitat Autònoma de Barcelona.
  32. Matushansky, O. and A. Marantz. 2013. Distributed Morphology Today: Morphemes for Morris Halle. Cambridge, MA: MIT Press.
  33. Ramchand, G. 1997. Aspect and Predication: The Semantics of Argument Structure. Oxford: Oxford University Press.
  34. ––. 1998. Deconstructing the lexicon. In M. Butt and W. Geuder (eds), The Projection of Arguments: Lexical and Syntactic Constraints. CSLI Publications.
  35. ––. 2008. Verb Meaning and the Lexicon. A First-Phase Syntax. Cambridge, UK: Cambridge University Press.
  36. Rappaport Hovav, M. and B. Levin. 1998. Building verb meanings. In M. Butt and W. Geuder (eds), The Projection of Arguments: Lexical and Syntactic Constraints. CSLI Publications.
  37. Rizzi, L. 2004. Locality and left periphery. In A. Belletti (ed), Structures and beyond. The cartography of syntactic structures (vol. 3), 223-251. Oxford, NY: Oxford University Press.
  38. Starke, M. 2002. The day syntax ate morphology. Class taught at the EGG summer school, Novi Sad, Serbia.
  39. ––. 2005. Nanosyntax. Seminar taught at CASTL, University of Tromsø, Norway.
  40. ––. 2009. Nanosyntax. A short primer approach to language. In Peter Svenonius et al (eds), Nordlyd 36.1, special issue on Nanosyntax. CASTL, Tromsø, Norway.
  41. ––. 2011. Towards an elegant solution to language variation: Variation reduces to the size of lexically stored trees. Unpublished Ms., CASTL, Tromsø, Norway.
  42. Svenonius, P., G. Ramchand, M. Starke, and T. Taraldsen (eds). 2009. Nordlyd 36.1, special issue on Nanosyntax. CASTL, Tromsø, Norway.
  43. Uriagereka, J. 2008. Syntactic Anchors. On Semantic Structuring. Cambridge, UK: Cambridge University Press.