Journal of Language and Linguistics Volume 2 Number 2 2003 ISSN 1475 - 8989

1. Introduction

The role of Lexical Conceptual Structure (LCS; Jackendoff 1990, 2002, Rappaport & Levin 1988, 1992) in derivation has received little attention. Recent work (Plag (1997), Lieber & Baayen (1993, 1998) and Lieber (1992, 1998)) proposes analyses of the interaction between the LCS of a base (e.g. modern) and an affix (e.g. -ize) in English derivatives (e.g. modern+ize). However, those models represent the base and affix LCSs as a merged unit (e.g., MODERN + -IZE: [EVENTCAUSE ([THING ], [EVENTBE (LCS of MODERN)])]; Lieber (1992)), without showing the process that led to the merger.

This paper presents an LCS-based, concatenative model that explains the process of [base+affix] merging in regular derivational suffixation in Spanish (e.g. demoler+cion [demolición 'demolition']). It is a model that incorporates semantic features (e.g. CAUSE, [+animate]) and is crucially based on the notion of lexical entries instantiated in a slot structure, as well as on the proposal that percolation manipulates the contents of the LCS of bases and affixes. Employing the mechanisms of subcategorization/selection (henceforward "subcat/select") and percolation, already available in the generative framework (cf. Lieber 1992, 1998), the model unifies, in a principled way, all the processes that take place during the formation of a complex word (e.g. plega+ble [fold+able] 'foldable') in a single representation.

Crucial to the model is that both percolation and subcat/select, as well as slot structure, take into account not only syntactic but also semantic information, including information about argument structure. Special emphasis is placed on this incorporation of semantic information into the representation of morphological processes. Furthermore, arranging features in a slot structure allows predictions to be made about the behavior of groups of features during and after derivation. The slot structure of bases and affixes as proposed in this study is a crucial factor in the derivational process because it helps determine the structure of the lexical entries of derivatives.

The present model reduces the machinery of the grammar by proposing only two morphosemantic1 mechanisms (percolation and subcategorization/selection) that depend on each other and act in concert. Additionally, several principles, conditions and constraints that have been proposed in the morphological literature (e.g. the Unitary Output Hypothesis (Scalise 1984) and the Word-Based Hypothesis (Aronoff 1976)) are incorporated in the mechanisms of percolation and subcategorization/selection.

In addition to Spanish derivational suffixation and prefixation (e.g. re+construir [re+construct] 'reconstruct'), the model adequately accounts for the regular derivational morphology of several languages genetically unrelated to Spanish, namely Mam, Turkish, and Swahili, which suggests that the notions of percolation, subcat/select, slot structure and the LCS are universal constructs.

The study focuses on regular derivation because of the rule-based nature of regular derivation as opposed to irregular derivation. As pointed out by Pinker (1999), there is considerable evidence that while regular forms (e.g. work+er) are computed by combinatorial rules, irregular or unpredictable forms (e.g. cranberry, raspberry, strength, salut+at+ion vs. *salut+ion) have to be memorized and are stored in a sort of analogical network. In this study, it is the combinatorial rules referred to by Pinker that are explored.

The present study investigates the semantics of derivation using data from contemporary Spanish,2 a language with a particularly rich derivational morphology. A total of 1250 derivatives formed with close to fifty productive Spanish derivational suffixes (approximately seventy suffixes if homonyms are taken into account) were analyzed. It must be clarified that even though Spanish data were used to explore the subject, the purpose of the paper is not only to shed light on Spanish morphology, but, as mentioned above, to model the mechanisms that operate in regular derivation across languages. Thus, Spanish has been employed here additionally as a vehicle to discover universal principles.

The model is presented in several parts throughout the article. In section 2 I present a reasoned justification for incorporating semantics in the morphological process, as well as a detailed description of the notion of slot structure. In section 3, the modified feature percolation conventions, a central part of the model, are discussed. Section 4 illustrates the model with tree representations of sample derivations. The application of the model to affixes other than derivational suffixes, as well as to derivation in languages unrelated to Spanish, is also discussed in section 4. The concluding remarks are presented in section 5.

2. Lexical Entries

As is shown in sections 3 and 4, arranging features in a slot structure (as described below), rather than just listing the features (in a lexical entry), allows predictions to be made about the behavior of groups of features during and after derivation. It will be seen that the slot structure of bases and affixes as proposed here is a crucial factor in the derivational process because it helps determine the structure of the lexical entries of outputs. Thus, the morphosemantic analysis presented in this proposal hinges on the notion of a highly organized slot structure.

2.1. Slot Structure

In the present model, the information contained in lexical items, namely ontological and syntactic category (e.g. EVENT and V, respectively),3 semantic features (e.g. [+human], Location),4 subcategorization frames, and selectional frames, is organized into groups of features that act as information blocks that percolate as units to the branching node. This arrangement of blocks of information located within their respective slots constitutes the "slot structure" of each lexical item (see sections 3 and 4 for examples and details).

It is generally assumed (cf. Grimshaw 1990) that verbs, adjectives and some nouns have argument structures. Since these lexical items subcategorize and select for arguments, they must have, in addition to blocks that contain their ontological, categorial, and core semantic information, blocks that contain syntactic subcategorization and selectional information. Suffixes that contribute arguments to an output must also contain syntactic subcategorization and selectional blocks aside from their morphological subcategorization and selectional blocks. Inasmuch as verbs, adjectives, nouns and suffixes differ idiosyncratically in their featural content and argument structure, they must also differ in their slot structure. Thus, the idiosyncratic information contained in lexical items is what determines their slot structure.

2.2. Lexical Conceptual Structure (LCS) and Slot Structure

The idea of a slot structure containing idiosyncratic information is compatible with the notion of an LCS. The LCS is the place in the lexical entry of an item where the syntactically relevant semantic content of the item is encoded (cf. Rappaport & Levin 1988, Jackendoff 1983, 1990, Speas 1990). The LCS is defined as the decomposition of the meaning of a word into conceptual primitives (e.g. primitive predicates such as CAUSE and GO) which are related to arguments that occupy slots and are also characterized by means of conceptual primitives (e.g. ontological categories such as [THING]). As Kornfilt & Correa (1993) assert, the LCS captures the core aspects of the meaning of a lexical item (that is, its core meaning), not the whole range of meaning associated with the item (i.e., encyclopedic information is not included in the LCS). For example, (1) is the LCS of the verb put.

(1)	LCS of put
	PUT: [EVENTCAUSE ([THING ], [EVENTGO ([THING ], [PLACEAT [PLACE ]])])]
	[adapted from Jackendoff 1990]

The organization of lexical entries as slot structures proposed in this study is a natural extension of formulations of the LCS such as in Rappaport & Levin (1988), where arguments occupy slots, as well as of Grimshaw's (1990) use of slots that store argument and aspectual information (see section 2.4.3). If some lexical information (i.e. argument structure) is stored in slots, it is not implausible to propose that all (non-encyclopedic) lexical information, including conceptual primitives, may be stored in such slots as well. Since non-semantic information (e.g. categorial features) is stored in lexical entries as well, it is natural to assume that this information is also stored in slots. Thus, slot structure contains the LCS and other idiosyncratic information stored in lexical entries.

2.3. Relation between (Lexical) Semantics and Morphology

In the literature, the role of semantics in morphology is usually treated separately from morphology per se. Lieber (1992), for example, explains the semantics of the derivational process using the LCS alone as the basis. When an affix attaches to a base, the LCSs of the concatenating morphemes merge and become a single LCS for the whole derivative, as in (2-3).

(2)	LCS of modern+ize
	MODERN + -IZE: [EVENTCAUSE ([THING ], [EVENTBE (LCS of MODERN)])]
	[Lieber 1992]
(3)	LCS of hospital+ize
	CAUSE ([ ], GO ([Thing ]Theme; [TO [Thinghospital]Base])])
	[adapted from Plag 1997]

The affix adds a layer of meaning in the form of semantic primitives (e.g. CAUSE in (2-3) and GO in 3) and open slots to the LCS of the base. In certain cases, the affix deletes (or suppresses) semantic primitives and slots from the LCS of the base (cf. Plag 1997, Lieber & Baayen 1993, and Lieber 1998 for a more elaborate representation of the meaning of the suffix -ize and its Spanish equivalent -izar, and a more detailed representation of the derived LCS). Thus, argues Lieber, changes in argument structure are the result of changes in the LCS, and have nothing to do with percolation, a morphological process (see sections 2-4). However, in sections 3 and 4 it is shown that there is explanatory power in the proposal that the elements of the LCS are manipulated by percolation.

The separation of semantics (the LCS) from morphology in derivation advocated by Lieber (1992) is undesirable given that the semantics figures so prominently in morphological processes. There are four compelling reasons for not separating lexical semantics from morphology, and thus for viewing these two modules as interacting in what can be considered the interface component of morphosemantics.

First, the selectional restrictions of affixes significantly limit the number of well-formed derivatives and are thus a needed constraint on derivation, as the examples with affixes in both English and Spanish in (4) illustrate.

(4)	Selectional restrictions of affixes
	a.	re+paint (change of state)
	b.	*re+punch
	c.	un+optimistic (positive connotation)
	d.	*un+pessimistic
	e.	ladr+ido 'barking' < ladrar 'to bark' (sound emission)
	f.	*camin+ido 'walking' < caminar 'to walk'
	g.	pian+ista 'pianist' (musical instrument)
	h.	*oll+ista 'cooking pan player' < olla 'cooking pan'
	i.	barb+udo 'heavily bearded' < barba 'beard' (body part)
	j.	*botell+udo 'having a large bottle' < botella 'bottle'
	k.	botell+azo 'blow w/a bottle' < botella 'bottle' (solid object)
	l.	cod+azo 'blow w/the elbow' < codo 'elbow' (solid object/body part)
	m.	*concept+azo 'blow w/a concept'

Notice, for example, how the ungrammatical (4b) suggests that the prefix re- selects for bases that involve a change of state (4a). Likewise, the Spanish suffix -azo selects for solid objects (some of which could be body parts; 4k, l), but not for bases denoting abstract entities (such as concepto 'concept'; 4m).

Second, the application of constraints on the operation of Word Formation Rules (WFRs) such as the Atom Condition and the Adjacency Condition, which have been proposed within the theory of generative morphology, suggest that semantic features percolate and are crucial for making decisions about affix attachment. The competition between these two conditions lends credibility to the hypothesis that derivational affixes contain semantic features that percolate to the branching node, and therefore also to the larger hypothesis that morphology and semantics operate in concert during the derivational process.

The Adjacency Condition is formulated by Siegel (1977:192) as in (5).

(5)	Adjacency Condition No word filter may involve A and SP unless SP is uniquely contained in the cycle adjacent to A. (A=Affix; SP=Substring Property=Feature)

What (5) means is that affix attachment is sensitive to the features of the most recently attached affix. Note that the condition refers to cyclic attachment rather than to strictly positional adjacency. That is, for an affix to be sensitive to the features of another affix, the affixes need not be contiguous. A first affix (be it a prefix or a suffix) must simply attach in a cycle immediately before a second affix (see examples below).

Williams (1981:253) formulates the Atom Condition as in (6).

(6)	Atom Condition A restriction on the attachment of affix to Y can only refer to features realized on Y. (Y=a derived or underived base containing head features)

(6) asserts that affix attachment is sensitive to the features of the head of a word.

When the first affix attaching to a base is a suffix (-able in (7-8)), that suffix is, in relation to a second affix being attached, both the most recently attached affix (Adjacency Condition) and the head (Atom Condition), so both conditions are satisfied; that is, both conditions make the correct prediction and therefore agree. (In the examples that follow, solid arrows denote morphosyntactic feature percolation, while broken arrows (see (10)) indicate semantic feature percolation.).

(7) Both Conditions agree5

(8) Both Conditions agree

On the other hand, since non-category-changing prefixes do not carry any morphosyntactic features to percolate, if a non-category-changing prefix is the first affix attaching to a base (un- in (9)), the prefix is the most recently attached affix (Adjacency Condition), but the base (happy in (9)) is the head of the word (Atom Condition).

(9) Atom Condition "wins"

In such cases, the last affix to attach (-ness) is sensitive to the head of the word (happy) and not to the most recently attached affix (un-), so the Adjacency Condition is violated and the Atom Condition "wins" because it makes the correct prediction.

However, there are cases, such as the anomalous derivative in (10), where even though the most recently attached affix (re-) is a non-category-changing prefix, the next affix to attach (un-) is sensitive to that prefix (re-), and not to the head (hook).

(10) Adjacency Condition "wins"

I argue that, for semantic reasons, the Adjacency Condition "wins" in such cases. Since re-, the first prefix to attach, does not carry morphosyntactic features, an explanation for why the second prefix to attach, un-, is sensitive to the first one, re-, is that the second prefix is sensitive to the semantic features of the first prefix, features which express "repetition" and apparently percolate to the branching node (shown with a broken arrow in (10)). It could well be that, given the grammatical un+hook, un- has a selectional restriction stating that it cannot attach to bases with a meaning of "repetition," such as re+hook. A similar case is seen in *trans+inter+oceanic. Given the grammatical trans+oceanic, it stands to reason that trans- (the last affix to attach) is sensitive to the semantic features of the most recently attached affix (inter-). Notice that the number of prefixes attaching cannot be the relevant constraint, since there are grammatical derivatives with more than one prefix (e.g. anti+dis+establishmentarian).

Thus, the difference that arises between the Atom and Adjacency Conditions in cases such as (10) is due to the separation of semantics from morphology. If semantics is taken into account, that is, if not only subcategorization (7-9) but also selection is taken into account, then all structures, including structures such as the one in (10), will be predictable. In other words, a theory that takes into account both subcategorization (syntactic features) and selection (semantic features) is able to account for the grammaticality or ungrammaticality of any regular morphological construction.

Finally, it is necessary to emphasize the importance of percolation. If percolation were not to operate as part of the Atom and Adjacency Conditions (see arrows in (7-10)), neither the features of bases nor of affixes would percolate to the relevant nodes where decisions about attachment (subcat/select) need to be made.

Consider another constraint on WFRs, Aronoff's (1976:21) Word-Based Hypothesis (WBH), whereby "all regular word-formation processes are word-based. A new word is formed by applying a regular rule to a single already existing word. Both the new word and the existing one are members of major lexical categories."6 Expressed in concatenative terms, the WBH reads as in (11).

(11)	Word-Based Hypothesis (WBH-concatenative version)
	(a)	all regular word-formation processes are word-based;
	(b)	a new word is formed by attaching a single affix to a single already existing word (a base);
	(c)	both the new word and the existing base word are members of major lexical categories.
		[adapted from Aronoff 1976]

The required attachment of a single affix to a base (11b) is called the "one affix, one rule" hypothesis by Booij (1977). Since the present is a concatenative model, I will leave out the word "rule" and call this the "single-affix attachment principle". The WBH (11), including the single-affix attachment principle, can be encoded in the subcat/selectional information of affixes. Thus, in a concatenative model subcat/select determines that an affix may subcategorize for only one base; only one affix at a time may attach to a base.

Since, as mentioned above, the mechanism of subcat/select must specify that an affix is sensitive to both the syntactic and semantic features of the base, a formal statement of the principle that underlies subcat/select can be formulated as in (12). Within the present model, this principle will determine the content of the SUBCAT/SELECT slot (see section 2.4.5).

(12)	Principle of Subcategorization/Selection The subcategorization/selectional frame of each affix encodes the following information: The affix attaches to a single word that contains a specific set of syntactic and semantic features.

The set of "syntactic and semantic features" that is specified for each suffix consists of idiosyncratic information that forms part of suffixes in their condition of (bound) lexical items.

It is important to recall that in this model percolation itself presupposes subcat/select; if there were no information for subcat/select in a suffix, there would be no instructions on what elements to concatenate, and therefore no concatenation at all. If no concatenation occurs, then percolation cannot operate. This is further evidence of the importance of subcat/select for morphosemantic theory.

This leads to the third reason for keeping semantics and morphology working in concert. Since the mechanisms of subcategorization and percolation are already employed for the manipulation of morphosyntactic features, it is advantageous to unify their function and use them to manipulate semantic features as well. The use of semantic subcategorization (i.e. selection) makes unnecessary a separate set of semantic rules (such as the selectional rules proposed by Lieber 1980) to account for regular derivation. And, as sections 3 and 4 show, percolation can handle not only morphosyntactic features but also semantic features and information about argument structure (cf. Marantz 1984). This use of subcategorization and percolation to handle syntactic and semantic information as well as argument structure results in a reduction in the machinery of the grammar in comparison with models which separate syntactic from semantic information. Such models (e.g. Beard 1995, Aronoff 1976, Lieber 1992) must have three separate sets of rules: those that handle subcategorization, those that deal with selection, and those that manipulate the LCS and argument structure.

Finally, since Scalise's (1984) Unitary Output Hypothesis (UOH) assumes that the semantic features contained in the lexical entry of an affix percolate to the branching node in a derivative, the UOH provides yet another argument for viewing morphology and semantics as acting in concert. Recast in concatenative terms, the UOH can be seen as being encoded in the semantic features of an affix if these features are allowed to percolate to the branching node. In concatenative terms, the UOH holds that for an affix to be considered a single, distinct item, it must produce an output with a unique category or meaning. Since in the concatenative model developed here an affix transfers its semantic features (along with the categorial ones) to the derivative as a whole by percolation, the output semantic content of each derivative is made unique by the distinct semantic features of each individual affix.

Consider derivatives with the Italian prefix s-, which attaches to verbs. The s- in s+bloccare 'to unblock' produces an output with a different meaning (although with the same category) from the s- in s+gocciolare 'to drip' because with the first prefix (13a) "negation" is the semantic content that percolates (from the prefix), while with the second prefix (13b) the percolating meaning is "repeated action." (The percolation of semantic features is represented by broken arrows in the tree diagrams in (13)).

(13) The Unitary Output Hypothesis (UOH)

Since each affix in (13) has unique semantic information that percolates, each affix produces an output with a unique meaning, and therefore the affix s- should be considered two different (homonymous) lexical items, in accordance with the UOH.

Since affix homonymy can be determined by means of percolation (13), the UOH may be incorporated into the mechanism of percolation (i.e. we do not need the UOH as such to account for homonymy). This incorporation allows percolation to account for homonymy, and provides a potential argument for viewing derivation as a process in which the modules of morphology and lexical semantics interact in concert within the interface module of morphosemantics, rather than acting separately (as Lieber 1992 proposes).

Summarizing, some of the seemingly independent principles and constraints that have been proposed in the generative literature (the WBH, the Atom and Adjacency Conditions, the UOH) may be incorporated into the more fundamental notions of feature percolation and subcat/select and thus do not need to be formulated as separate constraints on WFRs. Inasmuch as the two basic notions of percolation and subcat/select crucially need each other in order to produce well-formed words, they emerge as the two essential mechanisms governing word formation.

In sections 3 and 4 the LCS is represented in the form of semantic features within feature blocks which may percolate, along with blocks containing morphosyntactic features, to the mother node in a derivative. (In Marantz 1984 we find a precursor to the present approach in proposing that argument structure (which is contained in the LCS) may be manipulated by percolation.) As proposed in sections 3 and 4, the percolation of feature blocks from a head and a nonhead to a branching node explains the layering of meaning as well as the addition or deletion of semantic primitives and slots that Lieber describes as taking place in the LCS when affixation occurs (cf. (2) above). Lieber's (1992) own notions of Head and Backup Percolation (as modified in section 3) can handle the percolation of all types of features, including semantic features and those related to argument structure.

2.4. Composition of Slot Structure

What follows is a description of the contents and characteristics of the slots and blocks that make up slot structure. Sections 3 and 4 illustrate how these slots and blocks are organized in a lexical entry and how they participate in derivation.

2.4.1. CATEGORIAL Slot

The CATEGORIAL slot contains a block with the ontological and syntactic categories of bases and suffixes. These two types of features have been placed together in this slot given the close link between the two mentioned in section 2.1. Following Lieber (1992), it is assumed that only items that may function as heads (e.g. bases and derivational suffixes) have categorial information and therefore a CATEGORIAL slot.

2.4.2. CORE Slot

The CORE slot contains a block with the core semantic features of bases and content suffixes.7 Function suffixes, which lack core semantic features, have an empty CORE slot.

2.4.3. ARGUMENT Slots

The ARGUMENT slots contain blocks with the syntactic selectional information of verbs, adjectives, nouns that take arguments, and certain suffixes. Since this study makes a semantic rather than syntactic distinction between arguments, the external and internal arguments will be labeled ARGUMENT I and ARGUMENT II, respectively.

The arrangement of ARGUMENT slots proposed in this model coincides with, and is based in part on Grimshaw's (1990) theory of argument and aspectual prominence, which makes a semantic distinction between what are traditionally called external and internal arguments. Grimshaw suggests that argument structure is organized on the basis of a scale (or template) of argument prominence and a scale of aspectual prominence. The argument scale is characterized by the relative prominence of arguments in argument structure, the most prominent argument being the external argument. According to Grimshaw, argument structure is a lexico-syntactic representation assembled from a set of elements identified by the LCS, to which Grimshaw ascribes ultimate responsibility for selection.

According to Grimshaw, aspectual prominence is determined by event structure. An event structure is commonly composed of two subevents, the first (or initial) and more prominent subevent being an activity or a process, the second (or final) and less prominent subevent a state or change of state. For Grimshaw, the argument and aspectual prominence scales are related in that each one of the arguments in argument structure is associated with a subevent in event structure (cf. Pustejovsky (1992, 1995), Tenny 1994, and Smith 1991 for more elaborate views of event structure). The argument linked to the initial subevent tends to be more prominent (usually the external argument) than the argument linked to the second subevent (the internal argument). Grimshaw argues that each argument linked to a subevent occupies a slot in the aspectual prominence scale.

Rappaport & Levin (1992) likewise support a semantic distinction between the external and internal arguments of a predicate on the basis of an apparent strong link between the external argument and a prominent event position. And, like Grimshaw, Pustejovsky (1995) favors the inclusion not only of argument structure but also event structure into the representation of lexical entries.

Dowty's (1991) theory of thematic proto-roles helps support the semantically-based distinction between external and internal argument proposed by Grimshaw. Following Rappaport & Levin (1988) and Jackendoff (1990), Dowty argues that theta-roles are not primitives, and proposes that the only theta-roles are two fuzzy, non-discrete conceptual categories called Proto-Agent and Proto-Patient, each of which is characterized by a cluster of properties related to the event denoted by the predicate, as shown in (14).

(14)	Contributing properties for the Proto-Agent
	a.	volitional involvement in the event or state
	b.	sentience (and/or perception)
	c.	causing an event or change of state in another participant
	d.	movement (relative to the position of another participant)
	Contributing properties for the Proto-Patient
	e.	undergoes change of state
	f.	causally affected by another participant
	g.	stationary relative to movement of another participant
		[adapted from Dowty 1991]

Arguments differ in the degree to which they bear each proto-role, as determined by the Argument Selection Principle, presented in (15).

(15)	Argument Selection Principle (ASP) The argument with the greatest number of Proto-Agent properties will be the subject (ARGUMENT I), while the argument with the most Proto-Patient properties will be the direct object (ARGUMENT II).
	[adapted from Dowty 1991]

Dowty argues that theta-role hierarchies follow from his two proto-role definitions (14). Proto-Agent-like arguments, that is, arguments (such as Agent or Experiencer) which have Proto-Agent properties (14) will tend to be higher in the hierarchy than Proto-Patient-like arguments (e.g. Theme). Dowty also claims that the unergative/unaccusative distinction in verbs corresponds semantically to the Proto-Agent/Proto-Patient division, with unergative arguments tending to be Proto-Agents (e.g. the argument of swim), and unaccusative arguments Proto-Patients (e.g. the argument of arrive).

In conclusion, following Grimshaw (1990), Pustejovsky (1995) and Dowty (1991), in this study argument structure, aspectual structure and a classification of arguments (ARGUMENT I, ARGUMENT II) are incorporated in the representation of slot structure (sections 3 and 4).

2.4.4. PARTICIPANT Slot

According to Grimshaw (1990), participants are the entities related to a predicate. However, not all entities are equally important. While arguments are participants that stand in a grammatically significant relationship to predicates, participants per se are entities implied by the situations in which the predicates are used, but are not as grammatically important as arguments. For example, as Grimshaw notes, even though the noun exam has no arguments, the existence of a participant-the person who created the exam-is implied. Grimshaw further notes that the LCS defines the set of participants (arguments included) involved in the meaning of a lexical item. Thus, the PARTICIPANT slot may be said to be a part of the LCS. The PARTICIPANT slot in the present model contains a block with information relative to the participants associated with a lexical item.8

Bases may contain one or more PARTICIPANT slots, but suffixes systematically lack a PARTICIPANT slot. This, and the fact that suffixes (and not bases) contain morphological subcategorization and selectional information are two of the most important formal differences between suffixes and bases. In the present approach, these distinctions are inherent in the representation of slot structure (sections 3 and 4).

2.4.5. SUBCAT/SELECT Slot

The SUBCAT/SELECT slot contains a block with the morphological syntactic and selectional features of a suffix. Since suffixes are the items that subcategorize for the bases they attach to, it is suffixes, and neither bases nor derivatives, that have a SUBCAT/SELECT slot.

As explained in section 2.1.1, morphological subcategorization and selection are highly relevant in connection to several constraints on WFRs. One of those constraints, the WBH, can be incorporated into the subcat/selectional frames of affixes, while other constraints simply assume these frames (e.g. the Atom and Adjacency Conditions and the UOH). As mentioned above, this points to the importance of subcat/select for morphosemantic theory, and, as illustrated in the following section, leads to the conclusion that subcat/select, along with feature percolation, are the two central mechanisms that drive regular derivation.

3. Modified Feature Percolation Conventions

I have based my work on predecessors such as Lieber (1992), Selkirk (1982), and Marantz (1984), who have claimed tree-like structure for morphological representations. Of these, Lieber (1992) has the most well-developed theory, a theory that is based on feature percolation. Lieber's (1992) Percolation Conventions are taken as the basis for the analysis developed in this and the following section.

The major modifications to Lieber's (1992) Percolation Conventions9 made in the present model consist of the incorporation of semantic features, and the organization of lexical information into slot structures. Because of this, the Modified Feature Percolation Conventions proposed here (16) involve a re-definition of the features that are allowed to percolate by Head and Backup Percolation, in answer to Lieber's (1992:77) question of "what features percolate, [and] where features are allowed to percolate from."

(16)	Modified Feature Percolation Conventions10
	a.	Head Percolation: The affix (the head) percolates its non-subcat/selectional information (i.e. its CATEGORIAL, CORE and ARGUMENT slots and blocks) to the branching node.
	b.	Secondary Percolation: All the information blocks of the base (i.e. the CATEGORIAL, CORE, ARGUMENT, and PARTICIPANT blocks) percolate to the branching node and attempt to occupy slots. Once a slot has been occupied, a percolating information block may occupy that slot as long as it has compatible features (i.e. either morphosyntactic or semantic). If a percolating block does not find an empty or compatible slot, it may not occupy any slots in the output, and is discarded.

First, notice that Backup Percolation has been relabeled Secondary Percolation. The term "backup" implies that certain features percolate after other features (head features) have percolated. However, as conceived in (16), the process of percolation does not imply any ordering in its application, but rather reflects the idea that Head Percolation takes precedence over the percolation of features from the base (Secondary Percolation; see further discussion below in this section).

The notion of "compatible features" encoded in Secondary Percolation needs to be elaborated on. Aside from having to belong to the same type (morphosyntactic or semantic), features are compatible if they can coexist in a single slot without contradicting or canceling each other out. For example, since an entity cannot be both an [EVENT] and a [THING], nor a noun and a verb at the same time, the CATEGORIAL blocks of a base and a suffix cannot occupy the same slot. On the other hand, since the features that may be stored in the CORE slot (classemes, semes, general features and primitive predicates (see section 2.1)) do not necessarily contradict each other, features from different CORE blocks may coexist in the same slot. Thus, "doubly-filled" slots (as when two CORE blocks occupy the same slot in a derivative) are allowed when there is feature compatibility (see derivations in section 4).

Because Dowty's (1991) Argument Selection Principle (ASP) (15) can be viewed as a refined form of the compatibility specified for Secondary Percolation, the ASP is incorporated into Secondary Percolation. Recall that the ASP states that the argument with the greatest number of Proto-Agent properties will be ARGUMENT I, while the argument with the most Proto-Patient properties will be ARGUMENT II. Thus, assuming that the ASP is encoded in Secondary Percolation, Proto-Agent-like arguments, being more compatible with an ARGUMENT I, will occupy the ARGUMENT I slot, while Proto-Patient-like arguments, being more compatible with an ARGUMENT II, will fill the ARGUMENT II slot.

Secondary Percolation (16b) entails that the CORE, ARGUMENT, and PARTICIPANT blocks of the base may occupy slots in the output after having percolated. The CATEGORIAL block of the base will not occupy the CATEGORIAL slot of the output because that slot is already filled by the CATEGORIAL block of the affix. The CATEGORIAL block of the base cannot occupy any other slots (i.e. the CORE or ARGUMENT slots) because of feature incompatibility. The CATEGORIAL block of the base therefore cannot occupy a slot in the output and is discarded. If the slot structure of the suffix provides only one or no ARGUMENT slots, some of the ARGUMENT blocks of the base will not find any slots to occupy and will be discarded as well.

Head Percolation (16a) entails that the affix imposes its slot structure on the output, with the final location of each block in the derivative being dictated by this slot structure. That is, once head features percolate (within information blocks), they determine what features of the base and affix will occupy what slots in the branching node. The slot structure of the affix (the head) thus constrains the possible lexical content (i.e. slot structure) of the output.

The changes in argument structure (e.g. suppression of an argument) are included in the changes brought about by the imposition of the slot structure of the affix on the output. Thus, there is no need to explain this suppression by employing rules which state, for example, that the affix "absorbs" or "binds" an argument. The suppression of arguments follows from the operation of percolation on slot structure. Head Percolation thus allows us to predict what the feature composition and slot structure of the derivative will be, making the notion of "head" in this concatenative model crucial. The significance of the head in this model places it in stark contrast to models based on lexical rules, where the notion of "head" is non-existent.

Crucially, a head is characterized by the fact that it imposes its categorial features on the output,11 and affects argument structure by adding arguments or contributing to their suppression. It follows that the nonhead in a derivative can neither impose its categorial features on the output nor bring about changes in argument structure. The trees in the next section illustrate that, in contrast to syntax, in morphology the fact that an item has a (morphological) subcategorization frame does not necessarily make it a head. Affixes need a SUBCAT/SELECT block because they are bound morphemes, but what characterizes an affix as a head are the points just mentioned. An important conclusion that derives from the notion of head and the mechanism of Head Percolation is that they give rise to the prediction that the features of the nonhead (the base) do not overrule the features of the head (the affix) in the output.

In sum, the whole process just described can be conceived of as the percolation of both morphosyntactic and semantic information in parallel fashion from the head and nonhead to the branching node, with the features of the head preempting those of the nonhead.

The application of the modified version of feature percolation (16) is demonstrated with sample suffixes in the trees in the following section.

4. Derivational Trees

The trees in this section, which illustrate the model proposed, show that when the information in the lexical entries of bases and affixes is organized in slot structures, predictions can be made about the organization of information in derivatives, including their argument structure. The following are sample derivatives formed with suffixes that are representative of the more than 50 affixes (forming more than 1200 derivatives) whose meaning and morphological behavior were analyzed for this study.

4.1. Suppression of Arguments

What follows is a description of Tree 1, a derivation with the suffix -dora. The slots in the output that contain information blocks that have percolated to the branching node by Head Percolation have been set in boldface (in all the trees). To facilitate interpretation, arrows have been placed in lexical entries to indicate the filling of a given slot in the output by a given block. For example, in Tree 1, the arrow in the CORE block of the base signals that that block occupies the CORE slot in the output.

As noted in section 2, the LCS, which contains semantic information and from which argument and aspectual structure derive, is itself contained within slot structure. This is reflected in the trees. In Tree 1 and remaining trees, the shaded areas represent the LCS of each lexical item. The slots in each lexical entry are ordered vertically rather than horizontally in order to facilitate the representation of the merging of the lexical entries of the base and suffix. Note that although the slots in each lexical entry have been numbered, no ranking is implied. The slots have been arranged in an arbitrary order which, for ease of exposition and for the sake of consistency, is the order assigned to all lexical entries in the remainder of this study. Slots for arguments that are empty have been labeled but not numbered (e.g. the ARGUMENT II slot in the suffix in Tree 2).

In order to make clear the relation between the conventional (horizontal) LCS representation and the vertical one illustrated in each tree below, consider the LCS (in simplified notation, following Levin & Rappaport (1998)) of colar 'sift' in (17) as compared to the LCS in Tree 1.

(17)	colar 'sift': [x CAUSE [y BECOME SIFTED]]

The first subevent, with its accompanying argument, [x CAUSE], corresponds to block 6 in the tree, while the second subevent, [y BECOME SIFTED], corresponds to block 7. Block 5, the CORE block, as its name indicates provides the core meaning of sift. The same relation between the horizontal LCS and the vertical LCS should be interpreted in the trees that follow.

Note that the ARGUMENT I and ARGUMENT II labels represent Dowty's (1991) Proto-Agent and the Proto-Patient, respectively. Features such as CHANGE OF STATE, in slot 7, are meant to represent the association between arguments and Grimshaw's (1990) aspectual scale. The terms Agent and Theme are used merely as labels for arguments, and thus, following Lieber (1992), Rappaport & Levin (1988, 1992), and Dowty (1991), are not considered primitive notions.

The ARGUMENT slots of the base, with their blocks containing aspectual and argument features, constitute the argument and aspectual structure of the verb. Since the verb (colar) is an event composed of two subevents, each one of the two ARGUMENT blocks represents an argument associated with a subevent; each block occupies a position in both the argument and aspectual scales. The CORE slots of the base and suffix contain the non-argumental semantic features of the LCS, while the CATEGORIAL slots, which contain syntactic information, are a part of slot structure but not of the LCS (see section 2).

Even though the two ARGUMENT blocks of the base percolate to the branching node by Secondary Percolation, they cannot occupy slots in the output because the output, being a noun without an argument structure, does not have ARGUMENT slots. The failure of the two ARGUMENT blocks of the base to find slots in the output represents the operation where the LCS of the suffix completely "deletes" or "suppresses" the argument structure of the base (see Lieber's 1992 analysis of this in section 2). This is one of the ways in which Head and Secondary Percolation determine the interaction between the LCSs of the lexical items that participate in a derivation.

Finally, notice in Tree 1 how the SUBCAT/SELECT block specifies that the suffix may only attach to verbs that have an ARGUMENT I block. This ensures that derivatives formed with, for example, unaccusative verbs are ungrammatical (e.g. an instrument called *existi+dora 'exister'; the suffix -dor is similar, cf. *llega+dor 'arriver', *ocurri+dor 'occurrer'). Because it contains semantic information, the SUBCAT/SELECT block is considered to be a part of the LCS of derivational affixes.

4.2. Addition Accompanied by Transference of Arguments

Next consider the derivation in Tree 2. Here the suffix provides a filled ARGUMENT I slot to the output, and the base ARGUMENT II block fills the ARGUMENT II slot of the output, in what could be called transference (or inheritance) of arguments.

The single argument of the base, an ARGUMENT II as determined by the ASP, occupies the ARGUMENT II slot in the output because it remains Proto-Patient-like in the output. This is a case where, on merging, the LCSs of the base and suffix each contribute one argument to the derivation. Again, Head and Secondary Percolation determine the interaction between LCSs.

Notice in Tree 2 that the information contained in the SUBCAT/SELECT block indicates that the suffix -izar may attach to both adjectives and nouns, as well as to bases which are either [PROPERTY]s or [STATE]s. Since in both types of derivation -izar has the same meaning (causation), according to the UOH it is a single affix. However, -ize has other meanings (cf. Plag 1997) and an adequate representation of -ize and other polysemous affixes is still an unresolved issue at the heart of the debate on how to more appropriately represent the LCSs of polysemous items (cf. Plag 1997 and Lieber 1998).

4.3. Transference of Arguments

In Tree 3, after the merging of LCSs, only one argument occupies a position in the output. The ARGUMENT II block cannot occupy any slots in the output because there are none available, and is discarded. The ARGUMENT I block of the base remains Proto-Agent-like and occupies (i.e. is transferred to) the fully compatible ARGUMENT I slot of the output.

4.4. Summary

The list in (18) summarizes the operations on argument structure that result from derivation, and which have been illustrated in the trees above. (The arrows indicate transference by percolation of a base argument to a slot in the output.)

(18)	Operations on argument structure Suppression of one or more arguments (Trees 1, 3) Addition of an argument (Tree 2) ARGUMENT II ® ARGUMENT II (Tree 2) ARGUMENT I ® ARGUMENT I (Tree 3)

Notice that the operations ARGUMENT II - ARGUMENT I and ARGUMENT I - ARGUMENT II do not occur, an indication that when arguments percolate to the output, they tend to remain of the same type (i.e. either ARGUMENT I or ARGUMENT II). In certain causative constructions (e.g. Chichewa phík+íts 'cook' + 'cause' = 'cause to cook'), where the affix contributes to the output a strongly Proto-Agent-like argument (the Causer), the operation ARGUMENT I - ARGUMENT II does occur (see section 4.5), which leaves the operation ARGUMENT II - ARGUMENT I as the only one disallowed in morphology. This prohibition suggests that the semantic promotion of arguments from Proto-Patient-like to Proto-Agent-like in derivation is forbidden, a result that might be related to the ASP in conjunction with the notion of "head". I leave this matter for future research.

4.5. Other affixes

The model illustrated in the previous sections accounts for derivational suffixation, but can be extended to other types of affixes (in Spanish and other languages), namely, derivational prefixes, passives, causatives, applicatives, expressive suffixes (e.g. diminutives), inflectional affixes, and parasynthetics. Due to space limitations, only one sample derivation each for derivational prefixes and causatives suffixes will be illustrated in this section.

4.5.1. Derivational Prefixes

Only a few Spanish derivational prefixes yield outputs with a syntactic or ontological category that is different from that of their base (cf. Lang 1990, Scalise 1984, and see tree in this section). Those prefixes that do not yield outputs with a different syntactic category (the majority) do not yield outputs with a different ontological category either, which suggests that they do not contain any categorial information. For example, the prefix co- attaches to nouns that are [THING]s and yields output nouns that are [THING]s as well (e.g. co+autor 'co-author', co+piloto 'co-pilot'), and the prefix des- attaches to verbs that are [EVENT]s and yields output verbs that are also [EVENT]s (e.g. des+acelerar 'decelerate' < acelerar 'accelerate', des+amarrar 'untie' < amarrar 'tie').

Among derivational prefixes there does not appear to be a division between content and function items as in derivational suffixation, since all derivational prefixes appear to contribute objective meaning to the derivation. However, a distinction can be made between those prefixes that do not yield outputs with a different category than that of their base (the majority), and those that do.

As is the case for derivational suffixes, derivational prefixes have stricter selectional restrictions than non-derivational (e.g. inflectional) affixes. The prefix re-, for example, selects for bases that have the feature CHANGE OF STATE in their lexical entry (cf. re+paint vs. *re+punch [someone]), as shown in Tree 4.

The prefix, which does not yield an output with a syntactic or ontological category different from that of the base, adds semantic information (CORE block) to the output via Secondary Percolation. Notice that the prefix does not affect the argument structure of the base. That is, the prefix neither adds an argument to the output nor contributes to the suppression of an argument of the base. This suggests that the prefix is not a head, since, as noted in section 3, a head is characterized by the property that it imposes its slot structure on the output, thereby affecting argument structure.

However, some derivational prefixes do behave as heads. The prefix pro- in pro+amnistía 'pro-amnesty', for example, yields an output (an adjective) with both syntactic and ontological categories different from those of the base (a noun). The features of the base, amnistía, percolate via Secondary Percolation. In addition, just like typical derivational suffixes, the prefix contributes an argument to the output (we omit the tree in the interest of space).

4.5.2. Causative Suffixes

The key semantic information of causative affixes consists of the single primitive predicate CAUSE, which is associated with two arguments, a causer and a causee. The subcategorization information of causatives may make reference to both transitive and intransitive verbs. Causatives typically attach to bases that are [EVENT]s, but in certain languages (e.g. Malayalam, Madurese) they also attach to bases characterized by the ontological categories [PROPERTY] or [STATE]. In all cases, causatives yield [EVENT] outputs, which indicates that the causative affix contains the ontological category [EVENT] in its CATEGORIAL block (cf. Jackendoff 1983). Since [EVENT]s prototypically correspond to verbs, it stands to reason that the syntactic category of causatives is that of verb.

Tree 5 illustrates a construction involving a causative suffix attaching to an intransitive verb.

There are no changes in ontological category, but there is an important effect on argument structure that is typical of causative constructions (cf. Marantz 1984). Since the ARGUMENT I of the base has fewer Proto-Agent-like properties in the output, it is Proto-Patient-like in relation to the ARGUMENT I contributed by the suffix. In other words, since the ARGUMENT I of the suffix, being a causer is more strongly Proto-Agent-like than the ARGUMENT I of the base (an agent), the argument contributed by the suffix occupies the ARGUMENT I position in the output, relegating the ARGUMENT I of the base to the less Proto-Agent-like ARGUMENT II position in the output.

The generalization, based on the ASP (see section 2.4.3), seems to be that even though a strong Proto-Agent-like argument (e.g. a causer) may demote a somewhat less Proto-Agent-like argument to an ARGUMENT II position, it seems unlikely that a Proto-Patient-like argument (of the base) may acquire enough Proto-Agent-like properties to demote a Proto-Agent. However, the prohibition (*ARGUMENT II - ARGUMENT I) may be due to a broader generalization that results from Head Percolation and the notion of "head." Since, as noted in section 3, the information of the nonhead (the base) may not override that of the head (the affix), an ARGUMENT II contributed by the base may not overrule an ARGUMENT I (or any other argument) contributed by the affix. Thus, the Modified Feature Percolation Conventions predict the prohibition of a semantic promotion from ARGUMENT II to ARGUMENT I in derivation.

4.6. Other languages

Since a highly valued theory should be applicable cross-linguistically, this section provides examples of how the theory developed in this study may apply to derivation in several languages genetically unrelated to Spanish, namely Mam, Turkish and Swahili. It is shown that these languages, like Spanish, have both content and function suffixes and, overall, have a derivational component closely resembling that of Spanish.

4.6.1. Mam

England (1983) states that Mam, a Mayan language, has a large class of derivational affixes (up to seventy-seven) which yield outputs with a word class, meaning, or both, different from those of the base they attach to. According to England, these affixes contribute meanings such as "repetitive," "instrumental," "locative," "remainder," "abstract," "facility," "measure," and "direction," strongly reminiscent of what I have labeled general features (e.g. Instrument), and other features (e.g. classemes such as [-concrete]="abstract") (see section 2.1). The vast majority of Mam bases are [EVENT]s, followed by [PROPERTY]s. A few bases are [THING]s. A few sample Mam derivational suffixes, with their associated meanings, are shown in (19).

(19)	Mam derivational suffixes
	Content suffixes
		Repetition
			qej+najee7 'lie down every little while' < qej 'lie down'
		Location
			poom+b'il 'place for performing rites' < poom 'perform rites'
		Facility
			txik+b'aajal 'easy to cook' < txik 'cook'
	Function suffixes
		[STATE]
			yaab'+il 'sickness' < yaab' 'sick'

Tree 6 illustrates a derivation with the content suffix -b'aajal.

As with derivatives formed with -izar (see section 4.2), the ARGUMENT II of the base occupies the ARGUMENT II slot in the output. In this case, however, the base has an ARGUMENT I, which, finding no available slots, is discarded.

4.6.2. Turkish

Lewis (1967) observes that in Turkish, an Altaic language, there are close to forty derivational suffixes, contributing meanings such as "occupation," "result," "action," "instrument," "abstract," "intensive," "agent," and "inchoative," which, again, coincide with some of the features mentioned above. As in the case of Mam, there are content and function suffixes in Turkish, as seen in (20).

(20)	Turkish derivational suffixes
	Content suffixes
		Occupation
			ögret+men 'teacher' < ögret 'teach'
		Instrument
			ele+k 'sieve' < ele 'sift'
		Location
			dur+ak 'stopping-place' < dur 'to stop'
	Function suffixes
		[STATE]
			kIskan+ç 'jealousy' < kIskan 'to envy'
		Incoahative
			serin+les 'become cool' < serin 'cool'

Tree 7 illustrates a derivation with the content suffix -ak.

The core features of the base transfer to the core block of the suffix, and the ARGUMENT I is discarded.

4.6.3. Swahili

According to Myachina (1981) and Polomé (1967), Swahili, a Bantu language, also has derivational suffixes with meanings such as "instrument," "habitual," "occupation," and "abstract." Like Mam and Turkish, Swahili has both content and function suffixes, as shown in (21).

(21)	Swahili derivational suffixes
	Content suffixes
		Habitual
			m+wind+i 'occasional hunter' < winda 'hunt'
		Occupation
			m+wind+aji 'professional hunter' < winda 'hunt'
	Function suffixes
		[ACTION]
			u+haribif+u 'destruction' < haribu 'destroy'

Tree 8 illustrates a derivation with the content suffix -aji.

This section has shown that the morphosemantics of several languages unrelated to Spanish, namely Mam, Turkish and Swahili, closely resembles that of Spanish. The model proposed for Spanish in this study is applicable in its entirety to the derivational morphology of these languages, which suggests that the mechanisms of percolation and subcat/select, as well as the notion of slot structure, are universal constructs.

5. Conclusion

This paper has shown how a model based on the key notions of slot structure, subcat/select, and the manipulation of LCS by percolation can account adequately for Spanish regular derivation. The model successfully unifies, in a principled way, all the processes that take place during the formation of a complex word in a single representation. By utilizing only two formal mechanisms (percolation and subcat/select), the model considerably reduces the machinery of the grammar. In addition, the application of these mechanisms gives rise to the central prediction made by the model: the slot structure of the head, given Head Percolation, predicts what is possible and impossible in a derivative by determining the slot structure of the output.

The proposal may be extended to other types of affixes (in Spanish and other languages) (e.g., derivational prefixes, causatives), as well as to regular derivational suffixation in languages genetically unrelated to Spanish (e.g. Mam and Turkish). This suggests that the model has cross-linguistic, universal application. Future research could focus on further formalization of the interface component of morphosemantics and the role of the present approach in that interface module.


Notes

1 Morphosemantics is considered a sort of interface module concerned with the interaction between morphology and lexical semantics. (see definitions of the term in Dressler & Kiefer 1990, Mighetto 1981, and Gordon 1978.)

2 I draw heavily on the data provided by Lang (1990). Other sources include Spanish corpora provided by native speakers, Bosque & Pérez' (1987) and Stahl & Scavnicky's (1973) reverse dictionaries, and several Latin American and Spanish online newspapers such as the Honduran La Prensa, the Colombian El Tiempo, and the Spanish El País.

3 Grimshaw (1981) hypothesizes that certain semantico-cognitive categories such as "object" and "action" have a Canonical Structural Realization (CSR), the assignment on a one-to-one basis of a syntactic category to a word of a certain semantico-cognitive type (what Jackendoff 1983 terms "ontological category"). Thus, "objects" are prototypically assigned the category N, while "actions" are assigned the category V. In general, adds Grimshaw, "a word belongs to its CSR, unless there is evidence to the contrary."

4 According to Lieber (1992) and Selkirk (1982), lexical items contain both morphosyntactic (i.e. grammatical) and semantic features. Inasmuch as the manipulation of semantic features is central to this study, it is necessary to take into account "componential semantics" (also known as Feature Semantics), the theory of lexical meaning that is concerned with semantic features (cf. Goodenough 1956, Hjelmslev 1961, Coseriu 1977, Lyons 1977, Lipka 1990, Wierzbicka 1996, Jackendoff 2002). The importance of the features resides not in their classification but in the fact that they serve to characterize the meaning of lexical items in relative detail, which will allow us to make an explicit analysis of derivation. We must keep in mind Kornfilt & Correa's (1993) observation that the exact set of semantic primitives has not been identified, a situation that makes the classification of semantic features an open problem. The following are the features that are employed in this study.

Semantic features
a. classemes (e.g. [+concrete])
b. semes (e.g. [+has four legs])
c. general features (e.g. Abundance)
d. ontological categories (e.g. THING)
e. primitive predicates (e.g. CAUSE)
f. aspectual features (e.g. ACTIVITY)

5 Note that both conditions are also able to rule out ungrammatical forms. For instance, if the suffix -er were to attach to read yielding read+er, the ungrammaticality of the derivative *reader+ity would be predicted by both conditions. The suffix -ity is sensitive to the features of -er, which is both the head (Atom Condition) and the most recently attached affix (Adjacency Condition) of read+er. Since -er contributes nominal features but -ity only attaches to adjectives, both conditions predict that -ity will not be able to attach to the noun read+er to yield a grammatical derivative.

6 The WBH in effect disallows the attachment of an affix to a non-existent word, or to another affix in isolation. It works to make predictions about what is possible in language and what is not, a characteristic that makes it a central constraint in morphology. See section 2.1.5 and chapter 5 for how this constraint can be applied in a concatenative model.

7 According to Lehrer (1996), derivational suffixes may be classified into "content" suffixes and "function" suffixes, on the basis of a comparison of suffixes with full words. Lehrer argues that just as function words have more limited meanings than content words, so function suffixes have more limited meanings than content suffixes.

8 McKoon & MacFarland (2000) differentiate participants on the basis of the distinction between structural meaning (roughly core meaning) and content meaning (roughly world knowledge). Participants associated with the structural meaning of a lexical item are considered arguments, while participants associated with the content meaning are considered secondary and are usually optional participants. In John broke the window with a hammer, for example, John and window are the arguments of break because they are part of its structural meaning, whereas the instrument participant hammer, which is optional, is part of the content meaning.

9 Lieber's (1992) Feature Percolation Conventions
a. Head Percolation: Morphosyntactic features are passed from a head morpheme to the node dominating the head. Head Percolation propagates the categorial signature (see definition below).
b. Backup Percolation: If the node dominating the head remains unmarked for a
given feature after Head Percolation, then a value for that feature is percolated from an immediately dominated nonhead branch marked for that feature. Backup Percolation propagates only values for unmarked features.
The categorial signature is a set of morphosyntactic features (such as [±Plural], [±Fem], [±Masc], [±I] and [±II]) headed by the category features [±N] and [±V].

10 Two of Krieger's (1994) principles have a role nearly equivalent to the one played by these conventions. These principles, which are based on principles formulated by Pollard & Sag (1994) within the framework of Head-Driven Phrase Structure Grammar (HPSG), are the Morphological Head Feature Principle, which states that the category of the mother (the branching node) is the same as that of the head daughter, and the Morphological Semantics Principle, which holds that the semantics of the mother is equal to the semantics of the head daughter. When taken together, Krieger's principles perform a function that is similar to the function of the Modified Feature Percolation Conventions. Also note that the merging of information blocks in the branching node brought about by the conventions closely resembles what in HPSG is termed "unification," a process whereby the information contained in the feature structures of two lexical items combine to form a single feature structure. A "feature structure" is the set of features and their values (e.g. [number: singular]) that make up the content of a lexical item or constituent (cf. Shieber 1986, Pollard & Sag 1994).

11 This part of the characterization of the notion of head is the relevant one for HPSG. In HPSG, the head is the element of a phrase or complex word whose category is assigned to the whole (cf. Pollard & Sag 1994).

About the Author

Dr Benavides works in the Department of Foreign Literature and Languages at the University of Massachusetts, Dartmouth, USA.

Email: cbenavides@umassd.edu

References

Aronoff, Mark. 1976. Word formation in generative grammar. Cambridge, MA: MIT Press.

Beard, Robert. 1995. Lexeme-morpheme base morphology. Albany: SUNY Press.

Booij, Gert. 1977. Dutch morphology: A study of word formation in Generative Grammar. Lisse: Peter de Ridder.

Bosque, Ignacio, and Manuel Pérez Fernández. 1987. Diccionario inverso de la lengua española. Madrid: Gredos.

Cinque, Guglielmo. 1990. Ergative adjectives and the lexicalist hypothesis. Natural Language and Linguistic Theory 8.1-39.

Coseriu, Eugenio. 1977. Principios de semántica estructural. Madrid: Gredos.

Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67.547-619.

Dressler, Wolfgang, and Ferenc Kiefer. 1990. Austro-Hungarian morphopragmatics. Contemporary morphology, ed. by Wolfgang Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison, 69-77. Berlin: Mouton.

England, Nora. 1983. A grammar of Mam, a Mayan Language. Austin: University of Texas Press.

Giorgi, Alessandra, and Giuseppe Longobardi. 1991. The syntax of noun phrases. Cambridge: Cambridge University Press.

Goodenough, Ward H. 1956. Componential analysis and the study of meaning. Language 32.195-216.

Gordon, W. Terrence. 1978. Morphosemantics: A neglected chapter in linguistics. Glossa 12.3-15.

Grimshaw, Jane. 1981. Form, function, and the Language Acquisition Device. The logical problem of language acquisition, ed. by C. L. Baker and John J. McCarthy,165-182. Cambridge, MA: MIT Press.

--. 1990. Argument structure. Cambridge, MA: MIT Press.

Hjelmslev, Louis. 1961. Prolegomena to a theory of language. Madison: The University of Wisconsin Press.

Jackendoff, Ray. 1983. Semantics and cognition. Cambridge, MA: MIT Press.

--. 1990. Semantic structures. Cambridge, MA: MIT Press.

--. 2002. Foundations of language: Brain, meaning, grammar, evolution. New York: Oxford University Press.

Kornfilt, Jaklin, and Nelson Correa. 1993. Conceptual structure and its relation to the structure of lexical entries. Knowledge and Language, vol. 2, ed. by Eric Reuland and Werner Abraham, 79-118. Dordrecht: Kluwer.

Kratzer, Angelika. 1995. Stage-level and individual-level predicates. The generic book, ed. by Gregory Carlson and Francis Jeffry Pelletier, 125-75. Chicago: The University of Chicago Press.

Krieger, Hans-Ulrich. 1994. Derivation without lexical rules. Constraints, language and computation, ed. by C.J. Rupp, M.A. Rosner and R.L. Johnson, 277-313. London: Academic Press.

Lang, M.F. 1990. Spanish word formation. London: Routledge.

Lehrer, Adrienne. 1996. How lexeme-like are affixes? MS.

Levin, Beth, and Malka Rappaport. 1988. Nonevent -er nominals: A probe into argument structure. Linguistics 26:1067-1083.

--. 1995. Unaccusativity. Cambridge, MA: MIT Press.

--. 1998. Morphology and lexical semantics. The handbook of morphology, ed. by Andrew Spencer and Arnold Zwicky, 248-271. Oxford, UK: Blackwell.

Lewis, G.L. 1967. Turkish grammar. Oxford: Clarendon.

Lieber, Rochelle. 1980. On the organization of the lexicon. Cambridge, MA: MIT dissertation.

--. 1992. Deconstructing morphology. Chicago: The University of Chicago Press.

--. 1998. The suffix -ize in English: Implications for morphology. Morphology and its relation to phonology and syntax, ed. by Steven Lapointe, Diane Brentari, and Patrick Farrel, 12-33. Stanford, CA: CSLI.

Lieber, Rochelle, and Harald Baayen. 1993. Verbal prefixes in Dutch: A study in lexical conceptual structure. Yearbook of Morphology 51-78.

--. 1998. Nominalizations in a calculus of lexical semantic representations. Yearbook of Morphology, ed. by Geert Booij and Jaap van Marle, 175-197. Dordrecht: Foris.

Lipka, Leonhard. 1990. An outline of English lexicology. Tübingen: Max Niemeyer.

Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.

Marantz, Alec. 1984. On the nature of grammatical relations. Cambridge, MA: MIT Press.

Mighetto, David. 1981. Ajuste morfosemántico y determinación morfológica: ¿-ión o -miento? Studia Neophilologica 53.347-354.

Myachina, E. N. 1981. The Swahili language. London: Routledge & Kegan Paul.

Pinker, Steven. 1999. Words and rules. New York. Basic Books.

Plag, Ingo. 1997. The polysemy of -ize derivatives: On the role of semantics in word formation. Yearbook of Morphology 219-42.

Pollard, Carl, and Ivan Sag. 1994. Head-driven phrase structure grammar. Chicago: The University of Chicago Press.

Polomé, Edgar. 1967. Swahili language handbook. Washington, D. C.: Center for Applied Linguistics.

Pustejovsky, James. 1992. The syntax of event structure. Lexical and conceptual semantics, ed. by Beth Levin and Steven Pinker, 47-81. Cambridge: Blackwell.

--. 1995. The generative lexicon. Cambridge, MA: MIT Press.

Rappaport, Malka, and Beth Levin. 1988. What to do with q-roles. Syntax and semantics, vol. 21, ed. by W. Wilkins, 7-36. New York: Academic Press.

--. 1992. -er nominals: Implications for the theory of argument structure. Syntax and semantics, vol. 26, ed. by Tim Stowell and Eric Wehrli, 127-53. New York: Academic Press.

Scalise, Sergio. 1984. Generative morphology. Dordrecht: Foris.

Selkirk, Elisabeth. 1982. The syntax of words. Cambridge, MA: MIT Press.

Shieber, Stuart M. 1986. An introduction to unification-based approaches to grammar. Stanford: Center for the Study of Language and Information.

Siegel, Dorothy. 1977. The Adjacency Constraint and the theory of morphology. NELS 8, 189-97.

Smith, Carlota. 1991. The parameter of aspect. Dordrecht: Kluwer.

Speas, Margaret. 1990. Phrase structure in natural language. Dordrecht: Kluwer.

Stahl, Fred A., and Gary E.A. Scavnicky. 1973. A reverse dictionary of the Spanish language. Urbana: University of Illinois Press.

Tenny, Carol. 1994. Aspectual roles and the syntax-semantics interface. Dordrecht: Kluwer.

Wierzbicka, Anna. 1996. Semantics: Primes and universals. Oxford: Oxford University Press.

Williams, Edwin. 1981. On the notions 'lexically related' and 'head of a word'. Linguistic Inquiry 12.245-274.