Chapter 3
Distributed Morphology
Introduction
Distributed Morphology (DM), a lexical realizational theory[1] (Sproat, 2007, pg 65) was proposed by Halle and Marantz (1993, 1994). The theory believes that the ‘words’ are structurally not different from other constituents like- phrases or sentences, and are formed and manipulated using syntactic rules. This suggests that word formation is primarily a syntactic operation, i.e., the morphological structure of a word or a word form is generated using syntactic operations. It is syntax that provides features and the structures upon which morphology operates. This view is opposed to the one that believes that morphology operates in an entirely separate component that generates words or word forms outside syntax that are later fed into syntax. Unlike other morphological theories such as lexicalism and affixless morphology, DM believes that various components of morphology are distributed among various levels in the process of word computation. The chapter will discuss the major components of the theory and their distribution among various levels in the process of word computation. The theory stands against various postulates of the earlier morphological theories (such as lexicalist, word-and-paradigm based, a-morphous morphology, etc.) and presents a strong case to handle various complex phenomena especially syncretism and syntax-morphology mismatch better than others.
The Basic Architecture
For DM, ‘morphology’ neither lies wholly within the lexicon as articulated by lexicalists nor does it happen in syntax as with the proponents of lexeme-based morphology. According to DM, the constituent components of morphology are distributed among various levels in the architecture of grammar that work in close connection with syntax and phonology. Halle and Marantz postulate a separate level of representation called Morphological Structure (MS) (shown in Figure1 below) that operates in between Syntactic Structure (SS) and Phonological Form (PF). This level receives hierarchical structures from syntax that contain ‘abstract’ morphemes as the terminal nodes; abstract, because at this level, these nodes only have morphosyntactic and semantic features and lack any phonological features associated with them. The DM grammar can be represented as below:
Figure 1: The Architecture of Grammar in DM (Adapted from Halle and Marantz, 1993)
Embick and Halle (2005) talk about two types of terminals or morphemes that take part in the course of derivation. These are abstract morphemes (placeholders for functional categories such as Tense, Case, Number, etc.) and roots (which represent open class items or indeterminate lexical category whose categorial features are determined by their syntactic context (Marantz, 1997)). The syntactic and semantic features for these nodes are provided by UG (Universal Grammar) whereas the phonological features are supplied by language vocabulary in the form of vocabulary items (consisting of roots and affixes). The items are listed in the form of rules that contain a phonological form along with information regarding its morphosyntactic context. The insertion of vocabulary items into the terminal nodes takes place at Spell out and is named as Vocabulary Insertion. In addition, words with specialized meanings are treated as idioms and are stored as encyclopaedic entries. Here are some examples of morphemes, vocabulary items and encyclopaedic entries in English and Hindi:
Morphemes
a) Roots
In English:
1. Class X: [play], [cry]
2. Class Y: [bring], [think]
3. Class Z: [sing], [drive]
In Hindi:
3. Class X: [खेल], [सोच]
4. Class Y: [रो], [ला]
b) Functional Categories
5. In English: [Tense, +past]
6. In Hindi: [Tense, +past, +masc, +plural]
Vocabulary items
In English:
7. null <-> /+perfective] for X
8. /-t/ <-> [+perfective] for Y
9. /-d/ <-> [+perfective]
In Hindi:
10. /-yaa/ <-> [+perfective, +masc, -pl] for Y
11. /-aa/ <-> [+perfective, +masc, -pl]
Encyclopaedic Entries (Noyer, 2006)
12. [[ lɛg Root] n] <-> ‘leg’
13. [[hæi Root] tæn RootP] <-> ‘ride on a surfboard’
Basic Components
The framework of DM relies on three core properties that set it apart from other morphological theories. These properties or the key components of the theory as identified by Halle and Marantz (1993, 1994) are:
Late insertion
‘Early insertion' described in other theories (like lexicalism) maintain that lexical items are combined in the lexicon and contribute their features to the features of the words, which then combine in the syntax (Halle and Marantz, 1994). This essentially means that whatever features are available at the terminal nodes, come only from the lexical items. As opposed to the notion of ‘early insertion’, in DM, the phonological features are subject to late insertion and do not add or delete any semantic or syntactic features already present at the terminal. Phonological content is provided to the syntactic terminals (abstract morphemes) only at PF. Both roots and affixes appear as empty slots at the level of syntax and vocabulary items are inserted late in the course of derivation. According to Embick (2000), roots are inserted first, may be before or after syntax, while insertion of affixes or vocabulary items takes place only at spell out just before PF.
Underspecification
Underspecification means that the items need not be fully specified for the features present at the node of insertion. Their features can rather be a subset of those present at the node. This results in competition for insertion. The highly specified item wins over less specified ones and gets inserted. The default affixes are least specified or underspecified and are inserted only when more specific features are not present. See an example of an underspecified entry to form subjunctive in Hindi. Other theories (like lexicalism) demand full specification of lexical entries before they are inserted in the syntactic structure. These entries need to have all features that are necessary to satisfy the constraints like subcategorization frame, etc.
Syntactic Hierarchical Structure All the Way Down
Both phrases and words represent similar hierarchical structures. In other words, the hierarchical structure exists at every level of word computation. The structure is provided and regulated only by the rules of syntax. The only operations (morphological) that are allowed are those that work with strict local constraints and otherwise do not alter the already available structure. These operations include head-to-head movement, merger of the adjacent nodes, fission of a node, etc., The operations will be discussed in detail below.
The view contrasts with other theories that believe in the formation of words outside syntax using morphological templates or level ordering in lexicon or using subcategorization frames (Halle and Marantz, 1994). DM, on the other hand, maintains strict separation between the rules of vocabulary insertion and those of morphophonological allomorphy rules. For example, once the root sit and the affix null or zero morpheme for [+perfective] are inserted, phonological rule is applied that changes /i/ to /æ/ and derive sat for [+perfective]. It will not be correct to say that the change /i/ to /æ/ leads to the spell out of [+perfective], rather it is [+perfective] that allows the insertion of real morphemes which later undergo phonological change due to their association.
Morphological Operations
At the level of MS, before Vocabulary Insertion, the terminal nodes containing morphosyntactic features undergo various morphological operations such as merger, fusion, insertion, fission, and impoverishment (see Halle and Marantz 1993, Harley and Noyer 1999 for details). These operations follow strict locality conditions and syntactic hierarchical principles and have the capacity to modify the terminal nodes as derived from syntax.
Merger generally joins a category head with the head of its complement XP (Halle and Marantz, 1993). As a result of merger, a new word is formed from the heads of independent phrases. Merged nodes may or may not fuse depending upon the morphology of a language. Without fusion, two separate morphemes are inserted while a single morpheme is inserted if fusion takes place.
Hindi is morphologically quite rich in terms of verbal morphology and shows inflectional marking of tense, aspect, mood or agreement on to the verb itself. Hindi thus offers a case where unlike English, the verbal head raises to tense or aspect or mood in order to get adjoined to the other syntactic head. Merger may or may not lead to fusion. The first example of merger given for the verb ‘khaayaa’ (ate-masculine-singular) leads into fusion as a single morpheme –yaa appears as an exponence for both aspect and agr. The second merger example diagrammed for the verb ‘khaaegaa’ (will eat-singular-masculine) does not end into fusion as the future morpheme -g and –aa though appear in the same word form, exist as two separate morphemes. Here are the examples of merger in Hindi verb inflectional morphology:
a) khaayii (ate-singular-feminine) – Merger of Verb and Aspect in Figure 2
Figure 2: Merger of Verb and Aspect nodes
After vocabulary insertion of the entries khaa (eat) and yii the tree will generate the form khaayii.
b) khaauuM (go-1st person-singular) – Verb and Mood (subjunctive) to form root subjunctive in Figure 3
Figure 3: Merger of Verb and Mood nodes and fusion of mood and AGR1
After vocabulary insertion of the entries khaa and -uuM, the tree will generate the form khaa-uuM.
Fusion combines two sisters under a single category node and produces a single terminal node. The operation leads to the insertion of a single vocabulary item that has the combined morphosyntactic features of the two fused nodes. Fusion results in fewer numbers of independent morphemes in a tree.
Examples: khaayaa (ate-singular-masculine), khaayii (ate-singular-feminine), khaaye (ate-plural-masculine): Asp (perfective) and AGR2
jaauuM (go-1st person-singular), jaaye (go-2nd or 3rd person, singular), jaayeM (go-plural): subjunctive and AGR1
Here is an example of fusion of the aspect (perfective) and AGR2 (gen-num) nodes. Figure 4 illustrates the formation of the word khaayii:
Figure 4: Fusion of Aspect and AGR nodes
After vocabulary insertion of the entries khaa (eat) and yii the tree will generate the form khaayii (ate-singualar-feminine).
Few Other Operations
Fission splits a single morpheme into two (resulting into more number of morphemes), certain morpheme features can be modified or deleted using impoverishment (resulting into the insertion of a less specified vocabulary item).
Implications of Late Insertion, Underspecification, Syntactic Hierarchical Structure All the Way Down and Impoverishment
Halle and Marantz (Halle and Marantz, 1994) and Noyer (2006) explain in detail how the above-mentioned three core components of DM help avoiding overgeneration of word forms and in addition, reject assembling word into paradigm groups that allow only one word occupying each paradigm slot.
In lexicalist theories, as opposed to late insertion, the underspecified lexical entries in the lexicon compete to occupy slots in paradigm structures. Once inserted into a paradigm cells, these entries pick up full set of syntactic and semantic features that are already available in those cells. This results in the competition among lexical entries in which highly specified entries win to fill up a paradigm cell. This amounts to lot of redundancy and demands huge effort in creating word paradigms. This can be avoided if vocabulary items are allowed to insert late, as DM does. In DM, impoverishment deletes features before vocabulary insertion, rejects a set of vocabulary items that are specified for the deleted features, and finally picks up and inserts the least specified or the default item in the node of competition. The insertion of vocabulary items, as roposed by Halle (1997) is subject to Subset Principle that says
‘The phonological exponent of a Vocabulary Item is inserted into a morpheme... if the item matches all or a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the Vocabulary Item contains features not present in the morpheme. Where several Vocabulary Items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen.’
Halle and Marantz illustrate this by giving an example where two vocabulary items namely A and B compete for insertion at a node called X that has features F1, F2 and F3. Assuming that during the course of derivation and after syntax, one feature, say F2, is deleted by an impoverishment rule, i.e., F2 -> ɸ / [X Y]. The rule says that the feature F2 from the feature set of X whenever it is followed by Y. This can be understood by giving an example from Hindi.
Root subjunctive verb forms in Hindi are marked cumulatively for number and person using the following suffixes:
1st person, singular: /-uuM/
1st person, plural: /-eM/
2nd person (intimate), singular: /-e/
2nd person (semi-honorific), singular/plural: /-o/
2nd person (honorific), singular/plural: /-eM/
3rd person, singular: /-e/
3rd person, plural: /-eM/
Impoverishment rule:
person -> ø / V [plural] or [singular, honorific]
The rule says that for any verb that is marked for plural number need not be specified for person. This reduces the competition of the vocabulary items that wait for insertion at the target node. Impoverishment creates room for underspecified entries that are the most frequent ones and always act as default affixes. In Hindi root subjunctive verb forms, /-em/ appears as the default affix (elsewhere rule) and thus, is least specified for the features present at the target node. Thus, the number of vocabulary entries is reduced (which would otherwise be twelve; with all possible combinations of three persons (1st, 2nd and 3rd), two numbers (sg, pl) and three degrees of honoroficity (intimate, semi-hon and hon) in second person), as we do not need entries that are specified for all the features at the terminal node. The Vocabulary items that compete for insertion after applying the impoverishment rule are:
[2p, semi-hon] <-> /-o-/ / V _
[1p, -pl] <-> /-uuM/ / V _
[-pl] <-> /-e/ / V _
[+pl] <-> /-eM/ / V _
In cases where subset principle fails to make a choice for a vocabulary item, either the one that comes first in the order is picked (Halle and Marantz, 1993) or the Universal Hierarchy of Features is considered (Noyer, 1997). This means that the Vocabulary Item that has the feature that appears highest in the hierarchy is inserted.
The Hierarchy of Features
1 person > 2 person > dual > plural > other features
Vocabulary Items
A vocabulary item is a relation between a phonological string or 'piece' and information about where that piece may be inserted. These items provide the set of phonological features available in a language for the expression of abstract morphemes. The set of all Vocabulary Items is called the Vocabulary (Harley and Noyer, 1999). The vocabulary items are representing using a scheme that shows a phonological exponent or signal with the context where it can be inserted. See examples below (see :