Abstract
This paper considers the NP vs. DP debate from the perspective of dependency grammar (DG). The message is delivered that given DG assumptions about sentence structure, the traditional NP-analysis of nominal groups is preferable over the DP-analysis. The debate is also considered from the perspective of phrase structure grammar (PSG). While many of the issues discussed here do not directly support NP over DP given PSG assumptions, some do. More importantly, one has to accept the widespread presence of null determiner heads for the DP analysis to be plausible on PSG assumptions. The argument developed at length here is that the traditional NP-analysis of nominal groups is both more accurate and simpler than the DP-analysis, in part because it does not rely on the frequent occurrence of null determiners.
1 Introduction
Before the 1980s, the root/head of the nominal group was almost unanimously assumed to be the noun. 1 Since the 1980s, however, the position has been argued and has long since solidified among many that the determiner, not the noun, is the root/head of the nominal group (cf. Venneman 1977; Hudson 1984, 90–92, Hellan 1986; Abney 1987). The syntax community remains split on this issue, however. Within the tradition of phrase structure grammar (PSG), the dominant position today is that nominal groups are determiner phrases (DPs); the corresponding analysis is called the DP-analysis here. 2 There remains a significant number of prominent PSG people who did not, or who have not, adopted this position, however; these grammarians assumed, or continue to assume, the traditional NP-analysis of nominal groups (e.g. McCawley 1998; Huddleston & Pullum 2002; Sag et al. 2003; Culicover & Jackendoff 2005; Quirk et al. 2010).
Switching to dependency grammar (DG), the dominant stance was and still is that the noun, not the determiner, is the root of the nominal group. A majority of DG people assume the NP-analysis of nominal groups (e.g. Hays 1964; Robinson 1970; Kunze 1975; Schubert 1987; Mel'čuk 1988; Sgall et al. 1986; Starosta 1988; Engel 1994; Jung 1995; Heringer 1996; Bröker 1999; Groß 1999; Mel'čuk 2009; among many others). This is so even for approaches that vary drastically in other respects. For instance, computational linguists using dependency for the automated parsing of sentences are split into two broad camps concerning differences in the annotation schemes they employ; the one camp is associated with the Universal Dependencies (UD) project and its annotation scheme (https://universaldependencies.org/), and the other camp with the syntactic alternative to the UD scheme, the Surface-Syntactic Universal Dependencies (SUD) (https://surfacesyntacticud.github.io/). While these two camps disagree significantly in their annotation choices, they nevertheless agree concerning the status of determiners, which are consistently positioned as dependents of nouns in nominal groups. Worth noting as well is that there are also a couple of DG people who have assumed, and who may still assume, the DP-analysis of nominal groups (e.g. Hudson 1984, 90–92, 1990, 268–276; Hewson 1991; Lobin 1993, 38–64; Hudson 2003; Anderson 2006, 184–186, and most recently, Ninio 2019).
Given that the DP-analysis is more rarely assumed in DG circles than in PSG circles, the question arises as to why: Why has the DP-analysis of nominal groups gained little traction in the DG community? This paper endeavors to answer this question. It demonstrates that given DG assumptions about sentence structure, the DP-analysis of nominal groups is implausible. A number of linguistic insights, some of which are new, are brought to bear on the matter. The discussion also refutes some observations that have been deemed supportive of the DP-analysis within the context of DG, mainly those produced by Hudson (1984, 90–92; 1990, 268–271; 2003) and Ninio (2019). The discussion also considers a recent argument from Salzmann (2020) in the area in the context of PSG.
After examining the NP vs. DP debate within the context of DG, the discussion turns to PSG. The main point established is that many of the arguments produced in favor of NP in the context of DG do not transfer smoothly to PSG. The willingness to assume null determiners in PSG disarms many of the key arguments produced in the DG context. The point is emphasized in this regard, though, that positing null determiners in the syntax is a powerful device that opens up the PSG analysis of nominal groups to the objection of too much reliance on null elements. The take-home message is therefore that the DP-analysis of nominal groups is broadly implausible and that if one nevertheless chooses to assume DPs, then one should concede that doing so is only defensible in the event that one readily acknowledges the frequent occurrence of null determiners. In this respect, an approach that does without the null elements is certainly simpler, and if it can offer explanations of the relevant data that are as good or better than the approach with the null elements, then it should be preferred. This is Occam's Razor.
This paper is organized as follows. Section 2 frames the debate by considering some criteria employed for identifying heads more generally. Section 3 establishes some terminology that promotes the understanding of the NP vs. DP debate in the context of DG. Section 4 enumerates the arguments in favor of NP over DP. Section 5 scrutinizes and refutes some of the arguments that have been produced in the context of DG in favor of DP. Section 6 turns to PSG, reflecting on the extent to which the insights gained in the DG context are transferable to PSG. Section 7 then sketches how the relatively flat DG structural analyses of nominal groups can nevertheless accommodate the organization of functional categories in nominal groups. Section 8 concludes the paper.
2 Criteria for identifying heads
Identifying heads has long been a central goal of much syntactic theory. Various criteria and concepts have been proposed and developed as a basis for accomplishing this goal. The well-known exchange between Zwicky (1985) and Hudson (1987) illustrates well the use of such criteria. Zwicky proposed eight notions for distinguishing heads: 1. semantic argument, 2. determinant of concord, 3. morophosyntactic locus, 4. subcategorizand, 5. governor, 6. distributional equivalent, 7. obligatory appearance, and 8. ruler. Hudson then critiqued Zwicky's understanding and use of these criteria. He argued that some of the conclusions Zwicky had reached based on these criteria were open to debate and could in fact be interpreted in a different way.
No attempt is produced here to evaluate each of the criteria from the Zwicky-Hudson exchange, since doing so would not shed much light on the NP vs. DP debate. It can suffice to merely point out that Zwicky's and Hudson's interpretations of the eight criteria differed so much with respect to determiner+noun combinations that they essentially reached opposite conclusions. Zwicky interpreted most of the criteria as supportive of nouns as heads whereas Hudson interpreted most of them as supportive of determiners as heads. Many of Hudson's arguments in favor of DP will be scrutinized below in Section 5. At present what is important is the fact that attempts to establish a single absolute definition or principle for identifying heads in every case, or even a group of such definitions and principles (cf. Melˈčuk 2009, 25–33), has proven to be impossible and that attempts to do this cannot succeed in general. The intent here is therefore not to produce such absolute definitions or principles, but rather a more heuristic approach to identifying heads is pursued.
The danger of relying too much on a single (group of) notion(s) or diagnostic(s) for identifying heads is worth briefly illustrating. Let us take the omission test as an example. Omission is the diagnostic for shedding light on the sixth and seventh of Zwicky's notions, i.e. distributional equivalent and obligatory appearance. Omission is easily employed because it involves the simple omission of the one or other of the two words that are connected by a dependency and under scrutiny. If the resulting structure is acceptable, then the one word that remains is deemed to be head over the word that was omitted. Like other diagnostics for sentence structure, though, omission can deliver unexpected and contradictory results (cf. Hudson 1987, 122). Consider what omission reveals in the following cases:
| (1) | a. | The milk is white. |
| b. | Milk is white. | |
| c. | *The is white. |
| (2) | a. | The bottle is brown. |
| b. | *Bottle is brown. | |
| c. | *The is brown. |
When the noun at hand is a mass (or an indefinite plural) noun, omission suggests that the determiner depends on the noun as illustrated in (1a-c). When the noun at hand is a singular count noun, however, omission does not help one way or the other, since neither determiner nor noun can be omitted, as illustrated in (2a-c).
Consider next what omission reveals about the nature of an optional function word such as the subordinator that. At times that can be omitted without changing the meaning or affecting grammaticality, yet at other times it cannot:
| (3) | a. | Frank emphasized repeatedly (that) he was hungry. |
| b. | That he was hungry Frank emphasized repeatedly. | |
| c. | ??He was hungry, Frank emphasized repeatedly. |
The appearance of the subordinator that is optional in (3a), which suggests that it should be viewed as a subordinate of the verb was. This suggestion is contradicted, however, by examples (3b) and (3c), where the appearance of that is mandatory. The reason for this difference in optional vs. obligatory appearance is clear. In (3b-c), the presence of that is necessary to ensure that he was hungry is interpreted as an embedded clause as opposed to as the matrix clause. In sentence (3a), in contrast, its appearance is not necessary in this regard, since there is no potential for such a misanalysis. The point to these observations, then, is that the omission diagnostic is delivering inconsistent results and should therefore not be used without awareness of its potential for inconsistency.
“… although each of the criteria is useful in its own right, they nevertheless ought to be applied with a fair amount of caution. A better option may be to use a cluster approach, which basically means that the more criteria an element complies with, the more prototypical a head it is. Where two elements compete for headedness, the one fulfilling most criteria wins out. This will mean that none of the criteria will, by itself, be decisive; nor will the non-fulfillment of any one criterion mean that an element cannot be the head. More importantly, perhaps, it will mean that the result need not always be unequivocal: headedness will be regarded as a matter of degree; where differences in degree of headedness between two elements are small, the outcome may not always be clear, and the question of headedness remain unresolved …” (Keizer 2007, 21) 3
The rest of this paper examines the NP vs. DP debate with awareness of this sentiment. The expectation is that neither noun nor determiner will fulfill all criteria for identifying heads. However, the overall position adopted and argued for extensively below is that among the elements of nominal groups, the noun fulfills more head criteria than the determiner.
3 Preliminary DG matters
Before presenting the insights that favor NP over DP, some preliminary DG matters need to be addressed. Aspects of terminology use and of the DG assumed are established. The next two trees serve as the starting point:
| (4) |
|
The primary issue discussed and decided in this paper concerns these two competing analyses, that is, whether the noun is head over the determiner as in (4a) or vice versa as in (4b). The message is that the NP-analysis in (4a) is well motivated, whereas the DP-analysis in (4b) is not.
The term nominal group is used throughout this paper to denote phrases that, depending on one's assumptions, can be viewed as rooted/headed by either a noun (the NP-analysis) or a determiner (the DP-analysis). The intent in this regard is to employ terminology in a way that promotes a neutral vantage point from which the NP vs. DP debate can be evaluated. Another important term in the current DG is phrase; any complete subtree consisting of two or more words is a phrase. The discussion here also draws an important distinction between heads and roots. The head of a given phrase is the one word outside of that phrase that immediately dominates the phrase. In contrast, the root of a given phrase is the one word within the phrase that dominates all the other words in the phrase. Thus, the root of the phrase the house in (4a) is house, and the root of the phrase the house in (4b) is the. The head of the house is not shown in (4), because it necessarily appears outside of the house. Observe, however, that house is head over the in (4a), and the is head over house in (4b). 4
Most DGs assume a one-to-one ratio of words to nodes. Each word corresponds to exactly one node in the dependency structure, and vice versa. Some DGs are explicit about this, stating directly that dependency is a strict one-to-one mapping of words to nodes (e.g. Mel'čuk 1979, 76; Mel'čuk & Pertsov 1987, 48, 57–58; Schubert 1987, 78–86, 129; Engel 1994, 25, 28; Kahane 1996, 45; Bröker 2003, 297; Hudson 2003, 520, 2007, 183; Carnie 2010, 177). This one-to-one mapping is evident in (4a) and (4b), where there are two words and two nodes present each time. The next dependency tree further illustrates this one-to-one mapping:
| (5) |
|
There are eight words and one clitic in this tree, which means there are 9 (= 8+1) atomic units of syntax present. This matches the number of nodes in the structure, namely 9. The completely vertical, lightly dotted projection lines point clearly to the strict one-to-one mapping: each word (or clitic) is mapped to a single node, and vice versa. Observe that the analysis given as (5) assumes NP (not DP), since the determiners the majority leader's and its are positioned as dependents below their respective nouns.
The status of possessive ’s is particularly noteworthy and important for the discussion of determiners below; it plays a central role in a couple of the arguments in favor of NP and against DP. Example (5) illustrates key assumptions about the nature of possessive ’s: it is a clitic, which means it is prosodically dependent on a host. At the same time, possessive ’s is syntactically independent, which means it occupies its own node in the syntactic structure. The status of possessive ’s as a clitic is indicated in (5) in two ways: the hyphen to the immediate left of s indicates that ’s cliticizes to the word that immediately precedes it, and the absence of a lightly dotted projection line extending up to s indicates that s lacks word status. The analysis in this paper of possessive ’s and of reduced auxiliary verbs such as ’s and ’ve (see examples 42a-c below) follows the approach to clitics in Groß (2014).
The strict one-to-one mapping of dependency syntax is important for the NP vs. DP debate. Strict adherence to one-to-one mapping discourages DGs from positing phonologically null nodes. What this means is that the NP vs. DP debate is simply stated, since there are just two main possibilities for the analysis of determiners, as in (4a) or (4b). Phrase structure syntax, in contrast, can and usually does assume more nodes than words. The fact that nodes already outnumber words in phrase structures means that increasing the number of nodes with phonologically null elements does not strike one as problematic. The discussion returns to this point below in Section 6.
A final preliminary matter for the discussion concerns language-specific aspects of the debate below. Most of the arguments produced below for and against NP or DP are from English, and therefore the data concerning determiners in English are particularly relevant. Table 1 provides an overview of the words that qualify as determiners in English.
List of determiners and corresponding pronouns in English
| Determiner | Pronoun form | Examples |
| a | one | You own a house, and I also own *a/ ✓ one. |
| any | any | He won't eat any fish, and she also won't eat any. |
| both | both | Both boys were present, and both were hungry. |
| each | each | Each boy was present, and each was hungry. |
| either | either (of…) | Susan has two cats, but I don't like ? either / ✓ either of them. |
| every | every one of… | Every boy was present, and *every / ✓ every one of them was hungry. |
| her | hers | His brother was present, and *her/ ✓ hers was, too. |
| his | his | Her brother was present, and his was, too. |
| my | mine | Your brother came, and *my/ ✓ mine, too. |
| neither | neither (of…) | Susan has two cats, but I like ? neither / ✓ neither of them. |
| no | none | You own no pets, and I also own *no/ ✓ none. |
| our | ours | Your pets are friendly, and *our/ ✓ ours, too. |
| …’s | …’s | Jill likes John's dog, and John likes Jill's. |
| some | some/one | You bought some milk, and I also bought some/*one. |
| I heard some dog barking, and you also heard *some / ✓ one barking. | ||
| that | that (one)5 | This is tasty, and that, too. |
| This cat is friendly, and *that / ✓ that one, too. | ||
| the | the one | The cat over there is friendly, and *the / ✓ the one over here, too. |
| these | these | Those cats are friendly, and these, too. |
| their | theirs | Your dogs are friendly, and *their/ ✓ theirs, too. |
| this | this (one) | That is tasty, and this, too. |
| That cat is friendly, and *this/ ✓ this one, too. | ||
| those | those | These cats are friendly, and those, too. |
| what | (what N) | What claim bothers her, and ?? what / ✓ what claim bothers him? |
| which | which (one) | Which cat did he see, and ? which / ✓ which one did she see? |
| whose | whose | Whose cat is black, and whose is gray? |
| your | yours | My cat is friendly, and *your/ ✓ yours, too. |
The distinction between determiner and pronoun included in this table will play a role in the argumentation below in a couple of areas. Moreover, the role of possessives (e.g. her/hers, your/yours, etc.) is particularly relevant, since a number of the arguments below in favor of NP employ insights gained from the distribution of possessive forms. 5
Possessives are determiners in English and some related languages. This fact is evident in the following data set:
| (6) | a. | the cat / a cat / each cat / this cat / etc. |
| b. | *the a cat / *each this cat / *the this cat / *each the cat / etc. | |
| c. | Bill's cat / his cat / our cat / my cat / etc. | |
| d. | *the Bill's cat / *a his cat / *our each cat / *my this cat / etc. |
These examples establish that typically only a single determiner can introduce a noun in English and that possessives (including the Saxon ’s) are clearly determiners in this regard. Possessives are, though, unlike other determiners in that they are semantically loaded in a way that other determiners are not. Possessive determiners (e.g. his, her, our, their, etc.) are fully referential, picking out a referent in the world of discourse. The same is certainly true of complex determiner phrases such as the woman wearing a hat's, my old friend's, etc. Other determiners (e.g. the, a, etc.) are not referential in the same way.
4 In favor of NP
The following eight subsections present observations and arguments in favor of NP. The first two of these subsections are concerned with reasoning that is more semantic than syntactic in nature, whereas the latter six build on insights that are more syntactic than semantic – insofar as distributional criteria are key. Section 5 then presents and refutes eight arguments that have been produced in favor of DP. Some of the arguments below in favor of NP can be found in varying forms in Payne (1993), Van Langendonck (1994), and Huddleston & Pullum (2002).
4.1 Selection
Huddleston & Pullum (2002, 357–358) state two motivations for their decision in favor of the NP-analysis of nominal groups. The first of these is selectional restrictions and the second is determiner-less nouns. Selectional restrictions are considered in this section, and determiner-less nouns are considered in Section 4.5 below.
Predicates select their arguments. Selection has both a semantic and a syntactic component. To illustrate the general semantic component of selection, consider Tesnière's and Chomsky's famous sentences concerning the autonomy of syntax:
| (7) | Le | silence | vertébral | indispose | la | voile | licite. | (Tesnière 1966, 40–42) |
| the | silence | vertebral | indisposes | the | sail | licit | ||
| ‘The vertebral silence indisposes the licit sail.' | ||||||||
| (8) | Colorless green ideas sleep furiously. | (Chomsky 1957, 15) |
These sentences are nonsensical because they violate the selectional restrictions of the predicates involved. The predicates vertebral, indispose, and licite in Tesniére's sentence and the predicates colorless, green, and sleep in Chomsky's sentence are combined with semantically inappropriate arguments. While one can argue that these sentences are syntactically well formed, there is no dispute about the fact that they are quite nonsensical.
There is also a syntactic component to selection, and it is in this area that we encounter reasoning that supports NP over DP. Consider first the manner in which certain predicates require the presence of a particular preposition. Certainly verbal predicates can demand the presence of a specific preposition (e.g. laugh at, work on, think over, etc.) and so can adjective predicates (e.g. proud of, interested in, frustrated with, etc.). These sectional restrictions are accommodated in the syntax by subordinating the preposition directly to the verb or adjective. Concerning nominal groups, the inclination is also to establish a direct dependency between the words that influence each other in terms of selection. Thus, the relevant question is whether a given predicate selects the noun of a nominal group, or the determiner. The answer should be clear: it is the noun that the predicate selects.
Huddleston & and Pullum (2002, 357–358) illustrate the point with the predicate assassinate and the noun dog. They observe that a prominent person can be assassinated, but a dog is killed or put down. What this means is that the transitive verb assassinate places semantic restrictions on its object noun. It does not, however, put semantic restrictions on the determiner introducing the object noun. The next dependency analyses help illustrate the point:
| (9) |
|
The fact that on the NP analysis, assassinated and president are directly linked to each other by a dependency accommodates the manner in which assassinated puts semantic restrictions on the noun president. On the DP-analysis, however, the direct link is no longer present. In this regard, the DP-analysis must stipulate that the determiner the is transparent to these semantic restrictions; it somehow conveys the semantic restrictions coming from the verb assassinate above to the object noun president below.
While the DP analysis can reach to this stipulation – i.e. that the determiner is transparent and hence conveys the selectional restrictions downward to its complement – this stipulation becomes implausible when the determiner at hand is no longer a clear function word, but rather has referential content, such as a possessive, e.g.
| (10) |
|
The notion that the possessive his, which has full referential content, is transparent to selectional restrictions coming from above is implausible. The NP-analysis therefore is preferable in these cases, since it allows the link between the verb and object noun to be direct, the former semantically selecting the latter.
The reasoning just illustrated with respect to the verb assassinate and its complement argument extends to predicates and arguments of all types. Selectional restrictions are accommodated best in syntax if the dependencies that connect semantically and/or idiosyncratically related words are as direct as possible. The NP-analysis of nominal groups allows these dependencies to be more direct than the DP-analysis.
4.2 Parallelism across semantics and syntax
Haegeman & Guéron (1998, 411–417) establish that there is a measure of parallelism across clause and nominal group in the distribution of predicates and their arguments. The predicate-argument structures of simple clauses can appear in nominal groups as well. In dependency syntax, the NP-analysis can preserve this parallelism in predicate-argument structures, whereas the DP-analysis destroys it (cf. Van Langendonck 1994, 253). A general aspect of dependency syntax that has long been acknowledged is its closeness to semantic organization in terms of predicates and their arguments. Thus, the NP-analysis helps preserve this closeness, whereas the DP-analysis increases the distance between syntactic and semantic analyses.
Consider the structure of a simple sentence and the competing analyses of the corresponding nominal group:
| (11) |
|
The semantic parallelism across clause and nominal group is obvious here. The following alignments is are present:
| (12) | Predicate: invaded or invasion |
| Agent argument: Caesar or Caesar's | |
| Patient argument: Gaul or of Gaul's |
The NP-analysis in (11b) has the predicate-argument structure of the NP matching that of the clause in (11a). In both (11a) and (11b), the predicate dominates both its arguments, the subject argument appearing as a pre-dependent and the object argument as a post-dependent. In contrast, the DP-analysis in (11c) destroys the parallelism, since it positions the subject argument Caesar's in such a manner that the predicate invasion does not dominate it.
While examples (11a-c) have been constructed to effectively illustrate structural parallelism (or lack thereof), the parallelism is largely maintained when the arguments are shuffled, for instance with passive voice in the clause, e.g.
| (13) |
|
The DP-analysis shown as (13c) follows Hellan (1986, 99–102) and Hudson (1990, 276–282) in positioning possessive ’s as the head/root of the entire nominal group. Assuming the matrix predicate in (13a) is was invaded (not just invaded), then the NP-analysis again maintains the semantic parallelism in predicate-argument organization across clause and nominal group insofar as the matrix predicate consistently dominates both its arguments. The DP-analysis, however, destroys the parallelism because the predicate invasion does not dominate its argument Gaul's.
The parallelism in predicate-argument structures just examined is likely important for the DG analysis of various phenomena of syntax, such as scope (cf. Osborne & Reeve 2018) and binding. Correct choices in this area are hence crucial for developing DG accounts of these phenomena.
4.3 Proform substitution
Proform substitution is one of the most widely employed diagnostics for identifying sentence structure (e.g. Carnie 2010, 19–20; Quirk et al. 2010, 75–7; Miller 2011, 54–5; Sobin 2011, 32; Carnie 2013, 98; Denham and Lobeck 2013, 262–5; Sportiche et al. 2014, 50; Müller 2016, 8). If a given string can be replaced by a definite proform, then that string is deemed a complete subtree in the starting sentence. Consider (14):
| (14) |
|
The strings the girls, the boys, at the pool, and Saturday morning are deemed complete subtrees in (14a) by virtue of the fact that they can be replaced by they, them, there, and then, respectively. Proform substitution used in this manner is one of the most widely employed tests for identifying constituents. Since the complete subtrees of dependency structures correspond to the phrasal constituents of PSGs, proform substitution is also an insightful means of identifying dependency structures.
When applied to phrases rooted/headed by possessive ’s, proform substitution delivers evidence in favor of NP. Observe the next two competing analyses of the nominal group the woman with a hat's dog:
| (15) |
|
These two analyses agree insofar as possessive ’s is head over the woman with a hat. They disagree, however, regarding the placement of the noun dog, as head over ’s or as a dependent of it. When one checks to see how this phrase behaves with respect to proform substitution, the proform her can replace the woman with a hat's, resulting in her dog. On the NP-analysis, her has replaced a complete subtree, whereas it has replaced a string that is not a complete subtree on the DP analysis.
To further establish the point, consider the competing dependency analyses of the phrase Frank's older sister:
| (16) |
|
| (17) |
|
On the NP-analysis, the proforms his and she replace complete subtrees. On the DP-analysis, in contrast, the proform his replaces a string that is not a complete subtree. Furthermore, she in (17c) replaces a complete subtree the root of which is possessive ’s. In this regard, the expectation should be that the correct proform to replace the entire nominal group would be the possessive pronoun his, but this is clearly incorrect, as demonstrated next:
| (18) | Frank's older sister arrived first. | |
| a. | She is a mother of two. | |
| b. | *His/*Hers is a mother of two. | |
The personal pronoun she can easily refer back to and hence replace the complete nominal group Frank's older sister. The possessive pronoun his or hers in contrast, cannot do the same.
The message, then, is that one of the most widely employed tests used to identify sentence structure, i.e. proform substitution, supports the NP-analysis over the DP-analysis. Van Langendonck (1994, 247–248) delivers a similar message, but concerning interrogative proforms – his question-word test.
4.4 Coreference and binding
The phrase his older sister references two individuals, the one being a male individual and the other a female individual. Syntactic analyses typically indicate these referential possibilities with indexes. The relevant question in this regard concerns the referential index of the entire phrase: does this index correspond to the male or female individual? Observe the competing analyses:
| (19) |
|
Unlike the DP-analysis, the NP-analysis is consistent with the referential possibilities of the entire phrase. This phrase points to the female individual, not to the male individual. The root of the phrase on the NP-analysis is sister, a word denoting a female individual, whereas the root on the DP-analysis is his, a word denoting a male individual. The NP-analysis is therefore consistent with the referential possibilities; the referential status of the root node determines the reference of the entire phrase (cf. Salzmann 2020: 21).
The importance of this distinction in referential potential becomes evident when considering basic binding data.
| (20) |
|
These examples demonstrate that the nominal group his girlfriend should bear the index corresponding to girlfriend, not to his. The NP analysis is consistent with this state of affairs since it positions the noun girlfriend as the root of the subject phrase. If the possessive his were the root of the subject phrase as shown in (20b), we would expect himself to be acceptable instead of herself.
The same point is also evident with the simple pronoun:
| (21) | Hisi girlfriendj praises *herj/✓himi. |
In this case, coreference is possible with his, but not with his girlfriend. This is expected on the NP analysis, since his is embedded inside the subject phrase and is hence not a co-argument of the object. To state the point in other words, the reflexive pronoun is obligatory in such cases when the co-indexed expressions are co-arguments, whereas the simple pronoun appears when the co-indexed expressions are not co-arguments. Unlike the DP analysis, the NP analysis is consistent with these facts.
In Government and Binding Theory (GB) (Chomsky1981, 1986), data such as (20–21) here give the basic facts around which the traditional binding theory is constructed. The subject NP c-commands the object and everything inside the object, but not vice versa. Anything embedded inside the subject does not, however, c-command the object. The traditional binding theory was established assuming NPs, not DPs.
4.5 Determiner-less nouns
McCawley (1998, 214) and Huddleston & Pullum (2002, 357–358) point to determiner-less nouns as a source of motivation for the NP-analysis. Many languages lack definite and indefinite articles, and even in languages that have them, determiner-less nouns occur frequently. Mass nouns and indefinite plurals often appear without a determiner in English. From a DG point of view, the frequent occurrence of determiner-less nouns supports NP over DP in a straightforward manner. In particular, stating subcategorization generalizations is simpler on the NP-analysis, since nominal groups are consistently NPs, never DPs (cf. Payne 1993, 129–130).
The next examples illustrate what is at issue. The N (= noun or pronoun) and the D (= determiner) draw attention to the category status of the complement of the verb:
| (22) |
|
| (23) | a. | The kids played in the water. |
| b. | The kids played in water. |
A transitive verb like order consistently takes a noun (or pronoun) such as lobster as its dependent. Similarly, a preposition like in consistently takes a noun (or pronoun) like water as its dependent. If determiner dominates noun, however, then at times the verb order takes a determiner like the as its immediate dependent and at other times, it takes a noun (or pronoun) such as lobster. Similarly, at times a preposition like in takes a determiner like the as its dependent, and at other times it takes a nominal like water as its dependent.
Consider the matter also from the other point of view, i.e. from that of the determiner or noun. If noun dominates determiner, then a determiner consistently takes a noun as its head. If determiner dominates noun, however, then at times the determiner takes a verb as its head and at other times it takes a preposition as its head, yet at other times, it takes potentially some other category as head.
The DP-analysis might address the issue by positing the presence of a covert determiner in all those cases where an overt determiner is absent, and certainly many PSGs do exactly that. Such a move can establish consistency in subcategorization requirements insofar as nominal groups consistently qualify as DPs. Such a move complicates the grammar in another way, however, namely by adding an additional syntactic element, a null determiner. In contrast, DGs have traditionally rejected such phonologically null elements (except in cases of ellipsis). 6 The discussion returns to this point below in Section 6.
4.6 Possessive determiners vs. pronouns
There are two forms of many determiners/pronouns in English and related languages. The one form appears when the word is functioning as a pure determiner, that is, when it is introducing a nominal group, and the other form occurs when the word is functioning as a pronoun, meaning that there is no common noun following it. This distinction occurs with the possessives. Table 2 shows the distinct forms:
Possessive determiners and corresponding pronoun forms in English
| Possessive determiner | Possessive pronoun |
| her | hers |
| my | mine |
| our | ours |
| their | theirs |
| your | yours |
An examination of the distribution of these competing forms casts light on the NP vs. DP debate.
The NP-analysis easily accounts for the distinct forms of these words. Possessive determiners appear when the head word is a common noun, whereas the pronoun occurs in all other cases. The next trees illustrate this state of affairs:
| (24) |
|
| (25) |
|
On the NP-analysis, the distribution of the distinct forms is a function of the head word. When the head is a noun, such as dog here, the possessive determiner appears. When the head word is a verb (or a word of some other non-noun category), such as brought here, the possessive pronoun appears instead. This simple explanation accounting for the distribution of the distinct forms does not work on the DP-analysis, since in both cases, the head of the possessive form is a verb (or a word of some other non-noun category), as is the case in (25a-b) with brought.
An aspect of possessive pronouns that is worth considering further is the fact that the affixed -s many of them contain (hers , ours , theirs , yours ) is likely essentially the same element as the clitic -s of complex determiner phrases (e.g. Bill's , the woman's , etc.) – see example (5) above. The structural analyses of these two types of possessive -s can also be essentially the same assuming that one acknowledges intra-word dependencies (cf. Groß & Osborne 2013). In both cases, the -s takes a predependent and can turn the whole into a pronoun, e.g.
| (26) |
|
The heavily dotted dependency edge connecting her to -s identifies that dependency as existing within the word (as opposed to between words). Groß & Osborne develop this sort of approach to intra-word structure extensively.
Returning to the DP-analysis, it must look down the hierarchy to see if a complement noun is present (cf. Hudson 1987, 123) instead of appealing to the category status of the head word. It assumes the determiner form appears if there is a dependent noun present, otherwise the pronoun appears. The problem with this line of reasoning is that other forms that optionally take a complement do not vary in form in this manner. For instance, many verbs are optionally transitive in English, yet the form of the verb does not vary depending on whether a complement is present or absent, e.g. We have eaten pizza vs. We have eaten. Consider prepositions in this regard. When they combine with a motion verb, the complement of the preposition often appears optionally, yet the form of the preposition never changes, e.g. Please come in (the room), You should climb up (the tree), Get out (of here)!, When are you going to stop by (my place)?.
One might object, as an anonymous reviewer has, that verb prefixes in many languages (such as German and French) change the valency of the verb, rendering it obligatorily transitive. The prefix be- in German is a case in point, since it is often viewed as a transitivizing prefix. Interestingly, however, the impact that be- actually has on the verb is to change the type of complement that it takes, as demonstrated with the examples in Table 3:
Impact of prefix be- on the meaning and valency of verbs in German
| Verb without be- prefix | Verb with be- prefix |
| (auf etwas) antworten ‘to reply (to something)’ |
etwas beantworten ‘to answer something’ |
| (an etwas) arbeiten ‘to work (on something)’ |
etwas bearbeiten ‘to work something over’ |
| etwas (irgendwohin) setzen ‘to set something somewhere’ |
etwas besetzen ‘to occupy something’ |
| (mit jmdn) sprechen ‘to speak (with someone)’ |
etwas besprechen ‘to discuss something’ |
| (irgendwohin) gehen ‘to go somewhere’ |
etwas begehen ‘to walk along something’ |
The meaning of the verb changes when the be- prefix appears, which means be- cannot be viewed purely as a transitivizing prefix. Furthermore, the verb without be- already takes a complement. The verbs antworten, arbeiten, sprechen, and gehen can take a prepositional complement, meaning that the prepositional complement is optional. The verb setzen takes two complements, one of which is obligatory. Crucially, the form of the former four verbs does not change based upon the presence/absence of the prepositional complement, a fact that further supports the claim here. 7
4.7 Fronting of pre-noun modifiers
Ross (1967/1986, 127–134) observed that a determiner cannot be separated from its noun. He characterized the insight in terms of his Left Branch Condition. A constituent on a left branch underneath a noun cannot be moved away from that noun. Ross' insight is also valid for attributive adjectives. These adjectives appear on left branches below their nouns and accordingly, they cannot be separated from their nouns. The NP-analysis accommodates these facts in a straightforward manner because it positions both determiners and attributive adjectives on left branches below their nouns. The DP-analysis, in contrast, loses the generalization because determiners are no longer on left branches below their nouns.
The next two data sets establish the extent to which determiners and attributive adjectives behave the same with respect to fronting (topicalization and wh-extraction). Neither category can be fronted away from their nouns:
| (27) | a. | Frank does like your big house. |
| b. | *…but your Frank does like big house. | |
| c. | *…but big Frank does like your house. | |
| d. | *…but your big Frank does like house. | |
| e. | *…but house Frank does like your big. | |
| f. | *…but big house Frank does like your. | |
| g. | …but your big house Frank does like. |
| (28) | a. | Frank likes which big house? |
| b. | *Which does Frank like big house? | |
| c. | *Which big does Frank like house? | |
| d. | *Which house does Frank like big? | |
| e. | Which big house does Frank like? |
These examples illustrate that fronting mechanisms, i.e. topicalization and wh-fronting, cannot separate determiners and attributive adjectives from their nouns. They also show that nouns cannot be fronted away from their determiners and/or attributive adjectives. If the one or other is fronted, then they all must be fronted together.
These facts are compatible with the NP analysis, but problematic for the DP analysis. On the NP-analysis, a simple generalization captures the facts. This generalization is that pre-dependents of nouns cannot be displaced away from their nouns. Or, taking the opposite perspective, a head noun cannot be displaced away from its pre-dependents. The DP analysis, in contrast, must reach to more than one stipulation to account for the same facts. It must state that a determiner cannot be displaced away from its dependent noun or that the dependent noun cannot be displaced away from its head determiner, and that an attributive adjective cannot be displaced away from its head noun or that a head noun cannot be displaced away from its dependent attributive adjective.
The next two trees illustrate the competing analyses:
| (29) |
|
On the NP-analysis, both your and big are on left branches below house. On the DP-analysis, in contrast, the one is head over the noun, and the other a dependent below it. The point, then, is that the DP-analysis forces one to augment Ross' Left Branch Condition to address the relevant data. It necessitates a more complex account of the empirical facts.
There is hence again evidence that the grammar is simpler if one assumes NPs instead of DPs. Generalizations about discontinuities can be discerned and expressed more efficiently given NPs.
4.8 Pre- and post-noun positioning
In many languages, possessives can precede or follow their noun (cf. Payne 1993, 130–132). This point is established here by examining possessive -s in German (cf. Van Langendonck 1994, 251–252 concerning -s in Dutch). Possessive -s in German is an affix, not a clitic as it is in English; it appears on proper nouns marking them with genitive case. German orthography therefore writes the -s without an apostrophe. The interesting thing about possessive -s in German is the flexibility in position. The possessive can precede or follow the common noun, e.g.
| (30) | a. | Franks Haus | ‘Frank's house’ |
| b. | das Haus Franks | ‘the house of Frank's’ | |
| c. | *das Franks Haus | ‘the Frank's house’ | |
| d. | *Franks das Haus | ‘Frank's the house’ |
These examples establish a couple of important facts. One of these facts is that Franks in (30a, c, d) is a determiner and as such it competes with das ‘the’ for the one pre-noun position that accepts a determiner. This competition results in robust ungrammaticality in (30c) and (30d).
The argument in favor of NP is present in the ability of Franks to precede Haus as in (30a) or follow Haus as in (30b). When it follows Haus, I think most would agree that it is a dependent of Haus (and not head over Haus). The argument in favor of the NP-analysis should be apparent. On the NP analysis, Franks is consistently an immediate dependent of Haus. The two phrases, which are basically synonymous, are hierarchically consistent insofar as in both, Franks is a dependent of Haus. In contrast, the DP-analysis would have Franks as head over Haus in (30a), but probably as a dependent of Haus in (30b). Note that the difference in meaning across (30a) and (30b) is merely a matter of style.
To make the point entirely clear, the DP analysis would have to choose between one of the following two possibilities:
| (31) |
|
The second analysis here is implausible because it would necessitate acknowledging two determiners in the nominal group, the one (i.e. das) being a determiner that precedes the noun, and the other (i.e. Franks) a determiner that follows the noun. That leaves (31a) as the best analysis assuming DPs. But the problem with (31a) should be evident, since it positions a determiner that follows its noun as a dependent of the noun instead of as head over the noun.
This problem does not arise for the NP analysis, e.g.
| (32) |
|
The possessive Franks is consistently a dependent of Haus. There is hence no inconsistency in its hierarchical status depending on whether Franks precedes or follows the noun Haus. The point, then, is that the NP analysis allows for consistency of structural analysis. It is therefore simpler and more efficient, and hence better.
Note that this argument is valid in other languages, for instance in Dutch where the data are much the same as in German.
5 In favor of DP
The following eight subsections present and refute arguments that have been produced in favor of DP, mostly within a DG context. The arguments considered are largely syntactic in nature (as opposed to semantic) and are from Hudson (1984, 1987, 1990, 2003), Ninio (2019), and Salzmann (2020). Note that Hudson (2003) alters his position from earlier, arguing that determiner and noun are in fact interdependent – rather than the noun being dependent on the determiner outright.
5.1 Parallel to aux-verb combinations
“The relation between a determiner and its common-noun is in many respects similar to that between a polarity-verb and the next verb – e.g. the first word carries much less semantic content than the second, the first word is sometimes required for purely syntactic reasons, the second word is generally optional, the two words both define different parts of the meaning of the same concept, both words belong to the same more general word-type. Since the polarity-verb is head of the next verb, it is natural to assume the same for the determiner in relation to the common-noun.” (Hudson 1990, 271)
The parallelism Hudson is pointing to is concerned with the hierarchical relationship between functional and lexical material. Determiners, many of which (but not all!) contribute functional information, should appear above nouns, which contribute lexical information, similar to the manner in which auxiliary verbs appear above content verbs.
The parallelism is most evident when considering verb chains that include one or more auxiliaries, e.g.
| (33) |
|
On the DP-analysis, the distribution of functional and lexical material in the nominal group the work is parallel to the distribution of the similar material in the verb chain is working. Just like the function verb is appears as head over the lexical verb working, the determiner the appears as head over the lexical noun work on the DP-analysis in (33c), but not on the NP-analysis in (33b).
The parallelism pointed to is indeed present (in part) with many determiners. Possessive determiners, however, cancel the parallelism. Possessive determiners such as my, your, his, their, etc. are semantically loaded in a way that other determiners are not – as stated above at the end of Section 3. These determiners have full referential content and can serve as arguments of the noun, e.g. your explanation, my description, etc. Possessive ’s is even such that it can combine with a phrase that contains an entire clause, e.g. the people who have children's concerns. In such cases, the determiner cannot be interpreted as contributing just functional information in a manner that is similar to how auxiliary verbs contribute just functional information.
The counterargument, then, is that the parallelism Hudson points to is not complete. Possessive determiners in fact undo the parallelism.
5.2 Optionality of common noun
“The criterion of optionality points clearly to any as the head of student in any student (cf. I haven't seen any/any student/*student). Similar remarks apply to all those determiners after which the common-noun is optional, which means the vast majority of determiners as we have just seen.” (Hudson 1990, 271)
Hudson's point here oversimplifies the data in the area. Determiners in English are an idiosyncratic bunch. Many of them can indeed function straight up as pronouns, yet others cannot function as pronouns, and yet others experience a change in form when they function as pronouns, possessive forms being a case in point – see Section 4.6.
Hudson does have a point, though, if the focus remains on those determiners that do not change at all when they no longer function as determiners but rather as pronouns. The next examples illustrate Hudson's reasoning in this area:
| (34) |
|
On the DP-analysis, the root of the nominal group is consistently any, whereas on the NP-analysis, the root varies, i.e. fish in (34a) but any in (35b). In this respect, the DP-analysis does seem preferable. However, Hudson's argument in this area comes up short for all the determiners that vary in form depending on how they function, i.e. as determiners or pronouns.
Table 1 above provides an overview of determiners in English and of the extent to which they can function as pronouns. The determiners a, the, every, and no cannot function as pronouns. With the exception of his, the possessives vary in form systematically depending on whether they function as determiners or pronouns. The forms either and neither can function as pronouns, but when they do, they prefer to take an of-complement PP (or one). The singular quantifier some cannot function as a pronoun unless it denotes a mass or an indefinite plural quantity, e.g. He bought some bicycle, and she also bought *some/ ✓ one. When used anaphorically (as opposed to deictically), the singular demonstratives this and that cannot generally be used as pronouns, e.g. This cat is friendly, and *that / ✓ that one, too – a major exception in this area is the anaphoric use of this or that to refer back to an entire proposition, e.g. They all arrived hungry, and this was a problem, in which case the appearance of one is not possible, e.g. *They all arrived hungry, and this one was a problem.
In sum, the strength of Hudson's argument is weakened by the idiosyncratic nature of determiners and the corresponding pronouns. The numerous cases in which the pronoun form is distinct from that of the determiner directly contradict his reasoning.
5.3 Optional of-phrase
“Many of them [determiners] can be followed by an optional of phrase (e.g. all/three/which of the boys), with the lexical noun in a clearly modifying position, inside the of phrase, and this fact has led other linguists (e.g. Jackendoff 1972, Hogg 1977) to adopt an analysis for at least some determiners similar to the one I am proposing here.” (Hudson 1984, 91)
Hudson's statements here point to the only plausible analysis of such cases: the of-phrase is a post-dependent of the determiner, the determiner itself having taken on a pronominal function, that is, it has become a pronominal determiner.
While Hudson's analysis of such cases seems indisputable, the relevant question is whether such cases provide evidence for the analysis that the determiner is also head over the noun when of is not present, i.e. in cases such as all boys, three boys, which boys, etc. The point of debate is schematized as follows:
| (36) |
|
Hudson's point is that since which is clearly head over boys in (36a), so too it should dominate boys when of is not present in (36c).
Hudson's reasoning in favor of DP in this area is unconvincing, as argued by Van Langendonck (1994, 245–246) using data from a number of languages. The appearance of the preposition of in cases such as (36a) is an indication that the determiner is no longer functioning as a determiner, but rather it has become a pronoun. It's status as a pronoun motivates the appearance of of. Observe in this regard that when the nominal at hand is not a common noun but rather a definite pronoun, the of-phrase is obligatory, e.g. *any us vs. ✓ any of us, *neither you vs. ✓ neither of you, *which them vs. ✓ which of them. The NP-analysis accounts for this fact: the determiner cannot attach as a pre-dependent to a definite pronoun, but rather it must become a pronoun itself in such a manner that it can take the definite pronoun in a post-dependent of-phrase.
5.4 Locus of important properties
“The determiner is the locus of many properties which are crucial to the whole phrase's distribution – negation (no student), person (you students), definiteness, number (these students, some sheep), distributivity (every student).” (Hudson 1990, 271)
The counterargument in this area is that nouns too are the locus of similar important properties. They are the locus of number (person vs. people), countability and hence distributivity (time vs. season, love vs. lover), relationality (e.g. father, mother, top, bottom), and argument-assignment (Caesar's invasion of Gaul). The question in this area is therefore as follows: Which properties are most important for determining the distribution of the nominal group as a whole? A clear answer to this question is not apparent.
5.5 Position on left periphery
“4. The determiner is always on the periphery of the noun-phrase, which is easy to explain if it is the head of the common-noun and all other words are dependents of the latter, but much more puzzling if the determiner is yet another dependent of the common-noun.
5. The position of the determiner before the common-noun follows automatically if we take the latter as its complement, given that complements in English always follow their heads.” (Hudson 1990, 271)
The idea is that if the determiner, the left-most word in the nominal group, is the root of the phrase, then the architecture of the phrase is straightforward. The noun and all its dependents appear as a post-modifier of the determiner.
There are two problems with this argument in favor of DP. The first is that there is a class of systematic exceptions to the observation that the determiner is the left-most word in the nominal group. These exceptions occur in cases of the so-called big mess construction (BMC) (cf. Berman 1974; Van Eynde 2007; Kim & Sells 2011), e.g. so friendly a cat that…, that large a car, too big a mess to…, etc. The determiner in such cases is the indefinite article a, yet a does not appear at the left periphery of the nominal group, but rather it is preceded by the adjective.
The second problem facing Hudson's argument is more serious; it has to do with the recursive application of the reasoning. If the position of a given word at the left-periphery of a phrase is motivation for viewing that word as the root of the phrase, then all phrases should be head-initial, i.e. right-branching. Applied to a nominal group such as the two old friendly cats, this reasoning would proceed as follows:
| (37) |
|
Since the appears at the left-periphery of the entire phrase, it must be the root of the phrase as in (37a); since two is at the left-periphery of the next smaller phrase, it must be the root of that phrase as in (37b); since old is at the left-periphery of the yet next smaller phrase, it must be the root of that phrase as in (37c); and so on. Consistent application of the principle should therefore result in all phrases being entirely right branching, as is the case here in (37d). The analysis in (37d) is, though, implausible in a number of respects.
The point just established reveals that in a language such as English, which arguably has root-initial, root-final, and root-medial phrases, the position of a given word at the periphery of a phrase cannot be interpreted as an indication that that word is the root of the phrase. Linguistic observations of various sorts identify roots, not the linear position in the string of words constituting the phrase.
5.6 Fused preposition-article combinations
Hudson (2003, 21–23) draws attention to fused preposition-article combinations as evidence in favor of DP. A number of languages include fused preposition-article combinations, such as French, German, Spanish, etc.
| (38) | French |
| à+le = au, de+le = du |
| (39) | German |
| an+das = ans, in+das=ins, von+dem = vom, etc. |
| (40) | Spanish |
| a+el = al, de+el = del |
The fusion in these cases is easier to account for given DPs, since the pre-fusion structures have the preposition and article directly linked to each other, thus more easily allowing for fusion to occur. The next examples from German illustrate the reasoning behind this argument in favor of DP:
| (41) |
|
On the DP-analysis, there is a direct dependency linking the preposition in to the article das, whereas no such direct link exists on the NP-analysis. Hudson's argument, then, is that the presence of the direct link between the relevant words is responsible for licensing the fusion of the two words into one.
There is an alternative account for the observed process that does not rely on a direct link between the preposition and determiner. This alternative is that fusion is a mechanism of phonology and is hence sensitive to linear organization but insensitive to hierarchical organization. In other words, adjacency alone is enough to license the fusing of preposition and determiner; there need not be a direct dependency between the two words. This alternative account finds support from similar mechanisms that influence the phonological form of words, such as the distribution of clitics and of the appearance of the indefinite article an.
The distribution of clitic auxiliaries in English is not sensitive to hierarchical relationships between the clitics and their hosts. The host of the clitic auxiliary is always the immediately preceding word, regardless of the hierarchical relationship between that word and the clitic. The next examples illustrate this to be the case:
| (42) |
|
Concerning the particular tree conventions used to indicate clitic status, see examples (5) and (26) above and the surrounding discussions. In (42a), the auxiliary ’ve cliticizes to its dependent; in (42b) the auxiliary ’ve cliticizes to its head; and in (42c) the auxiliary ’ve cliticizes to its sibling. These examples therefore demonstrate that cliticization of auxiliaries in English is a function of adjacency alone; the hierarchical relationship between the clitic and its host does not influence the distribution of the clitic.
A similar mechanism from English that demonstrates the same insensitivity to hierarchical organization is the sandhi mechanism responsible for the appearance of an (as opposed to a) as the form of indefinite article. The next examples show that the appearance of an is evoked by the onset of the immediately following word, regardless of the hierarchical status of this word in relation to an. The NP-analysis is assumed for the illustration:
| (43) |
|
The onset of the head licenses an in (43a), the onset of the sibling in (43b), and the onset of the sibling's child in (43c). We therefore see again that the hierarchical relationship between the relevant words does not influence the distribution of the phonological mechanism. Organization in the horizontal dimension, i.e. with respect linearity, is alone what counts.
One might object at this point that there are other contraction phenomena that the appeal to linear order alone cannot sufficiently address, for instance so-called wanna contraction. Wanna-contraction is a well-known argument in favor of the existence of traces, e.g.
| (44) | a. | Who1 does Fred want to talk to t 1 ? |
| b. | Who1 does Fred wanna talk to t 1 ? |
| (45) | a. | Who1 does Fred want t 1 to see win? |
| b. | *Who1 does Fred wanna see win? |
Comparing examples (44a-b) with (45a-b), the grammaticality contrast across the (b)-examples is assumed due to presence/absence of the trace. Wanna-contraction is not possible if there is a trace present that intervenes between want and to. The point, then, is that there appears to in fact be evidence that that contraction phenomena are sensitive to the presence of null elements in the hierarchy of structure, which means linear order alone is not sufficient to address such phenomena.
There is a long-standing explanation for this curious aspect of the word form wanna that does not rely on the presence of a null element (cf. Sag & Fodor 1994). This explanation is that wanna is a lexicalized form that has particular valency properties, one of these properties being that it cannot take a small-clause-like complement. It is in a sense similar to the word forms won't and don't in that they are also lexicalized (cf. Don't you know? vs. *Do not you know? vs. Do you not know?). Support for this alternative explanation comes from the ability of contraction to occur in similar environments where a putative trace should block its occurrence, e.g.
| (46) | a. | Who1 do they think t 1 will arrive first? |
| b. | Who do they think'll arrive first? |
| (47) | a. | What1 did you say t 1 has happened? |
| b. | What did you say's happened? |
The trace in the (a)-sentences should, based on the reasoning applied to examples like (44–45) above, block contraction from occurring, yet contraction is such cases is fine. Given these further insights, any objection raised based on wanna-contraction is not a problem for the current account.
In sum, Hudson's argument in favor of DP collapses given the plausible assumption that the mechanism responsible for the fusion of preposition and article is similar to the phonological mechanisms responsible for clitic auxiliaries and a vs. an sandhi. Adjacency alone is what licenses or necessitates that fusion occur; there need not be a direct dependency connecting the words that fuse. The objection to this account based on wanna-contraction is weak due to the more plausible assumption that wanna is lexicalized form that, unlike want, does not subcategorize for a small-clause-like complement.
5.7 Children's acquisition
Ninio (2019) develops an argument in favor of DP based on Children's production and acquisition of two-word utterances. Very young children who are just beginning to produce multi-word utterances demonstrate certain patterns in the production of these utterances. In particular, a child that produces two-word determiner-noun utterances is more likely to also produce two-word complement-verb and verb-complement utterances than two-word adjective-noun utterances. These basic possibilities are illustrated next:
| (48) | a. | “the ball” | (Determiner+noun) |
| b. | “I play” | (Complement+verb) | |
| c. | “play ball” | (Verb+complement) | |
| d. | “big ball” | (Adjective+noun) |
If a child produces determiner-noun utterances like the one in (48a), then that child is more likely to also produce complement+verb and verb+complement utterances like the ones in (48b) and (48c) than it is to produce adjective+noun utterances like the one in (48d). Ninio interprets the tendency in this area as support for DP over NP.
Citing others (e.g. Mel’čuk, 1988; Sgall, Hajičová & Panevová, 1986; Starosta, 1988; Matthews 2007), Ninio draws a two-way distinction between complementation and attribution. Complementation occurs when a word takes a complement word (subject or object) as its dependent, whereas attribution occurs when an adjunct modifies another word. These two possibilities are illustrated next with the examples just above:
| (49) |
|
The question Ninio entertains concerns the dependency analysis of determiner-noun combinations: Are these combinations more like complementation or attribution? She sees two possibilities in this regard, and these two possibilities correlate with the distinction between the two analyses, DP vs. NP:
| (50) |
|
As these two diagrams suggest, Ninio sees the DP-analysis as an instance of complementation, whereby the noun is the complement of the determiner. In contrast, she construes the NP-analysis as an instance of attribution, whereby the determiner is an attribute and hence an adjunct modifier of the noun.
Having conducted a large-scale data evaluation project from a corpus of children's speech, Ninio finds that children producing determiner-noun combinations are indeed more likely to produce noun-verb and verb-noun combinations (and other types of complementation) than adjective-noun combinations. She concludes accordingly that determiner-noun combinations must be more like complementation than attribution.
The problem with the reasoning Ninio employs to build her argument in favor of DP resides in the limited analyses she entertains. In particular, she sees the NP-analysis as necessarily treating determiners as attributes that are essentially the same in syntactic status as attributive adjectives. I think some in the syntax world would object in this area. 8 Attributive adjectives are, namely, viewed as adjuncts by many, whereas determiners are deemed specifiers, specifiers being syntactically distinct in status from adjuncts. Given Ninio's limited, two-way distinction, the relevant question in this regard is therefore whether determiners are more complement-like or adjunct-like. I think many syntacticians view them as more complement-like because their appearance is often obligatory, whereas attributive adjectives usually appear optionally.
To state the problem with Ninio's analysis in other words, she mistakenly sees the NP-analysis as necessarily treating determiners as adjuncts. 9 An alternative interpretation of the NP-analysis sees determiners as more complement-like than adjunct-like. Hence, the following analysis of the nominal group the ball should also be entertained:
| (51) |
|
On this analysis, the determiner the is construed as more complement-like than adjunct-like.
In light of the alternative analysis just sketched, Ninio's argument in favor of DP disintegrates. The traditional NP-analysis of determiners allows one to construe them as being more like complements than adjuncts. The correlation Ninio documents in her data is therefore in fact congruent with the traditional NP-analysis of nominal groups.
5.8 Obligatory determiners in idioms
Salzmann (2020) examines the NP vs. DP debate in detail. He scrutinizes many arguments for and against DP or NP. His message in these areas is that some of these arguments were internal to the GB theory of the late 1980s and are hence not applicable to modern versions of PSG, or they are simply unconvincing in one respect or another. At the same time, he also considers insights from selection, subcategorization, form determination, and agreement, which he deems supportive of the DP-analysis. These areas have been touched on to varying extents above. One of Salzmann's arguments is particularly vivid and hence attention-worthy. This argument concerns the distribution of determiners in idioms and conventionalized expressions more generally.
Salzmann (2020: 30–32) points out that certain idiosyncratic word combinations involving verbs and nouns require the presence of a particular determiner. He cites weak definites in this respect, e.g.
| (52) | a. | read the newspaper |
| b. | read a/this/your/etc. newspaper |
| (53) | a. | take to the hospital |
| b. | take to a/that/some hospital |
| (54) | a. | play the piano |
| b. | play this/a/your/etc. piano |
The nominal groups in the (a)-examples do not function referentially, but rather they form a semantic unit with the verb (and preposition) to form complex predicates. The presence of the definite article is obligatory in this regard, since if some other determiner appears instead, the nominal group becomes referential and can no longer be construed as forming a complex predicate with the verb (and preposition), as suggested with the (b)-examples.
“Thus, while determiners may often be flexible in idioms, there are conventionalized expressions both in English and other languages where there is no flexibility at all and selection of D[eterminer]-elements is thus indispensable. This clearly favors the DP-hypothesis since one can directly refer to properties of D (or, in the case of bare nouns, select just an NP). Under the NP-hypothesis, one would have to have selection of Ns that in turn select a particular kind of D, which is doable, though not particularly elegant.” (Salzmann 2020: 32)
While Salzmann is correct that the appearance of a particular determiner is at times a necessary aspect of some idiosyncratic expressions, this fact should not be construed as evidence in favor of the DP-analysis. It is, rather, compatible with the NP-analysis in the broadest of ways. Consider in this regard that Salzmann acknowledges – as is evident in the first sentence of the cited passage – that many idioms exclude the determiner at the same time that they include the noun, a situation that supports the NP-analysis over the DP-analysis.
The point at issue is illustrated particularly well with dependency trees (cf. Osborne et al. 2012: 370–375).
| (55) |
|
In each of these trees, the words of the idiom are given, whereby the X (and possessive ’s) marks a constituent that is excluded from the idiom. The idiom each time is a single, multi-word predicate and the X is the second argument of that predicate. Most noteworthy is the fact that the words constituting the idiom are hierarchically continuous, that is, they form a subtree. Osborne et al. (2012) introduce the term catena (plural catenae) in this regard; the words that constitute idioms are stored in the lexicon as catenae (as opposed to as constituents). Examples (55c-e) are particularly relevant in the current context, because we see there that the possessive determiner does not interrupt in the vertical dimension the words constituting the idiom.
This situation is otherwise on the DP-analysis. The possessive determiners in these cases would interrupt the words of the idiom in the vertical dimension, preventing them from forming a subtree. This point is illustrated with example (55e), but according to the DP-analysis:
| (55) |
|
The possessive X's separates on from toes in the hierarchical dimension such that step, on, and toes no longer form a single subtree. What these examples suggest is that the NP-analysis is in a stronger position to account for the non-compositional meaning than the DP-analysis.
Turning now to Salzmann's examples involving weak definites, the words constituting the conventionalized meanings qualify as subtrees on both analyses, NP and DP, e.g.
| (56) |
|
The weak definite is present if the definite determiner the occurs. In that case, both analyses can view the words constituting the conventionalized expression as subtrees. That is, on both analyses the word combination take to the hospital is a (sub)tree. If some other determiner appears, such as the indefinite article a, the idiosyncratic meaning is lost such that the matrix predicate encompasses just take or perhaps take to, both of which are also subtrees on both analyses.
To summarize, Salzmann's insight that some conventionalized expressions include the determiner is correct. This fact does not, however, prevent the NP-analysis from accommodating such conventionalized expressions. On the NP-analysis, the words constituting the idiosyncratic meaning form subtrees regardless of whether the determiner is or is not part of the idiosyncratic combination. In contrast, the DP-analysis is such that it views only those idiosyncratic combinations as subtrees that include the determiner; in the event that the determiner is excluded as in (55c–e) above, the DP-analysis comes up short because the words constituting the idiom do not form a subtree.
6 Relevance for phrase structure grammars
Many of the points in favor of NP presented above in Section 4 have to do with the status of possessives as determiners. PSG analyses can side-step the evidence in favor of NP from some of the observations above by positioning possessives in the/a specifier position within DP, whereby the head D position is assumed occupied by a null determiner. Doing this increases the reliance of the DP-analysis on the role of null determiners.
There are, given DPs, two main possibilities for the analysis of possessives involving the Saxon ’s. These two possibilities are illustrated next:
| (57) |
|
The analysis in (57a) views the Saxon’s as the D head of the phrase, but the host of ’s, i.e. Jill, appears in the specifier position. The analysis in (57b), in contrast, has both the Saxon ’s and its host Jill in the specifier position, while the D head position is assumed occupied by a null determiner (cf. Adger 2003, 257). When a pronominal possessive determiner appears instead of the Saxon ’s, the dominant analysis is along the following lines:
| (57) |
|
The pronominal possessive determiner appears in the specifier position, and the D head is again occupied by a null determiner. While something like the analysis of the pronominal possessive determiner shown here in (57c) is often assumed, there is more variation concerning the analysis of the Saxon ’s.
The three structural analyses just shown side-step many of the problems discussed above facing the DP analysis in the context of DG. Selectional restrictions (Section 4.1) are no longer a problem because possessive ’s or the null determiner can be deemed transparent to the assignment of selectional restrictions coming from above; semantic and syntactic parallelism (Section 4.2) is not a problem because parallelism obtains insofar as the possessor is consistently in a specifier position of the head; coreference and binding (Section 4.4) is less of a problem because the index assigned to the possessor is no longer the index associated with the root node of the entire nominal group; and pre- and post-noun positioning of possessive phrases (Section 4.8) is no longer a problem because in both cases, the possessive phrase is technically a dependent of the head null determiner. Three of the other four arguments do, however, remain problematic for the DP-analysis: determiner-less nouns (Section 4.5), possessive determiners vs. pronouns (Section 4.6), and the inability to front pre-noun modifiers (Section 4.7). And concerning the one other argument, proform substitution (Section 4.3), it is a problem for the analysis in (57a), but not for the one in (57b), because on the analysis in (57a), the proform substitute would be replacing a non-constituent string (e.g. Jill's idea → her idea).
Beyond the specific difficulties facing the DP-analysis in PSGs, there is a significant conceptual problem. The DP-analysis typically places the most frequently occurring determiners (e.g. the and a) in the D head position, yet possessives are placed in the specifier position of D as illustrated in (57b) and (57c). Inconsistency of analysis concerning determiners therefore occurs. The fact that typically only a single determiner can introduce a nominal group in English suggests strongly that the one determiner should appear in the same syntactic position as the other – see examples (6a-d). The reason two or more determiners cannot introduce a given nominal group in English is because there is only a single position available for a determiner. Thus, the analyses in (57b) and (57c) incorrectly predict that phrases such as *Jill's the idea and *her the idea should be acceptable in English. To overcome this problem, the DP-analysis must posit the presence of the null determiner, which then blocks the second determiner from appearing. But now a situation occurs in which the existence of a null element is posited solely for the purpose “saving” the DP analysis from a glaring inconsistency. Consider further in this regard that if the null determiners were readily available in such cases, what would prevent it from licensing the appearance of a singular count noun, e.g. *Fred has Ø dog?
The main difficulty that possessives pose to the DP-analysis therefore concerns non-falsifiability. If the existence of these null elements cannot be verified conceptually and in an empirically testable fashion, then the DP-analysis is fundamentally flawed. The simple argument that the null determiner enables the DP-analysis to side-step the problems associated with possessives that would otherwise arise is not sufficient to motivate the null elements. More direct, testable evidence in favor of their existence is needed.
7 Monostratal DG
A primary reason why many PSGs assume DP instead of NP is certainly that it gives the grammar opportunities to address the nature and distribution of functional material inside the nominal group. Time spent with many modern PSG accounts of the structure of nominal groups reveals a tendency to posit highly layered structural analyses of the sort that allow one to acknowledge distinct hierarchical positions for various words within the nominal group – this tendency is most pronounced in so-called cartographic syntax (see for instance Cinque & Rizzi 2008 and Shlonsky 2010). The message produced in this section is that monostratal DG can accommodate some of the hierarchical organization of PSG accounts by locating this organization in the linear dimension of organization alone instead of in the hierarchical dimension as well.
The next example tree of a French nominal group is from Haegeman & Guéron (1999, 459). While the particular analysis that Haegeman & Guéron present is debatable in certain respects, it suffices here for current purposes because it illustrates well the type of layered analysis assumed to accommodate the order of determiner and adjectives:
| (58) |
|
There are two points here that are relevant in the current context. The first concerns the redundancy in the organization of the categories shown; this organization exists in both the hierarchical and linear dimensions simultaneously. The second is that the analysis assumes movement of the noun, as indicated by the trace and co-indexation.
The redundancy in organization is visible in the fact that starting with the left-most word, each word immediately precedes and is positioned higher than the next word. Thus, the article un both precedes and is higher than the adjective joli; the adjective joli both precedes and is higher than the adjective gros, etc. An approach to syntax that takes linear order as a primitive need not do this. To establish prominence in syntactic status, the one or the other, hierarchical or linear order, can be deemed enough. In particular, linear order alone suffices in many cases to establish concrete organization among the elements present in nominal groups.
Consider the rather flat DG analysis of the Haegeman & Guéron's example.
| (59) |
|
On this flat analysis, the determiner un is more prominent than its sibling dependents because it precedes them; the adjective joli is more prominent than its sibling dependents to its right because it precedes them; etc. The order of dependents is determined more generally by a linear principle of organization that adheres to the following sequencing: definiteness > evaluative > size > color.
Concerning the movement of the noun ballon indicated in (58), the DG approach has no means of accommodating such aspects of the analysis. The PSG assumption in (58) is that there is a universal organizational template responsible for the underlying organization of the nominal group. In the event that the surface order of elements in the nominal group does not match the template, a derivational mechanism, i.e. movement, is deemed necessary to establish the surface order (cf. Cinque 1996; Laenzlinger 2004; Cinque 2005). In the case of (58), upward movement of the noun ballon is responsible for positioning it in front of the adjective rouge. The DG analysis of the post-noun position of adjectives in French is necessarily different. The DG assumption is, rather, simply that languages differ in how they position dependents of the root noun, as either pre- or post-dependents (cf. Abels & Neeleman 2012). Templatic organization is still present, but it exists only in the linear dimension. Whether or not a given dependent precedes or follows its head noun is a minor distinction, an area in which the grammars of languages vary freely. 10
The relatively flat analysis of nominal groups proposed here should not be construed as extending to clausal organization. One might object, namely, that clausal organization requires that hierarchical and linear order play separate roles. An example structure one might cite in this regard is from German, where auxiliary verbs usually precede the main content verb in main clauses, but follow it in subordinate clauses. Thus, an appeal to precedence alone cannot account for the variation in verb order that occurs across matrix and embedded clauses in German. Indeed, the current account is in agreement with this point. Verbs in clauses are arranged with respect to each other in terms of both linearity and hierarchy. The next examples from German illustrate this situation:
| (60) |
|
Given the three-verb combination, the auxiliary verb haben ‘have’ of aspect necessarily follows the main content verb überrascht ‘surprised’. The modal auxiliary wird ‘will’, however, can precede or follow überrascht depending in part on whether a subordinator such as dass ‘that’ is present.
The variation in verb order illustrated with examples (60a-b) requires that the verbs also be arranged hierarchically with respect to each other (as shown). This difference in hierarchical organization across nominal groups, which tend to be relatively flat, and clauses, which tend to be more layered, is motivated by numerous syntactic differences. For instance, the word order associated with phrasal constituents in clauses is freer than the order of the elements of nominal groups, especially regarding the pre-modifiers of nouns, the order of which is rigid. In languages with case, verbs assign case in a way that nouns do not. Verb chains are such that the one type of verb subcategorizes for another type of verb; a similar subcategorization imperative does not exist in nominal groups. These and other differences motivate the differences in structure across clause and nominal group.
To summarize, the relatively flat DG analyses of nominal groups can accommodate the rigid order of pre-noun modifiers in nominal groups by appealing to linear organization alone. In contrast, many PSGs are redundant in this regard because they locate the relevant organization in both ordering dimensions simultaneously.
8 Conclusion
This paper has examined the NP vs. DP debate from the perspective of dependency grammar (DG). The strict one-to-one mapping (words to nodes) of dependency syntax forces one to interpret the debate in a different light from how it is understood in phrase structure grammar (PSG). In particular, the reluctance of dependency syntax to posit null elements blocks it from positing null determiners. Without the null determiners, the evidence in favor of the NP-analysis becomes compelling. In particular, the behavior of possessives in English and related languages is easily accommodated on the NP-analysis, but is problematic for the DP-analysis.
References
Abels, Klaus and Ad Neeleman . 2012. Linear asymmetries and the LCA. Syntax 15(1). 25–74.
Abney, Steven . 1987. The English noun phrase in its sentential aspect. MIT Ph.D. dissertation.
Adger, David . 2003. Core syntax: A minimalist approach. Oxford, UK: Oxford University Press.
Anderson, John . 2006. Modern case grammars: A retrospect. Oxford, UK: Oxford University Press.
Berman, Arlene . 1974. Adjectives and adjective complement constructions in English. PhD dissertation, Harvard University.
Bröker, Norbert . 1999. Eine Dependenzgrammatik zur Kopplung heterogener Wissensquellen. Tübingen: Niemeyer.
Bröker, Norbert . 2003. Formal foundations of dependency grammar. In Vilmos Ágel , Ludwig M. Eichinger , Hans-Werner Eroms , Peter Hellwig , Hans Jürgen Heringer and Hennig Lobin (eds.) Dependency and valency: An international handbook of contemporary research. Berlin: Walter de Gruyter. 294–310.
Carnie, Andrew . 2010. Constituent structure, second edition. Oxford, UK: Oxford University Press.
Carnie, Andrew . 2013. Syntax: A generative introduction, third edition. Malden, MA: Wiley-Blackwell.
Chomsky, Noam . 1957. Syntactic structures. The Hague: Mouton.
Chomsky, Noam . 1981. Lectures on government and binding: The pisa lectures. Dordrecht: Foris publications.
Chomsky, Noam . 1986. Barriers. Cambridge, MA: MIT Press.
Cinque, Guglielmo . 1996. On the evidence for partial N-movement in the Romance DP. In Italian syntax and universal grammar (cambridge studies in linguistics). Cambridge: Cambridge University Press. 287–309.
Cinque, Guglielmo . 2005. Deriving Greenberg's Universal 20 and its exceptions. Linguistic Inquiry 36(5). 315–332.
Cinque, Guglielmo and Luigi Rizzi . 2008. The cartography of syntactics structures. In Vincenzo Moscati (ed.) University of Siena CISCL Working Papers 2 (STiL – Studies in Linguistics). MIT Press. 43–59.
Culicover, Peter and Ray Jackendoff . 2005. Simpler syntax. Oxford, UK: Oxford University Press.
Denham, Kristin and Anne Lobeck . 2013. Linguistics for everyone: An introduction. Boston: Wadsworth Cengage Learning.
Engel, Ulrich . 1994. Syntax der deutschen Gegenwartssprache, third edition. Berlin: Erich Schmidt Verlag.
Gerdes, Kim . 2002. Topologie et grammaires formelles de l'allemand. Ph.D. thesis, Université Paris. 7.
Groß, Thomas . 1999. Theoretical foundations of dependency syntax. München: Iudicium.
Groß, Thomas . 2014. Clitics in dependency morphology. In Kim Gerdes , Eva Hajičová and Leo Wanner (eds.) Dependency linguistics: recent advances in linguistic theory using dependency structures. Amsterdam: John Benjamins. 229–251.
Groß, Thomas and Timothy Osborne . 2013. Katena und Konstruktion: Ein Vorschlag zu einer dependenziellen Konstruktionsgrammatik. Zeitschrift für Sprachwissenschaft 32(1). 41–73.
Haegeman, Liliane and Jacqueline Guéron . 1999. English grammar: A generative perspective. Oxford, UK: Blackwell Publishers.
Hellan, Lars . 1986. The headedness of NPs in Norwegian. In Pieter Muysken and Henk Van Riemsdijk (eds.) Features and projections. Dordrecht: Foris. 287–301.
Heringer, Hans . 1996. Deutsche syntax dependentiell. Tübingen: Staufenberg.
Hewson, John . 1991. Determiners as heads. Cognitive Linguistics 2(4). 317–338.
Hogg, Richard. M . 1977. English quantifier systems. Amsterdam: North Holland.
Huddleston, Rodney and Geoffrey Pullum . 2002. The Cambridge grammar of the English language. Cambridge, UK: Cambridge University Press.
Hudson, Richard . 1984. Word grammar. Oxford, UK: Basil Blackwell.
Hudson, Richard . 1987. Zwicky on heads. Journal of Linguistics 23(1). 109–132.
Hudson, Richard . 1990. English word grammar. Oxford, UK: Basil Blackwell.
Hudson, Richard . 2003. Are determiners heads? Functions of Language 11(1). 7–42.
Hudson, Richard . 2003. Word grammar. In Vilmos Ágel , Ludwig M. Eichinger , Hans-Werner Eroms , Peter Hellwig , Hans Jürgen Heringer and Hennig Lobin (eds.) Dependency and valency: An international handbook of contemporary research. Berlin: Walter de Gruyter. 508–526.
Hudson, Richard . 2007. Language networks: The new word grammar. Oxford: Oxford University Press.
Jackendoff, Ray . 1972. Semantic interpretation in generative grammar. Cambridge, MA: MIT Press.
Jung, Wha-Young . 1995. Syntaktische Relationen im Rahmen der Dependenzgrammatik. Hamburg: Helmut Buske Verlag.
Kahane, Sylvain . 1996. If HPSG were a dependency grammar…. In Philippe Blache (ed.) Actes de TALN (traitement automatique de Langues naturelles), Marseille, 22–24, May 1996. 45–49. Available at: https://kahanedotfr.files.wordpress.com/2017/06/dg-hpsg1996.pdf.
Kayne, Richard . 1994. The antisymmetry of syntax. Cambridge, MA: The MIT Press.
Keizer, Evelien . 2007. The English noun phrase: the nature of linguistic categorization. Cambridge, UK: Cambridge University Press.
Kim, Jong-Bok and Peter Sells . 2011. The big mess construction: Interactions between the lexicon and constructions. English Language and Linguistics 15(2). 335–362.
Laenzlinger, Christopher . 2004. French adjective ordering: Perspectives on DP-internal movement types. Lingua 115(5). 645–689.
Lobin, Henning . 1993. Koordinationssyntax als prozedurales Phänomen. Series: Studien zur deutschen Sprache 46. Tübingen: Gunter Narr Verlag.
Matthews, Peter . 2007. Syntactic relations. Cambridge: Cambridge University Press.
McCawley, James . 1998. The syntactic phenomena of English, second edition. Chicago: The University of Chicago Press.
Mel'čuk, Igor . 1979. Studies in dependency syntax. Ann Arbor: Koroma Publishers.
Mel'čuk, Igor . 1988. Dependency syntax: theory and practice .Albany, NY: State University of New York Press.
Mel'čuk, Igor . 2009. Dependency in natural language. In Alain Polguère and Igor Mel'čuk (eds.) Dependency in linguistic description. Amsterday: John Benjamins. 1–110.
Mel'čuk, Igor and Nikolai Pertsov . 1987. Surface syntax of English: A formal model with the Meaning-Text Framework. Amsterdam: Benjamins.
Miller, Jim . 2011. A critical introduction to syntax. London: Continuum.
Müller, Stefan . 2016. Grammatical theory: From transformational grammar to constraint-based approaches (Textbooks in Language Sciences 1). Berlin: Language Science Press. Available at http//langsci-press.org/catalog/book/25.
Ninio, Anat . 2019. Complement or adjunct? The syntactic principle English-speaking children learn when producing determiner-noun combinations in their early speech. In The role of grammatical words in syntactic development: Special issue. First Language 39 (Anat Ninio, guest editor), Vol. 1. 33–44.
Osborne, Timothy and Matthew Reeeve . 2018. Scope Domains: toward a dependency grammar account of the syntactic distribution of negative polarity items. Acta Linguistica Academia 65(1). 1–49.
Osborne, Timothy , Michael Putnam and Thomas Groß . 2012. Catenae: Introducing a novel unit of syntactic analysis. Syntax 15(4). 354–396.
Payne, John . 1993. The headedness of noun phrases: Slaying the nominal hydra. In Grevelle Corbett , Norman Fraser and Scott McGlashan (eds.) Heads in grammatical theory. Cambridge, UK: Cambridge University Press. 114–139.
Quirk, Rodney , Sidney Greenbaum , Geoffrey Leech and Jan Svartvik . 2010. A comprehensive grammar of the English language. Dorling Kindersley: Pearson.
Robinson, Jane . 1970. Dependency structures and transformational rules. Language 46. 159–285.
Ross, John Robert . 1967. Constraints on variables in syntax. Ph.D. dissertation, MIT.
Ross, John Robert . 1986 (1967). Infinite syntax! Norwood, NJ: Ablex Publishing Corporation.
Sag, Ivan and Janet Dean Fodor . 1994. Extraction without traces. In Raul Aranovich , William Byrne , Susanne Preuss and Marth Senturia (eds.) Proceedings of the thirteenth west coast conference on formal linguistics. Stanford, CA: CSLI Publications/SLA. 365–384. Avaiable at: http://lingo.stanford.edu/sag/papers/sag-fodor-wccfl.pdf.
Sag, Ivan , Thomas Wasow and Emily Bender . 2003. Syntactic theory: A formal introduction, second edition. Stanford, CA: CSLI Publications.
Salzmann, Martin . 2020. The NP vs. DP debate. Why previous arguments are inconclusive and what a good argument could look like. Evidence from agreement with hybrid nouns. Glossa: A Journal of General Linguistics 5(1). 1–46, 83.
Schubert, Klaus . 1987. Metataxis: contrastive dependency syntax for machine translation. Dordrecht: Foris Publications.
Sgall, Petr , Eva Hajičová and Jarmila Panevová . 1986. The meaning of the sentence in its semantic and pragmatic aspects. Dordrecht: D. Reidel Publishing Company.
Shlonsky, Ur . 2010. The cartographic enterprise in syntax. Language and Linguistics Compass, 4(6). 417–429.
Sobin, Nicholas . 2011. Syntactic analysis: the basics. Malden, MA: Wiley-Blackwell.
Sportiche, Dominique , Hilda Koopman and Edward Stabler . 2014. An introduction to syntactic analysis. Malden, MA: Wiley-Blackwell.
Starosta, Stanley . 1988. The case for lexicase: An outline of lexicase grammatical theory. London: Pinter Publishers.
Tesnière, Lucien . 1966. Éléments de syntaxe structurale, second edition. Paris: Klincksieck.
Van Eynde, Frank . 2007. The big mess construction. In Stefan Müller (ed.) Proceedings from the 14th international Conference on head-driven phrase structure grammar. Stanford University: CSLI Publications. 416–433.
Van Langendonck, Willy . 1994. Determiners as heads? Cognitive Linguistics 5. 243–259.
Vennemann, Theo . 1977. Konstituenz und Dependenz in einigen neueren Grammatiktheorien. Sprachwissenschaft 2. 259–301.
Zwicky, Arnold . 1985. Heads. Journal of Linguistics 25. 639–669.
The term root/head is clarified below in Section 3. One can understand it for the time being simply to mean ‘head’.
The following list gives the main acronyms and abbreviations used in this article: DG = dependency grammar, DP = determiner phrase, NP = noun phrase, PSG = phrase structure grammar
An interesting aspect of Keizer's lengthy study of heads in nominal groups in English is that she never entertains the notion that the determiner might be head over the noun. She instead takes it for granted that one of the nouns present is the head. She examines difficult cases in this regard, e.g. close appositions (the actor Orson Wells), binominals (a fool of a doctor), pseudopartitives (a number of objects), etc.
The term governor is also prominent in many DGs, whereby the governor of a given word is the one other word that licenses that word's appearance. This notion is, however, not important in the current context because the current article has nothing to say about discontinuities. Most of the time, the head and the governor of a given word are the same one word. Given a discontinuous structure, though, the head and the governor of the displaced expression can be distinct words. Since no discontinuities occur in the examples in this article, the terms head and governor can be viewed as synonymous and there is therefore no reason to use the latter term.
Concerning the forms that one and this one, an anonymous reviewer comments that one in such cases is not a pronoun, but rather a generic noun. The analysis of one is actually complex and disputed, confusion arising due to the fact that there are two distinct types of one, the cardinal number and the so-called pronominal count noun. In any event, one in the examples provided in the table is functioning anaphorically, which motivates the term pronoun to denote its meaning contribution.
An anonymous reviewer comments at this point that since the current approach accepts null elements in cases of ellipsis, more caution is needed concerning the non-falsifiable nature of null determiners. This point is well made insofar as the key issue is not whether or not null elements are necessary in the syntax of a given language such as English – they are – but rather whether null elements are necessary to account for the given phenomenon under scrutiny. The stance developed in this article is precisely that null determiners are not needed to account for the syntax of nominal groups.
The anonymous reviewer also points to the forms after (transitive) and afterwards (intransitive). These forms are an anomaly insofar as other subordinators do not behave in this manner, e.g. since and before. More importantly, the -wards in afterwards can be viewed as the complement of after if one acknowledges intra-word dependencies as in (26b). What this means is that the head is consistently after, the form of which does not change based upon whether -wards or an NP complement is or is not present.
Others might well not object, since there has long been debate concerning the distinction between specifiers and adjuncts in nominal groups, at least since Kayne (1994).
Note that the matter at hand is addressed by many DGs in terms of the function labels attached to dependencies. The argument vs. adjunct distinction is often acknowledged and encoded in terms of these labels. Thus, Ninio's argument in favor of the DP-analysis would have no validity for these DGs because determiners would not be interpreted as adjuncts in the first place; the function labels assigned to the determiner dependencies would prevent determiners from being interpreted as adjuncts.
Interestingly, German also allows an alternative ordering in such cases; the auxiliary verb could also precede the other verbs, i.e. dass das sie wird überrascht haben.
