No CrossRef data available.
Article contents
Grammaticalization as Economy. By Elly van Gelderen. (Linguistik Aktuell/Linguistics Today, 71). Philadelphia: John Benjamins, 2004. Pp. xiv, 320. Hardcover. $138.00.
Published online by Cambridge University Press: 12 April 2006
Abstract
- Type
- REVIEWS
- Information
- Copyright
- © 2006 Society for Germanic Linguistics
1. Introduction.
Grammaticalization as Economy is an important contribution to the grammaticalization literature. Grammaticalization is standardly defined as the creation of new functional material through the reanalysis of lexical material or existing functional material; for example, the development of English auxiliary verbs from lexical verbs (see, for instance, Lehmann 1985). In this review, I first give a brief synopsis of Grammaticalization as Economy and its major themes, and then I explore some of the issues raised by van Gelderen's account of grammaticalization. In particular, I discuss (i) the status of economy principles in models of language change, (ii) the interaction of internal and external forces in language change, and (iii) formal and functional explanations of grammaticalization.
2. Synopsis.
In van Gelderen's model, the two internal forces of linguistic change that drive grammaticalization are the HEAD PREFERENCE PRINCIPLE (HPP: “Be a head, rather than a phrase”), and the LATE MERGE PRINCIPLE (LMP: “Merge as late as possible”). Within the grammaticalization literature, antecedents of the LMP can be found in Tabor and Traugott's (1998) discussion of syntactic scope expansion in grammaticalization (not cited by van Gelderen), as well as Roberts and Roussou's (2003) claim that grammaticalization always involves upward reanalysis. Upward reanalysis is syntactic reanalysis along the clausal hierarchy CPTP-VP—a small, unproductive subclass of a lexical category is reanalyzed as a subclass of a functional category. To illustrate, for English modals a subclass of lexical verbs (in V), which were once moved to T, are directly merged in T; that is, reanalyzed as elements of T. The HPP and LMP are at “work in grammars of speakers, as well as in leading children to build their grammars in a particular way” (p. 12). I return to the dual status of these principles in Section 3. The HPP is reflected in Jespersen's (1917) cycle (see recent work by Willis, forthcoming and Condoravdi and Kiparsky 2005). For example, in French a negative marker in the head of NegP (ne) weakened into a verbal proclitic (n-), then reinforced by a negative element in Spec,NegP (pas), and then ultimately replaced by the reinforcement when the reinforcement becomes the head of NegP.1
Note that van Gelderen seems to assume that grammaticalization generally involves the reduction of a form, sometimes to zero, followed by emergence of a new form. However, as Hopper and Traugott (2003:124) point out, “this kind of model is extremely problematic, because it suggests that a stage of language can exist when it is difficult or even impossible to express some concept.” Rather, new forms always compete with older ones during periods of language change.
Building on work of Rizzi (1997) and Cinque (1999), van Gelderen argues that languages vary in whether they have an “unsplit” CP, as in 1, or an expanded or “split” CP, as in 2. A split CP contains multiple layers of functional projections, such as ForceP, TopicP, FinP.


In Modern English, the head of non-finite unsplit CPs (see 1) can be filled by for, as in I would like for you to do your homework. In Italian, which has a split CP, the finite complementizer che ‘that’ appears in the head of ForceP in 2, whereas the nonfinite complementizer di ‘for’ appears in the head of FinP.
Van Gelderen proposes that the English complementizer that is derived from a demonstrative pronoun in the specifier of CP, and that this change, schematically illustrated in 3, is ultimately a reflection of the HPP. The demonstrative pronoun se/pam/pat starts out in the specifier of CP and then becomes the head of CP after 1050. A wh-pronoun is introduced in the Spec,CP around the 12th century.2
As is well known, wh+that clauses are allowed in certain varieties of English, for example, Belfast English (Henry 1995). Zwicky (2002) discusses wh+that clauses (I want to tell you what experiences that I've had here in my work) in American English.

Van Gelderen argues, based on the development of the complementizer for and variation in the syntactic distribution of infinitival to, that the LMP played a role in the development of the CP domain in English.3
More precisely: from P to Asp (by Shakesperean English), and then to M, the head of MoodP.
There is considerable debate in the literature about the auxiliary or subordinator status of infinitival to. Unlike auxiliary verbs, infinitival to must immediately precede the ellipsis site (They suggested I call the police but I decided not to __ / *to not __ ; see Huddleston and Pullum 2002:1526). However, in order to establish that to is in C, it is necessary to provide an explanation for the fact that to, but not complementizers (for, that), can license VP ellipsis (They ordered us to leave, and we want to __ versus *Pat preferred for Sandy to get the job, and we preferred for __ , also; see Ginzburg and Sag 2000:51). Further, unlike complementizers such as that, to does not have to precede the constituent it marks, as in She taught her children always to tell the truth (Huddleston and Pullum 2002:1186), where always is interpreted in the subordinate clause. See Pullum 1982, Ginzburg and Sag 2000:51, and especially Huddleston and Pullum 2002:1183–87. I return to van Gelderen's analysis of to below.
Turning to IP and VP, van Gelderen assumes the structure in 4 for Modern English.

For the VP, van Gelderen assumes the extended VP-shell structure given in 5 (Larson 1988). EP stands for Extent Phrase, expressing the extent to which the action denoted by the head verb has been done and how much the object has been affected.5
Unfortunately, van Gelderen does not address arguments against the VP-shell hypothesis (Jackendoff 1990, Kiparsky 1997, Kuno and Takami 1993, Wasow 2002). See Culicover and Jackendoff 2005:50–56 for recent critical discussion.

The development of modal verbs such as can (V > v > Asp (> M)) and perception verbs (V > v > Asp), such as the evidential use of see with a bare infinitival complement, as in I saw him cross the street, is taken to be a reflection of the LMP. Similarly, as noted above, to underwent the trajectory P > Asp > M (> C), also taken by van Gelderen to be a reflection of the LMP. Example 6 schematically illustrates the development of see from a verb base-generated in the head of VP to an evidential auxiliary verb base-generated in the head of AspP. Evidence for this trajectory comes from gradual changes in aspect marking in English, such as the decline of bare infinitival complements after present tense see and the loss of the perfective aspect marking ge- prefix.

In the concluding chapter, van Gelderen discusses the difference between parametric change and changes involving principles. She observes that changes involving parameters take place faster than those involving principles. For example, the parametric switch in English from having unmarked aspect to unmarked tense (The Tense Aspect Parameter) took place within the 19th century, as evidenced by the rapid rise of the obligatory progressive to mark the simple present with durative verbs (*I read now versus I am reading now). In contrast, the change from demonstrative pronoun to complementizer that discussed above (involving the HPP) was gradual, taking place over several hundred years.
3. Discussion.
3.1. Economy Principles.
In van Gelderen's model, the two internal motivating forces of grammaticalization are the HPP and the LMP. These principles have dual status. On the one hand, these “are principles of Economy that work in grammars of speakers.” On the other hand, these principles “lead children to build their grammars in a particular way” (p. 12).6
Later, van Gelderen seems to claim that the HPP and LMP only play a role in generation: “These are synchronic processes, at work in the production of utterances by native speakers” (p. 261).
Further, although economy played a major role in early Minimalist theorizing, economy no longer plays a key role in current work. For example, the assumption that it is less economical to merge early and then move than to wait as long as possible before merging (that is, Merge over Move, van Gelderen's LMP) no longer holds true. Instead, now Merge comes in two flavors: Internal Merge and External Merge, replacing Move and Merge, respectively (Chomsky 2004)
On the second view, economy principles are biases on the language acquisition process. This view of economy has played a key role in generative work on morphosyntactic change (Lightfoot 1979, Clark and Roberts 1993, Roberts 1993, Kiparsky 1996, Roberts and Roussou 2003, Haeberli 2004). For example, Roberts and Roussou (2003) define the markedness of parameter values in terms of a simplicity metric, where, for example Merge is less marked than Move. In the absence of cues for marked parameter values such as inflectional morphology, the less marked option is taken in acquisition.7
The notion that economy principles alone drive language change has been attacked in the literature (Lightfoot 1999:215–220). I return to this issue in Section 3.2.
Do the two views on economy described above have different empirical consequences? The idea that (something akin to) economy principles are part of the processing mechanisms that employ linguistic competence has been explored in recent work, particularly in the area of word order variation and change. For example, Hawkins (1994, 2004) argues that parsing principles influence generation, and that more easily parsable variants will be used more frequently than less easily parsable variants. In contrast, Kirby (1999) equates parsability with learnability, as in the second view of economy described above. Briscoe (1998) discusses computer simulation work that suggests that either Hawkins' model or Kirby's model, or a combination thereof, accounts for language change in favor of more parsable variants.
Further, economy principles may also influence comprehension. Recent psycholinguistic work has explored the different empirical consequences of production/speaker-based versus comprehension/hearer-based accounts of linguistic variation (see especially Wasow 2002; also Jaeger 2005 and Jaeger and Wasow 2005 on that-omission in complement and relative clauses). It would be interesting to explore the diachronic consequences of competing production and comprehension interpretations of the HPP and the LMP.
3.2. Internal and External Principles.
Lightfoot (1979) and Roberts (1993) rely solely upon internal economy principles to capture directionality in language change. For example, Roberts (1993) (also Clark and Roberts 1993:335) proposes a principle of language acquisition, the Least Effort Strategy. The Least Effort Strategy claims that language acquirers will prefer grammars that assign the most compact or simplest representations to the input string. As Lightfoot (1999:219) points out, if economy principles determine language change, why should there ever be systems that violate these principles? Van Gelderen's answer is that internal factors interact with external ones.
As van Gelderen notes (pp. 12, 262), the notion that language change involves an interaction between internal and external forces has been around for a long time (Jespersen 1922: Chapter 14,§6). Keller (1994:100), Kiparsky (1996), and Wu (2004) also propose accounts of language change that involve the interaction between internal and external forces. However, it is possible to critique competing forces accounts on methodological grounds, since there is typically an ad hoc search for external forces to match the phenomena that need explanation. One way to avoid this problem is to use computational models of language change, use, and acquisition (Briscoe 2002, Cangelosi and Parisi 2002). Computer simulations enable us to model different explanations for language change and explore the consequences of varying side conditions. For example, Briscoe (1998) utilizes computer simulations to show that an adequate model of language change must include at least internal learnability factors and counteracting external pressures, such as expressivity.
In van Gelderen's model, internal forces such as the HPP and the LMP interact with external forces such as prescriptive prohibitions, borrowing, and expressivity. For example, as noted above, van Gelderen argues that to undergoes a category change from P to I, and from I to C. Attested examples of to in I and C are given in 7 and 8, respectively (p. 247).


Van Gelderen (pp. 242–244) provides several arguments that to appears in C in examples such as 7. However, her arguments are somewhat problematic.
First, she claims that to not does not occur with the complementizer for (versus, for example, she prefers for me not to stay on the phone for very long). Note, however, that sequences of the form X prefers for Y to not, as in The WWE prefers for wrestlers to not have a personal webpage may be easily found.
Second, van Gelderen provides evidence from the British National Corpus that in raising constructions seem does not occur with to not (where to in to not is claimed to be in C). This is taken by van Gelderen to favor her analysis because it is widely assumed that in raising constructions seem takes an IP complement, and never a CP complement. However, it is also easy to find naturally occurring instances of the sequence seems to not, as in Javascript seems to not be working.
Third, according to van Gelderen, to must immediately precede the ellipsis site in VP ellipsis, as in They suggested I call the police but I decide not to __ / *to not __ (see Huddleston and Pullum 2002:1526). As indicated above (see note 4), this provides support for van Gelderen's position that to is not an auxiliary verb, but does not necessarily provide support for her position that to is in the head of CP.
Fourth, van Gelderen argues that the existence of to not to sequences as in This is to try to not to overturn the … supports the claim that to can appear in either I or C. Sequences of this sort with to not to are indeed frequent. However, just like the previous arguments, it does not provide powerful evidence in support of analyzing to as a clausal subordinator in C.
In sum, (i) as noted in footnote 4, there are clear differences between clausal subordinators such as that and for) on the one hand, and to on the other hand, and (ii) there is compelling evidence that to is not an auxiliary verb, but a VP subordinator (Huddleston and Pullum 2002:1185–1186).
Putting the question of infinitival to's categorial status aside, the problem that 7 and 8 pose for the LMP is why to is not always in C, as in 7. One possible explanation van Gelderen suggests (p. 248) is the prescriptive rule against splitting infinitives in English (Quirk and Greenbaum 1973:312). This is an interesting claim, but much more has to be said in order to establish its plausibility.
First, it is interesting to note that the rule against the split infinitive did not arise until the second half of the nineteenth century (Huddleston and Pullum 2002:581). Split infinitives had been in use for several hundred years (since the 14th century), becoming popular in English writing during the nineteenth century. A way to provide support for van Gelderen's hypothesis would be to track the frequency distribution of the split and non-split infinitive variants and see if a change in the distribution of these variants is correlated with the emergence of the prescriptive rule.
Second, the split infinitive is obligatory in certain cases, for instance with prepositional modifiers, as in We expect it to more than double versus *We expect it more than to double (Zwicky 2004). Third, promiscuous application of the rule against split infinitives can create ambiguity, as seen in 9 and 10 (Huddleston and Pullum 2002:581–582). In 9, which violates the rule against split infinitives, immediately unambiguously modifies approve. In contrast, 10 is ambiguous.


Fourth, it is not clear that “the rule against splitting infinitives is still very strong in standard English” (p. 248). As Huddleston and Pullum (2002:582) point out, “modern usage manuals […] present the rule only with the qualification that a ‘split infinitive’ is acceptable if it improves clarity or avoids awkwardness.”
Summing up, the idea that prescriptive pressures compete with economy principles is an intriguing one, but there is a tremendous amount of empirical and theoretical work that needs to be done in order to demonstrate this interaction in particular cases.
3.3. Formal versus Functional Explanations of Grammaticalization.
Grammaticalization as Economy is a part of a recent trend in the historical linguistics literature to give formal accounts of grammaticalization within the Principles and Parameters framework (van Kemenade 2000, Roberts and Roussou 1999, 2003, Batllori, Hernanz, Picallo, and Roca 2005), thus attempting to bridge the gap between formal and functional accounts of language change. Under these approaches, a complete theory of grammaticalization involves only two components: (i) an account of how trigger experiences become ambiguous, and (ii) a theory of language acquisition (and/or generation, see section 3.1) that employs economy principles (that is, biases for certain derivations over others). For van Gelderen, as well as the other proponents of formal accounts of grammaticalization cited above, it is sufficient to have a model of how the cognitive process of language acquisition maps the trigger experience to knowledge of language.
However, this approach falls short of a complete account of language grammaticalization.8
See also Traugott 2003:20–21 and Brighton, Kirby, and Smith 2005 for similar arguments regarding language change and language universals, respectively.
The formal accounts of grammaticalization cited above typically focus entirely on the cognitive processes involved in language acquisition and/or generation. In contrast, Traugott (2003:21) argues that language change is “the result of strategic interaction, specifically of choice-making on the part of speakers/writers in interactional negotiation with addressees/readers. This includes, but is not limited to, conveying of information.” An account of grammaticalization like Traugott's does not exclude internal factors such as the ones identified by van Gelderen and others. Rather, such a view implies that a satisfactory model of grammaticalization cannot rely entirely on an explanation of how trigger experiences change and a theory of language acquisition, but must also include a model of the cultural process that mediates between the context of language use and the cognitive structures that encode linguistic competence.
4. Conclusion.
Grammaticalization as Economy is an important and provocative work that will be of interest to historical linguists of all stripes. On the empirical end, van Gelderen provides a range of novel data from early and contemporary English. On the theoretical end, van Gelderen has contributed a new model of grammaticalization that raises interesting questions about the role of economy in models of linguistic competence and change, the interplay of internal and external forces in language change, and the relative strengths of formal and functional models of grammaticalization. The research program that van Gelderen has begun here could be developed in several ways. The empirical and theoretical consequences for grammaticalization of different interpretations of economy principles, such as constraints on generation, need to be investigated. Van Gelderen's proposal that economy principles interact with external forces (for example, prescriptive pressures) also needs to be examined in greater detail in the case of the development of infinitival to. Lastly, formal and functional explanations of grammaticalization are typically perceived as irreconcilable. However, it seems entirely feasible to develop a model of grammaticalization that combines a formalist account of linguistic competence (van Gelderen's model, the model presented in Roberts and Roussou 2003) with a functionalist understanding of the process that mediates between the context of language use and the cognitive structures that encode linguistic competence (Traugott and Dasher's 2002 Invited Inferencing Theory of Semantic Change).