3.1 Deixis Impossible

Surface anaphora cannot be deictic; they must have an antecedent in their linguistic context (Hankamer & Sag 1976). For example, if I see a man eating dinner, pointing and saying 12 will not convey that I have eaten dinner already.

1. (12)

The relevant type anaphora must also have an antecedent in the linguistic context (Borthen 2004:1). For example, if I see a bike, pointing and saying 13 will not convey that I bought a bike of the same kind yesterday.

1. (13)

Not allowing a deictic interpretation is a necessary, but not a sufficient condition for being a surface anaphor. There are deep anaphora that cannot be deictic, for example he and she referring to inanimates in German (Kiparsky 2002) and in dialects of Norwegian.

3.2 Missing Antecedent

Hankamer & Sag (1976) use the “missing antecedent” phenomenon as a criterion for surface anaphora. One of their examples appears in 14 (Hankamer & Sag 1976:403).

1. (14)

The first occurrence of camel cannot establish a discourse referent because it is under the scope of negation. The pronoun it must refer to the camel that Ivan rode, and this means that it must have access to the structure underlying the surface anaphor.

The same argument can be given for Norwegian, as shown in 15 (see Houser et al. 2007 for Danish as well).

1. (15)

The relevant type anaphora also show the “missing antecedent” phenomenon, as example 16 demonstrates.

1. (16)

The antecedent of the pronoun den ‘it’ is in a sense the type anaphor det. However, it is important to note that the type anaphor itself cannot be the antecedent; it must be the structure underlying the type anaphor. The reason is that the pronoun den ‘it’ is non-neuter, as required by the antecedent of the type anaphor, the masculine noun hund ‘dog’. It would have been impossible to use the neuter form of the pronoun, even if this is the form of the type anaphor (Borthen 2005:121).

3.3 Structural Parallelism

Hankamer & Sag (1976) and Sag & Hankamer (1984) show that there are requirements for a structural parallelism between the sentence with the antecedent and the sentence with the surface anaphor. One of their examples is given in 17 (Sag & Hankamer 1984:327).

1. (17)

A corresponding example with the surface anaphor det is given in 18.

1. (18)

Parallelism requirements also hold for sentences with the relevant type anaphora. It is important to note that the interpretation of the relevant type anaphora is determined by their antecedents. The antecedent cannot be definite, but it can be specific or nonspecific, as in 19. It should be noted that Norwegian allows so-called bare singulars, singular noun phrases without determiners, which are always nonspecific (Borthen 2003:26–28), such as hund ‘dog’ in 19.

1. (19)

The potential interpretation of the relevant type anaphora is always the same as that of their antecedents. For example, hund ‘dog’ is always nonspecific, while en hund ‘a dog’ can be specific or nonspecific, and the same is true of the surface anaphor they antecede in 19.

This analysis is different from the one proposed by Borthen, who insists that the type anaphor is always type referring and nonreferential (see Borthen 2003:286–294). However, there seems to be no reason to differentiate between the type anaphor and a repeated noun phrase in terms of specificity. For example, the choice between a type anaphor and a repeated noun phrase does not affect the interpretation of 20. With either option, the sentence in 20 means that John bought a specific dog (different from the dog referred to in the first clause).

1. (20)

It is important to note that the interpretation of the antecedent and the interpretation of the type anaphor do not have to be identical. Modality and polarity have a role to play. For example, 21 means that John bought one specific dog, even if the antecedent en hund ‘a dog’ will primarily be interpreted as nonspecific.

1. (21)

The important point is that the potential interpretation of the anaphor is the same as that of the antecedent. This is exactly what is predicted when the anaphor is treated as a surface anaphor.

Even the antecedent's scope options are reflected by the interpretation of the relevant type anaphora. A bare singular object can only take narrow scope, and this is also true of the anaphor that it antecedes. In contrast, a singular noun with an indefinite article can take wide or narrow scope, and this is also true of the anaphor that it antecedes. An example is given in 22.

1. (22)

In 22, the bare singular noun godtepose ‘candy bag’ can only take narrow scope (that is, the kids shared one candy bag, and the adults shared one). In contrast, en godtepose ‘a candy bag’ is ambiguous with respect to scope. It must have the same scopal interpretation in the second clause as in the first clause, however. This fact would follow from a general semantic identity requirement on surface anaphora and their antecedents, as discussed by Williams (1977) and Sag (1980).

3.4 Object Shift and Presentational Sentences

Even if the regular personal pronoun det is definite, surface anaphora are not. Surface anaphora usually represent new information, being focused and accented (Lødrup 1994). They are often realized in the first position of a main clause. When they are not, they are normally realized in the canonical complement position, and are reluctant to undergo object shift (that is, cliticize to the finite verb across an adverb), as noted in Lødrup 1994. This is also true of the relevant type anaphora. Examples are given in 23a and 23b, respectively.

1. (23)

The relevant type anaphora can tolerate object shift when they follow another type anaphor with the same antecedent, thus representing given information. (Borthen does not discuss object shift, but she makes the relevant observation indirectly, see Borthen 2004:4.) An example is given in 24.

1. (24)

Presentational sentences cannot take the verbs discussed so far in this section. Even so, they seem to allow the same kind of type anaphora in their object position (Lødrup 1994). This position is subject to the well-known indefiniteness requirement that prohibits the regular personal pronoun det from occurring there. However, the relevant type anaphor det is not affected by the indefiniteness requirement, as shown in 25. What matters is the indefiniteness of its antecedent, as predicted when det is a surface anaphor.

1. (25)

To sum up, the discussion in section 3 showed that some type anaphora share syntactic and semantic properties with established surface anaphora. It must be concluded, therefore, that they are surface anaphora. The other type anaphora are discussed in section 5.

4.1 Surface Anaphora in Grammar

In section 3, I suggested that some type anaphora are surface anaphora. Surface anaphora must be licensed lexically by the relevant verbs as a realization of their complement function. The option of taking a surface anaphor is not simply a reflection of the verb's general selectional properties. Many verbs that take an object cannot take a surface anaphor (see the discussion in section 5.4). One example is given in 26.

1. (26)

Surface anaphora seem to be limited to the complement position of a verb. In some cases, the head is a multi-word predicate, such as få tak i ‘get hold of’.

Across languages, surface anaphora are found in a small family of constructions that includes VP ellipsis and VP pronominalization, predicate ellipsis and predicate pronominalization, and sluicing (and possibly the cases mentioned in notes 2 and 5). The surface type anaphor extends this family. It has one property that is especially interesting to a theory of surface anaphora: It realizes a nominal argument position. The previously established surface anaphora realize other positions:Footnote ⁹

1. • VP (VP ellipsis, VP pronominalization)

2. • Predicate (predicate ellipsis, predicate pronominalization)

3. • IP (sluicing)

The only other case of a surface anaphor in a nominal argument position that I am aware of is with the main verbs have and existential be, as in example 6b above, and 27.

1. (27)

The main verbs have and existential be have Norwegian equivalents that allow surface anaphora in an argument position. However, this might be related to the fact that they allow surface anaphora in this position when used as auxiliaries.

Even if the surface anaphora discussed here realize nominal argument positions, they share the basic properties of established surface anaphora, as was shown in section 3. In grammar, they can be treated like other surface anaphora. Following Hankamer & Sag 1976, the basic strategy to account for surface anaphora is to assume that a possible antecedent is present in the position of the surface anaphor at some level of representation. A difficult and to some extent theory-internal question is what level of representation is involved. It is a generally accepted view that this level must be more abstract than surface structure. One argument for this view comes from the behavior of negation and quantification. Consider example 28, from Sag 1980:157:

1. (28) John doesn't see anyone, but Bill does.

The second clause in 28 can only mean that Bill sees someone (as pointed out by Sag 1980:157–158). It cannot mean that Bill sees anyone, which is the expected interpretation if surface structure were the relevant level. The same kind of argument can be constructed for the surface anaphora discussed here. Consider example 29:

1. (29)

The object of the first clause in 29 is ingen pølser ‘no sausages’, with the negative quantifier ingen ‘no’. This quantifier is not a part of the antecedent of the surface anaphor, as shown by the fact that 29 does not have the interpretation that I bought no sausages. This means that the antecedent must take the place of the surface anaphor at a level where the quantifier is not a part of the object. (The quantifier must be represented as sentential negation, see Sells 2000.)

This article is not the place to evaluate different theories of surface anaphora. For concreteness, I adopt a syntactic approach à la Hankamer & Sag 1976. The basic assumption is that a possible antecedent is present in the position of the surface anaphor at a deep syntactic level.Footnote ¹⁰ This kind of approach can be found in Lexical Functional Grammar, LFG (Levin 1982 on sluicing, Lødrup 1994 on VP pronominalization). In LFG, the surface representation is the constituent structure. A phonologically realized surface anaphor is present there, while a zero anaphor is not. The deep grammatical representation is the functional structure, with syntactic functions and morphosyntactic features. This is the level where the antecedent is represented.Footnote ¹¹

An antecedent is grammatically active in the position of the surface anaphor. For example, selectional requirements of the predicate restrict possible antecedents through standard well-formedness checking. Example 30a is, therefore, unacceptable in the same way as 30b.

1. (30)

4.2 Features of the Surface Anaphor

A surface anaphor does not contribute any standard grammatical features.Footnote ¹² It might be argued that the surface anaphor discussed here has one grammatical feature, namely [–definite]. This feature would prohibit a definite antecedent, as shown in 31.

1. (31)

However, there is a better account available for 31 that does not involve this feature. If 31 were acceptable, it would have to mean that John bought the same dog as Marit. To express this meaning, however, one would have to use the regular personal pronoun den ‘it.non-neutral’. This pronoun has specifications that the surface anaphor does not have (such as non-neutral and singular). It is a general phenomenon in grammar that a more specific form must be chosen over a less specific form when both are compatible with the content to be expressed. This follows from current versions of the Elsewhere Condition and the principle of morphological blocking (see, for example, Andrews 1990, Halle & Marantz 1994).

Assuming [–definite] would seem to fit in with the fact that the surface anaphor can be used in the object position of a presentational sentence (see 25 above). However, when det is a surface anaphor the indefiniteness requirement is not really a restriction on the surface anaphor, but rather on its antecedent. Assuming [–definite] would also predict that the surface anaphor does not undergo object shift (see 23b above) because indefinite pronouns never do. However, object shift is sometimes possible (see 24 above), which indicates that an account must be more fine-grained and make reference to information structure (see Andréasson 2008).

5.1 The Problem with Like ‘Like’

It was shown above that some type anaphora are surface anaphora. The next question one should ask is whether all type anaphora are surface anaphora. The problem is that the term type anaphor is not defined clearly. It can be shown, however, that some anaphora that Borthen (2003, 2004, 2005) assumes to be type anaphora cannot be surface anaphora. One example is where det appears as the object of the verb like ‘like’. Consider 32 (examples 32–34 are from Borthen 2004:3):

1. (32)

In 32, the antecedent en jente ‘a girl’ is interpreted as specific. To refer to the same girl in the second clause, the personal pronoun henne ‘her’ is the only option. To refer to a different girl, one has to say en annen jente ‘another girl’; det is not acceptable. This situation is very different from the one illustrated by 19 above, where det appears with the verb kjøpe ‘buy’ and shows no restrictions on referring to indefinite noun phrases.

Even if 32 is ungrammatical, Borthen does not consider the possibility that like ‘like’ does not take a type anaphor as its object. The reason is probably that 32 would be acceptable with the plural antecedent jenter ‘girls’ instead of en jente ‘a girl’, as in 33.

1. (33)

An important idea in Borthen 2003, 2004, 2005 is that there is a close connection between type anaphora and bare singulars (that is, singular noun phrases without determiners, see 19 above). Borthen (2004:3) writes:

The verb like ‘like’ does not take a bare singular, as shown in 34. According to Borthen, this explains why the type anaphor cannot have a singular noun phrase as its antecedent (as shown in 32 above).

1. (34)

The problem with this reasoning is that example 32 is not representative. It is possible to construct acceptable examples with the same structure as 32, as shown in 35. This fact undermines Borthen's generalization.

1. (35)

The crucial difference between 32 and 35 is that the acceptable objects in 35 can receive a generic interpretation, while this is hardly possible in 32. The specific plural noun phrase noen jenter ‘some girls’ would also be unacceptable as an antecedent in 32, as opposed to the generic bare plural in 33. The generalization is that det can only receive a generic interpretation when it appears with the verb like ‘like’. This contrast cannot be explained if det is a surface anaphor, and the only conclusion possible is that it is not.

5.2 A New Definition

As mentioned above, it is difficult to determine if all type anaphora are surface anaphora when the term “type anaphor” is not defined clearly. A more appropriate question to ask would be whether all nonagreeing instances of det are surface anaphora. The discussion in the previous section has shown that det occurring with the verb like ‘like’ is not a surface anaphor. Further investigation demonstrates that very many cases of nonagreeing det do not behave like surface anaphora. These cases are discussed in the remainder of this section.Footnote ¹³

Det can be a subject in copular clauses such as 36, independently of the number and gender of its antecedent (Faarlund et al. 1997:331–332, Josefsson 2010:2113, Mikkelsen 2007).

1. (36)

It is clear that det in 36a does not have the properties of a surface anaphor. It can have a definite antecedent, and it can be deictic. The status of det in 36b is more problematic; Mikkelsen 2007 refers to it as a property anaphor.

The examples in 36 represent restricted options that are not discussed further. These cases aside, there remains a large class of nonagreeing det that do not have the properties of surface anaphora. Examples with the verb like ‘like’ were given in 32–35 above. Other examples appear in 37.

1. (37)

The nonagreeing det that is not a surface anaphor is always nonspecific.Footnote ¹⁴ For example, det in 37b could not be used to refer to a specific evening. The status of nonagreeing nonsurface det raises many difficult questions (see Josefsson 2010 for one approach). Tentatively, two separate variants could be postulated: the generic det and the eventive det.

The generic det typically has an antecedent that is low on the definiteness hierarchy. This is also true of the eventive det. It differs from the generic det in having an eventive reading. Its antecedents are “clause equivalent noun phrases” (Teleman et al. 1999:288, my translation). Some examples might receive the one or the other reading. For instance, the subject of 37b can be understood as ‘an evening’ or as ‘having an evening’.

The question is then what relation the generic/eventive det has to type anaphora and surface anaphora. Borthen (2004:3, 2003:41) says explicitly that det is a type anaphor in sentences with the verb like ‘like’ and in 37a above. The point to be established in section 5.3 is that the generic/eventive det is not a surface anaphor. This is shown through a comparison of the properties of surface anaphora with the properties of the generic/eventive det. The cases of nonagreeing det can then be classified as surface anaphora and as the generic/eventive det (putting 36 aside). This leaves no room for the term type anaphor that was used in this article for expository purposes only.

5.3 Properties of the Generic/Eventive det

The first property of the generic/eventive det to be discussed concerns deixis: The generic/eventive det can be deictic. For example, when pointing to a chicken, I can utter 38.

1. (38)

In 38, det can only be generic. If I wanted to say that I disliked that specific chicken, I would have to use the non-neuter personal pronoun den.

The second property involves missing antecedents. The generic/ eventive det cannot be used to refer to a missing antecedent, as in 14 above. Specifically, the generic/eventive det cannot introduce a discourse referent, as shown in 39a; it can only introduce a short-term referent in a modal context (Karttunen 1976), as shown in 39b.

1. (39)

The third property involves structural parallelism. The fact that the generic/eventive det is always nonspecific makes it different from the surface anaphor det, which can be specific or nonspecific. The generic/ eventive det can even have a definite antecedent, as in 40a.Footnote ¹⁵ Furthermore, the generic/eventive det allows a discontinuous antecedent, as in 40b. Neither of these is possible with a surface anaphor.

1. (40)

The fourth property concerns object shift. The generic det has a different prosody than the surface anaphor. It does not have to be accented in a sentence such as 41a, and it can undergo object shift, as in 41b.

1. (41)

   

The eventive det is different in this respect: It is often accented, and it does not undergo object shift, as shown in 42. This is also often the case when det has a clausal antecedent in a nonfactive environment (see the discussion of Danish and Swedish in Andréasson 2008, 2009).

1. (42)

This section has shown that the properties of the generic/eventive det differ from the properties of established surface anaphora that were discussed in section 3. It is, therefore, necessary to conclude that generic/ eventive det cannot be a surface anaphor.

5.4 Selection

An important difference between the surface det and the generic/eventive det is that the surface det is selected by a limited number of verbs, while this is not the case with the generic/eventive det. The generic/eventive det has the distribution of regular noun phrases, and it is not selected by its governing predicate. In general, the generic/eventive det can be a subject or an object of a variety of verbs, as illustrated in 43a,b. It can also be a part of an adjunct PP, as in 43c, and it can be embedded deeper than the complement level of the potentially relevant verb, as in 43d.

1. (43)

   

The question is then which verbs can take a surface anaphor. To provide an answer to this question, the surface anaphor det must be distinguished from the generic/eventive det, which has a wider distribution. Consider the examples in 44:

1. (44)

The preterite verbs in 44 require their objects to have a specific interpretation. As shown above, this is not possible with the generic det; only with the surface det: 44a, with the verb se ‘see’, is acceptable, while 44b, with the verb vurdere ‘assess’, is not. This shows that the former verb takes a surface anaphor, while the latter does not.

Transitive verbs that take the surface anaphor belong to the following verb classes, according to Levin 1993: “change of possession,” “hold and keep,” “creation and transformation,” “perception,” “desire,” “communication,” and “ingesting.” It is difficult to generalize over these verbs. One possibility is the following: It was mentioned in section 2 that a central group of surface anaphor-taking verbs is verbs with a relation of possession in their semantic structure. The group of surface anaphor-taking verbs could be seen as an extension of these verbs. When one makes something (for instance, a chair), or eats something, or sees something, one would “possess” these things in a more or less metaphorical sense. In the same way, when one tells something to somebody, then the other would “possess” this information.

This generalization does not cover verbs of existence and appearance, which can take surface anaphora in the presentational construction (Hagen 2002:197). An example appears in 45.

1. (45)

Identifying the group of verbs that take surface anaphora raises both descriptive and theoretical problems. It is not clear why an extended relation of possession—or existence and appearance—should be relevant to surface anaphora.

5.5 Predicate Nonagreement

An interesting property of noun phrases that can antecede the generic/ eventive det concerns subject agreement. Norwegian has no verb agreement, but an adjectival predicate usually agrees with its subject, as can be seen in 1 above. However, noun phrases that can antecede the generic/eventive det can appear with a neuter singular adjectival predicate independently of their number and gender, as shown in 46.

1. (46)

The example in 46 illustrates a classical problem in Scandinavian grammar. The analyses given could be divided into two groups:

1. (i) The traditional explanation for the neuter singular form is based upon the fact that this form is triggered by a clausal subject. An eventive noun phrase is clausal in some sense, and it is, therefore, natural that it triggers the same form as a clause (see, for example, Faarlund 1977, Borthen 2003:212–223, 336–339). This explanation presupposes that all relevant subjects are eventive.

2. (ii) An alternative view is that the neuter form is related to formal or referential properties of the subject in some way (see Hellan 1986, Enger 2004).

These analyses compete in the literature. However, Josefsson (2006, 2009) argues that they can both be correct. Her proposal is that there are two different types of nonagreeing subjects that seem to parallel antecedents of the generic and the eventive det.Footnote ¹⁶ Josefsson (2009:36) uses the Swedish sentences in 47 as examples.

1. (47)

   

Example 47a only has a generic interpretation; it says something about mustard in general (and not, for example, about some specific mustard, or about having mustard). Example 47b, by contrast, only has an eventive interpretation; it says something about having two lovers (and not, for example, about two specific lovers, or about lovers in general).