NorGram documentation » History » Version 100
- I Overview: Grammatical framework and basic clause structure
- II Phrase Structure Categories
- III Syntactic Variables
- IV F-structure Features
- V Adverbs
- 1. Particle adverbs (ADVprt)
- 2. Attitude adverbs (ADVatt)
- 3. Commitment adverbs (ADVcmt)
- 4. Sentence adverbs (ADVs)
- 5. Negation adverbs (ADVneg)
- 6. Locational adverbs (ADVloc)
- 7. Verb phrase adverbs (ADV)
- 8. Adjectives used adverbially (A)
- 9. Degree adverbs (ADVdeg, ADVdegloc, ADVdegnum)
- 10. Focal adverbs (ADVfoc)
- 11. Interrogative adverbs (ADVint)
I Overview: Grammatical framework and basic clause structure¶
The Norwegian Bokmål computational grammar NorGram, now also extended to Nynorsk, has been under development since 1999 as part of the long-term international Parallel Grammar Project (ParGram). The ParGram grammars are developed within the framework Lexical-functional grammar (LFG) on the computational platform Xerox Linguistic Environment (XLE). An LFG grammar assigns two representations to each reading of a sentence, a c-structure (a phrase structure tree) and an f-structure (an attribute-value graph).
The aim of ParGram is to achieve common principles across languages for the application of LFG. The ideal is that the same grammatical phenomenon in different languages should always be described in the same way, using the same grammatical features and values, so that all differences between the grammars should reflect real differences between the languages and not just different descriptive strategies. The ambition to achieve such parallellism only applies to the f-structure within the ParGram project. Hence the general design of the f-structures assigned by NorGram, including the choice of feature names and their possible values, has its root in the ParGram project. Concerning the c-structure and its relationship to the f-structure, NorGram applies the principles proposed by Joan Bresnan (2001). We give a brief summary of the principles here (section 2), and present their application to Norwegian in the later sections where the rules of Norgram are described.
2. C-structure to f-structure mapping principles¶
2.1 Basic X-bar Structures¶
C-structures are constructed according to the principles of X'-theory, whereby phrases of category XP are projected from lexical entries of category X, with a possible intervening category X':
Figure 1. The basic X' schemata
The specifier and complement phrases are projections of the same general kind as XP (but possibly with other categories instantiating Y 0 and Z 0). The left-to-right order is unspecified, i.e., phrases with the complement preceding the head or the specifier following the head are also in accordance with the schema.
The head X 0 may belong to categories such as N (noun), V (verb), A (adjective), P (preposition), C (complementizer, subjunction), I (‘infl’, i.e., auxiliary or other finite head of IP) or D (determiner). The first four of these are lexical categories, and the last three are functional categories. A lexical category comprises semantically potent lexical entries, whose semantic potency is captured in LFG by the fact that they carry a PRED-feature in their f-structure, with a semantic form as value. A functional category comprises grammatical form words without an independent semantic contribution of the sort captured by means of an LFG PRED-feature. The category ‘I’ is a special case here, since it may contain semantically potent verbal elements (i.e., lexical verbs) in cases where it heads a lexical phrase in addition to heading the functional IP phrase; such a double role is made possible by the concept Extended Head presented below (section 2.4).
Figure 2. Examples of lexical projections
Figure 2 shows some basic examples of lexical projections. The first two structures classically illustrate the parallellism between NPs and PPs captured by X' theory, where genitives and subjects are both treated as specifiers of the lexical projections, and selected PPs and objects are both treated as complements. However, as is frequently the case in grammatical studies, NorGram does not assume that lexical projections have specifiers, but rather assumes only two-level analyses of them, like the third structure in Figure 2. Genitives and subjects in such cases as these are then introduced by means of dominating functional projections embedding the NP or VP.
Figure 3. Examples of functional projections
Figure 3 shows some basic examples of functional projections. The first structure is an IP projected from the modal auxiliary may. English modals are analyzed as functional categories without their own PRED-values. The subject Mary is the specifier of the IP, and the VP leave John is the complement. In NorGram, IP is the only projection which has three levels and a specifier. The second structure in Figure 3 shows a projection of the determiner this, taking the NP theory as its complement. NorGram also treats determiners as heads of DPs rather than as specifiers of NPs. Similarly, the third structure in Figure 3 shows a projection of the complementizer that, taking the IP clause as its complement.
The justification for taking these functional elements as heads is that they determine the syntactic distribution of the phrase. The semantic head-like properties of the complements, which might make these analyses seem counter-intuitive, are still captured in the f-structure, see section ??.
Figure 4. Adjunction
Figure 4 shows the structure of adjunction, whereby a constituent WP such as an adverbial or a topic is adjoined at the edge of an XP phrase by doubling the top XP node and letting the new top node dominate the adjoined element. This is typically the analysis of fronted elements in English, such as unfortunately in the sentence Unfortunately, Mary may Leave John. Apart from the fact that Norwegian differs from English in the incidence of this phenomenon, NorGram does not use the structure shown in Figure 4 in cases of adjunction, but rather lets WP be dominated by the original XP as a sister of YP and X'.
2.2 X-bar-theory in LFG¶
X'-theory is common to many grammatical frameworks. Two principles are important in the incorporation of X'-theory into LFG: Lexical Integrity and Economy of Expression.
Morphological complete words are leaves of the c-structure tree and each leaf corresponds to one and only one c-structure node.
Lexical Integrity distinguishes the use of X'-theory in LFG from its use in grammatical frameworks employing syntactic movement, such as the chomskyan Principles-and-Parameters framework. In such frameworks functional categories in the tree structures may correspond to morphological endings that are attached to their lexical roots by subsequent movement (‘pre-lexical syntax’). Also, head nodes may be left empty. In LFG only overt, complete words can enter the tree structures, and there is no syntactic movement. As a consequence the inventory of functional categories in LFG tends to be much more limited than in the alternative frameworks.
Economy of Expression:
All syntactic phrase structure nodes are optional and are not used unless required by independent principles (completeness, coherence, semantic expressivity).
The principle Economy of Expression crucially hinges on the systematic many-to-one correspondence relation between c-structure nodes and f-structures in LFG. It amounts to a criterion for the pruning of nodes from the trees depending on their contribution to the f-structure. Of particular relevance to NorGram is the implied possibility of omitting the head of a phrase in cases where the f-structure information normally contributed by the head, such as a PRED feature, is contributed from elsewhere (see section 2.4 below about the Extended Head concept). Consequently, not only the first structure in Figure 5, but also the second structure, where the head V node is omitted, is a legitimate structure, provided that certain conditions are met.
Figure 5. Omitted VP head under Economy of Expression.
This means that potential daughters, and not only actual daughters in an individual case, are taken into account when determining the category to which a phrase belongs.
2.3 The Mapping Principles¶
In an LFG analysis, f-structures are mapped from c-structures by means of a function Φ, which is many-to-one. In figure 6, five nodes are mapped to the outermost f-structure, while the two remaining nodes map to one embedded f-structure each. The mapping is encoded by means of equations which are annotated to the c-structure nodes, and included in Figure 6. In an equation annotated to a node n, the symbol ‘↑ ’ refers to the f-structure projected by the mother node of n, and the symbol ‘↓’ refers to the f-structure projected by n itself. Thus, the equation ( ↑ SUBJ)=↓ annotated to the leftmost NP node states that the f-structure projected by IP has an attribute SUBJ with the f-structure projected by the NP as its value.
Figure 6. The F-mapping from c-structure to f-structure
The fact that four nodes project the same f-structure as the top IP appears from the equation ↑ =↓ on those nodes. The equation states that the f-structure projected by the node and the f-structure projected by its mother are the same. A set of c-structure nodes projecting the same f-structure is called a functional domain.
An f-structure is an attribute-value structure where the attributes include grammatical categories such as TENSE and syntactic functions such as SUBJ. The syntactic functions in LFG are divided into subclasses as shown in Figure 73:
Figure 7. Subclasses of syntactic functions
COMPL in Figure 7 covers syntactic functions such as COMP (closed sentential complements with internal subjects) and XCOMP (open sentential complements with controlled subjects). a-fns are argument functions, i.e. those functions that can be referred to in the argument structures of verbs, prepositions, etc. (cp. the value of PRED in Figure 6). TOP (topic) and FOC (focus) are not argument functions. This doesn’t mean that arguments cannot be topics or foci, but such constituents will then also fill an argument function such as SUBJ, OBJ etc. in addition to filling the function TOP or FOC, i.e., they are not arguments in virtue of being topics or foci. Adjuncts, i.e., adverbial elements, also are not arguments. d-fns are discourse functions, i.e., functions whose interpretation essentially relates to what is given or salient in the discourse situation. As shown in Figure 7, SUBJ is the only function which is both an argument function and a discourse function. These subcategories of syntactic functions are referred to in the Mapping Principles spelled out below.
In feature structure grammars a central property of a ‘head’ of a phrase is that it shares its features with its mother. This feature sharing is precisely what is captured by means of the equation ↑ =↓; see, e.g., the annotation of the V daughter of VP in Figure 6. But this means that we now have two ‘head’ concepts: the X’-theory based concept whereby X is the head of XP (e.g., V of VP), i.e., a c-structure head, and the feature-based concept whereby any node with the annotation ↑ =↓ is a head, i.e., an f-structure head. If we allow unconstrained use of these and other notational possibilities emerging from the combination of X’-theory and f-structure projection, the result is a theory which allows monstrously unmotivated syntactic representations, i.e. a theory too unconstrained to retain much linguistic interest. The function of the Mapping Principles is precisely to introduce such constraints in order to capture basic linguistic assumptions and provide the representations with linguistic content. Bresnan formulates the following Mapping Principles:
a. C-structure heads are f-structure heads.
b. Specifiers of functional categories are the grammaticalized discourse functions DF.
c. Complements of functional categories are f-structure co-heads.
d. Complements of lexical categories are the nondiscourse argument functions CF.
e. Constituents adjoined to phrasal constituents are nonargument functions A̅F or not annotated.
We will review the Mapping Principles in turn.
a. C-structure heads are f-structure heads
This constraint means that all nodes that are heads according to X'-theory are annotated with the equation ↑ =↓. This means that for all phrases XP the annotations in Figure 8 follow from the principles.
Figure 8. C-structure heads are f-structure heads.
b. Specifiers of functional categories are the grammaticalized discourse functions DF
This means that all functional phrases FP with a specifier will be annotated as shown in structure (a) in Figure 9, where DF is a discourse function. (b) shows an example.
Figure 9. Specifiers of functional categories are discourse functions TOP, FOC or SUBJ
c. Complements of functional categories are f-structure co-heads
According to this constraint the complement ZP in all functional phrases FP is also annotated with the equation ↑=↓, as shown in Figure 10 (a). (b) and (c) shows two examples.
Figure 10. Complements of functional categories are f-structure co-heads
The examples illustrate how this constraint allows us to cater for potentially conflicting ways of understanding the concept of ‘head’. Thus, the DP analysis in Figure 10 (c) identifies this as a c-structure head because it determines the syntactic properties of the phrase and may replace it in a sentence (This car is new – This is new – *Car is new). At the same time, theory is the semantically central word in the phrase, which is also a head property. The constraint in question lets this semantic head-property of the noun be reflected in the f-structure, since in virtue of the ↑=↓ equation the PRED-value of theory is going to be the PRED-value of the entire phrase as well. Thus, with lexical entries for this and theory along the lines of (2a) and (2b), the result would be the f-structure in Figure 11 for the DP phrase.
(a) this D ( SPEC DET PRED)='this'.
(b) theory N ( PRED)='theory'
Figure 11. The f-structure of Figure 10 (c), given the entries in (2)
Clearly, this double-head analysis works only if at most one of the heads provide a PRED feature on the outermost (or top) level. In both the examples in Figure 10 the top PRED value is provided by the complement, while the functional category occurring as c-structure head does not carry a top-level PRED feature. In the case of Figure 10 (b) this reflects the common analysis of English modals as grammatical markers only providing grammatical features, rather than as fully-fledged verbs providing semanttic predicates. However, the converse is also possible, with the c-structure functional head providing a top-level PRED-feature, while the co-head complement does not. Examples will be presented and explained in the analysis of Norwegian finite constructions in section ??.
d. Complements of lexical categories are the nondiscourse argument functions CF
According to this constraint the complements in all lexical phrases LP are annotated with one of the non-discourse argument functions, i.e., all argument functions except SUBJ. This is shown schematically in Figure 12 (a), with (b), (c) and (d) as examples. As mentioned above we do not assume specifiers in lexical phrases; hence the examples have only two phrase levels.
Figure 12. Complements of lexical categories are the argument functions except SUBJ
e. Constituents adjoined to phrasal constituents are nonarguments functions A̅F or not annotated
According to this constraint an adjoined constituent either carries one of the functions TOPIC, FOCUS or ADJUNCT or is not assigned any function by the rules of syntax. In the latter case the function may be assigned by morphology. This does not occur in Norwegian and will be disregarded here. The former case is shown and exemplified in Figure 13.
Figure 13. Constituents adjoined to phrasal constituents are nonarg functions A̅F
As mentioned at the end of section 2.1, apart from the fact that the Norwegian structures translationally corresponding to (b) and (c) in Figure 13 are not cases of adjunction because of the V2 property of Norwegian (see section ??), NorGram does not use the structure shown in Figure 13 (a) in cases of adjunction, but rather lets WP be dominated by the original (lower) FP as a sister of YP and F'.
Figure 14 illustrates how the mapping principles constrain the functional annotations in a c-structure for the English sentence Unfortunately I believe that Mary will leave John.
Figure 14. Annotations constrained by the mapping principles.
Some of the nodes have been indexed for ease of reference. The head V of the lexical phrase VP:7 is annotated with ↑=↓ according to principle a, and the complement NP is annotated with one of the non-SUBJ argument functions, ( ↑ OBJ)=↓, in accordance with principle d. IP:5 is a two-level functional phrase with specifier and complement (the latter dominated by the intermediate head I':6). Accordingly, the complement daughter of I':6 is annotated with ↑=↓ according to principle c, and its head daughter V is also annotated with ↑=↓, according to principle a. Furthermore the specifier NP daughter of IP:5 is annotated with (↑ SUBJ)=↓ in accordance with principle b, which restricts the annotations to one of SUBJ, TOP or FOC. CP:4 is another functional phrase (without specifier this time), whose head C and complement IP:5 are both annotated with ↑=↓ in accordance with principles a and c. The lexical phrase VP:3 has head and complement annotations in accordance with principles a and d, just like VP:7, except that the argument function of the complement is COMP rather than OBJ. IP:2 is a functional phrase whose head has been pruned away in accordance with the principle Economy of expression discussed in section 2.2. Specifically, there is no auxiliary in this main clause, and therefore only the specifier NP and the complement VP:3 of IP:2 are found in the tree, annotated in accordance with principles b and c, respectively. This exemplifies the point made in 2.2 that a phrase label such as ‘IP’ is chosen based on potential daughters, and not only based on the actual daughters in the given tree. Finally, IP:1 is a ‘duplication’ of IP:2 by adjunction, and the annotation (↑ ADJUNCT)=↓ of the AP daughter is in accordance with principle e, while IP:2 gets the annotation ↑=↓, since by adjunction the duplex nodes can in a sense be seen as two versions of the same node, and hence share all features.
2.4 On the nodes I and V and the notion ‘Extended Head’¶
In virtue of the annotation ↑=↓ on all the dominated nodes, all the encircled nodes in Figure 14 project the same f-substructure. The nodes accordingly constitute a functional domain. Thus the auxiliary will and the main verb leave belong to the same functional domain, with the consequence that only one of them can have a PRED feature. A central feature of English syntax, distinguishing it from Norwegian, is that it is the distinction between auxiliary and main verb which determines the distribution between the I position and the V position in this structure in English, while finiteness is secondary. In English, the main verb is always under V – i.e., adjacent to its complements (OBJ, COMP etc.) – whether it is finite or not. Thus, the finite form believe in Figure 14 is analyzed as a V under VP:3, and not as the head I of IP:2. This is supported by the fact that negation and other sentence adverbials always occur before the main verb, allowed to do so by do-support in cases where no other auxiliary is motivated; cf. Figure 15.
Figure 15. Negation always occurs before the main verb in English.
Hence there is no motivation related to word-order for assuming that the main verb ‘moves’ to I in simple sentences like Mary leaves John or I believe this.
In Norwegian, the distinction which determines the distribution of forms between I and V in main clauses is the one between finite and non finite, while the distinction betwen auxiliary and non-auxiliary is irrelevant. In order to give some more typological background for the analysis of Norwegian we will briefly review Bresnan’s examples from Welsh.
Welsh is a verb-initial language: in main declarative clauses the finite verb, main or auxiliary, is the first element in the sentence. This is illustrated in Figure 16, which shows the analysis of two ways of expressing the proposition ‘John saw the dragon’ in Welsh.
Figure 16. Two ways of saying John saw the dragon in Welsh.
The form weld in (a) in Figure 16 is a non-finite form of the verb ‘to see’, while the form gwelodd in (b) is the finite, past form. Thus, Welsh gives the alternatives of either saying ‘John did see the dragon’, with do-support, or ‘John saw the dragon’, without. We may notice the following features of the analyses.
- The language being verb-initial, there is no specifier of IP, which dominates I and its complement directly.
- The complement of I is S, not VP, an exocentric phrase (see below), since it dominates the subject as well as the VP.
- Both the finite auxiliary and the finite main verb (in the absence of an auxiliary) occur in I.
- When the main verb is in I (as in (b)), the VP does not dominate any V head (cp. Figure 5 with discussion).
The phrases constructed according to X’-theory discussed so far are all endocentric: They are projected by a lexical or functional head. However, the theory also needs the concept of a non-projected, exocentric phrase, especially in order to cope with non-configurational languages that may (although must not) lack a VP phrase altogether. S is such a phrase, and once introduced in the theory, it may also turn out to be relevant in the analysis of basically configurational languages. In Welsh it is motivated to assume S as a complement of IP, since the node dominates subject and predicate, but not a finite verb, and is therefore not an IP.
The fact that the VP in (b) in Figure 16 does not dominate a V head is licenced by the fact that the f-structure projected by the VP still gets an f-structure head with a PRED-value in this configuration. In Figure 17 the relevant functional domains are encircled in the trees.
Figure 17. Functional domains and extended heads in the Welsh examples.
The systematic relationship between c- and f-structure in LFG enables us to define an extended head concept, extending the concept found in X’-theory. Bresnan, building on work by Jar, Zaenen and Kaplan, defines it as follows:
Given a c-structure containing nodes N, C, and c- to f-structure mapping Φ, N is an extended head of C if N is the minimal node in Φ-1(Φ(C)) that c-commands C without dominating C.
(A c-commands B if every node properly dominating A also dominates B.)
To see how this definition applies to the structures in Figure 17, supplying the VPs with heads, let first C in the definition be the VP in (a). The functional domain of VP, Φ-1(Φ(VP)), is encircled in the tree. Four nodes in (a) c-command the VP without dominating it: the I, the V, and both NPs. Notice in particular that the two nodes dominated by VP, V and NP(draig), also c-command their mother, since all the nodes properly dominating V and NP(draig) – IP, S and VP – also dominate the VP. (A node dominates itself, but properly dominates only its daughters and their descendants.) Among these four c-commanding nodes only two are within the functional domain of the VP, i.e., I and V. Of these two, the minimal node c-commanding VP is the one which does not c-command any other relevant c-commanding node. Since I c-commands V, the minimal node is V, which is hence identified as the head of VP – in this case identical with the head identified by the unextended head definition. In (b), on the other hand, the VP has no daughter within its functional domain and therefore no head according to the unextended definition. Consequently I is the minimal c-commanding node within the functional domain of the VP, and is hence identified as the head of the VP according to the Extended Head definition. Thus, the notion of Extended Head captures the intuition which is expressed by head movement in a transformational grammar: Metaphorically speaking, the head of VP has been ‘raised’ to the I position. The many-to-one relation between c- and f-structure elements allows the expression of this relationship without the assumption of syntactic movement.
The notion of Extended Head also entails an extension of the definition of functional phrases in X'-theory, which so far has been characterized as phrases with functional elements – grammatical form words – as heads. Lexical elements, such as the verb ‘see’, can occur as heads of functional phrases after all, in cases where they also are extended heads of lexical phrases such as VP. Intuitively, the finite lexical verb has a dual role as head of VP and as carrier of the finite categories of the sentence, and it is in its latter capacity that it occurs as he head of IP.
3. The Basic Clause Structure in Norwegian¶
The structure of Norwegian main declarative clauses is exemplified in (3):
We may note the following properties:
- There is a position before the finite verb in I (unlike Welsh).
- There is only one position before the finite verb: No adjunction of adverbs or topics (unlike English).
- There is a subject position after the finite verb (like Welsh, but unlike English).
- The finite verb – main or auxiliary – is always in the leftmost verbal I position, before negation (like Welsh, but unlike English).
This motivates representations like those in Figure 18 for the sentences Dessverre ville deltagerne ikke lære syntaks and Deltagerne ville ikke lære syntaks.
Figure 18. Main declarative clauses with two subject positions.
The V2 property of Norwegian is captured by the specifier position of IP ([SPEC, IP]) in combination with the second possible subject position after I. With an adverb in the [SPEC, IP] position the subject occurs after I: there is no adjunction of the type found in English. As in Welsh, the complement of IP is S rather than VP, since the complement has a subject position. However, unlike Welsh the VP under S in main clauses always is without a head daughter, the head always being provided from I by the Extended Head definition – even when the verbal element in I is an auxiliary (as in the examples in Figure 17). The reason for this is that Norwegian auxiliaries are fully-fledged verbs able to take their own complements, and hence expressing predicates of their own with their own PRED features. Therefore the verb phrase lære syntaks in Figure 18 is analyzed as an XCOMP complement under the main VP, and not as the main VP itself, whose extended head is the finite auxiliary ville, not the ‘main verb’ lære.
The syntax of typical subordinate clauses differs from that of main clauses; consider the examples in (4):
We may note the following properties:
- Negation and other sentence adverbs occur before the finite verb – main or auxiliary.
- The subject can only occur before the finite verb.
- Hence the finite verb is always adjacent to its complements: subordinate clauses are not V2.
These properties can be captured by basing the analysis on the same S phrase as the one found in main clauses, whereby the differences will follow automatically from the embedding context of the S. Consider Figure 19:
Figure 19. The same S phrase in main and subordinate clauses.
The S phrase in Figure 19 (a) is identical with the S phrase in Figure 18 (a). In Figure 19 (b) the S is inserted as the complement of CP, with no intervening IP. Hence there is no I node to provide an extended head for the VP, which therefore has to dominate its own V head. In this way the relative position of adverb and finite verb in subordinate clauses follows from the absence of a dominating IP.
The assumption that we are faced with the same S phrase in main and subordinate clauses is corroborated by the fact that the phrases share further properties, e.g., scrambling possibilities. Thus, in both clause types the order of the negation and the subject may be switched; cf. (5):
Dessverre vil [S deltagerne ikke [VP [VP lære syntaks]]]
Dessverre vil [S ikke deltagerne [VP [VP lære syntaks]]]
hvis [S deltagerne ikke [VP vil [VP lære syntaks]]]
hvis [S ikke deltagerne [VP vil [VP lære syntaks]]]
Some nominal subclauses (at-clauses) have main clause word order, but this posibility is restricted to contexts where the speaker can be said to endorse the proposition expressed by the subclause. Thus main clause word order is acceptable in (6a), but seems strange in (6b):
(a) Kari sa at hun var ikke syk
(b) ?? Jeg tviler på at Kari var ikke syk
(c) OK: Jeg tviler på at Kari ikke var syk
The possibility illustrated in (6a) is taken care of by letting IP be a possible complement of nominal CP, in addition to S. This also opens the possibility for nominal subclauses with topics occuring as specifiers of IP: Kari sa at syk var hun ikke.
This connection between the presence of an IP and the endorsement of the proposition pinpoints the semantic import of the IP domain. The IP is the grammaticalized modal core of the sentence; this is where the speech act «happens».
Go on to II Phrase Structure Categories
1 NorGram has a larger inventory of lexical and functional categories than the basic types mentioned above, making some more fine-grained distinctions especially among the determiners (D).
2 Bresnan 2001:92
3 Bresnan 2001:91
4 Bresnan 2001:97 f.
 Bresnan 2001:102
 The principle of Lexical Integrity discussed in section 2.2 renders irrelevant the arguments for such movement in the Principles and Parameters theory, where tense categories are generated under I and verbs need to move there in order to be tensed.
 NorGram uses a more fine-grained inventory of node labels than the ones used here, as described in section 4.
 For one thing, Norwegian auxiliaries have the full morphology of verbs, including non-finite forms (unlike English modals), for another, they can also take nominal complements: Han vil dette, Han må det.
 CP is the category of subordinate clauses, a functional phrase projecting the complementizer C.