1. Takafumi Maekawa.hwp Linguistic Research 30(3), 407-433 An HPSG analysis of ‘a beautiful two weeks’*1 Takafumi Maekawa (Ryukoku University) Maekawa, Takafumi. 2013. An HPSG analysis of ‘a beautiful two weeks’. Linguistic Research 30(3), 407-433. In this article we investigate a type of noun phrase in English, which is exemplified by a beautiful two weeks and a lucky three students. We call such examples determiner-modifier-numeral-noun constructions (DMNNCs). DMNNCs look similar to what we call the numeral-modifier-noun constructions (NMNCs) such as two beautiful weeks and three lucky students, but differ in several respects: e.g., an indefinite article can occur in DMNNCs but cannot in NMNCs; the numeral follows the modifier in DMNNCs but precedes the modifier in NMNCs. In this article we will mainly discuss the syntax of DMNNCs in some detail and consider how they might be analysed within Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1987, 1994). While sketching some analyses which have been proposed for DMNNCs, we look at some data which seems problematic to these analyses. We then develop an analysis of DMNNCs in which the peculiarities of the construction are attributed to the special constructional constraints. We argue that HPSG can provide a satisfactory account of these properties of DMNNCs. (Ryukoku University) Keywords Head-driven Phrase Structure Grammar, noun phrases, numerals, determiners 1. Introduction In this article we investigate a type of noun phrase in English, which is exemplified by the examples in (1). * Earlier versions of this article were presented at the 36th Annual Meeting of the Kansai Linguistic Society, Osaka, Japan, 11 June 2011, the 14th Korea-Japan Workshop on Linguistics and Language Processing at Seoul, South Korea, 8-9 March, 2013, Linguistics Society of Kobe, Kyoto, Japan, 20 April 2013, the 9th Seminar on English Grammar and Usage, Osaka, Japan, 5 August 2013 and the 20th International Conference on Head-​driven Phrase Structure Grammar at Berlin, Germany, 28-29 August 2013. I am grateful to members of the audience for their feedback and comments. I would like to thank Bob Borsley and Frank Van Eynde for their valuable comments and discussions. Thanks are also due to the anonymous reviewers for their insightful comments and suggestions. Any shortcomings are my responsibility. 408 Takafumi Maekawa (1) a. a beautiful two weeks b. a lucky three students We call such examples determiner-modifier-numeral-noun constructions (DMNNCs). DMNNCs look similar to what we call the numeral-modifier-noun constructions (NMNCs), which is illustrated by the following examples. (2) a. two beautiful weeks b. three lucky students These two constructions look similar in that a modifier and a numeral expression appear together, but differ in the following respects. First, NMNCs cannot have an indefinite article but DMNNCs requires one though the head noun is plural. The following examples illustrate that NMNCs cannot have an indefinite article. (3) a. (*a) two beautiful weeks b. (*a) three lucky students The examples in (4) show that DMNNCs are ungrammatical without a determiner. (4) a. *(a) beautiful two weeks b. *(a) lucky three students The second difference is the order of the numeral and the modifier: the numeral precedes the modifier in NMNCs but the numeral follows the modifier in DMNNCs. DMNNCs cannot have the same order as NMNCs. (5) a. *a two beautiful weeks b. *a three lucky students In this article we will mainly discuss the syntax of DMNNCs in some detail and consider how they might be analysed within Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1987, 1994). The following are some examples of DMNNCs from BYU-BNC. An HPSG analysis of ‘a beautiful two weeks’ 409 (6) a. This shrub bed was a full sixty yards from where they now stood, but just over half that distance from the treeline away over to their left. b. Major students take a further two units and joint students an additional one, in Theatre Production. c. He must have had the equivalent of a good twelve cups of black coffee. d. Even without Jackson, Aberavon chalked up a comfortable 24-0 success to leave the Police rooted to the foot of the table having conceded a massive 180 points in just four games. e. Reggae star Shabba Ranks has finally scored his first solo hit after releasing a staggering 30 singles on 15 different labels since 1988. It will be argued that HPSG can provide a satisfactory account of various properties of DMNNCs. The organisation of the article is as follows. In the next section we will sketch some analyses which have been proposed for DMNNCs and at the same time look at some data which are problematic for them. Section 3 introduces the framework of HPSG. In section 4 we will develop an analysis of DMNNCs within HPSG, and then we will look at how it might be able to deal with the facts. In section 5 we will look at some further data which we will argue is no problem to our approach. Section 6 is the conclusion. 2. Analyses of DMNNCs There have been some discussions of DMNNCs in the syntactic literature, but it seems that there are no fully worked out analyses so far. Jackendoff (1977: 128-130) assigns the following structure to DMNNCs (See also Ohna (2003)). 410 Takafumi Maekawa (7) N''' | N'' N''' N' | Art N'' N | | a A''' N' weeks | | beautiful N | two Jackendoff (1977) argues that numerals are nouns. As a singular noun, the numeral two in (7) requires a specifier and the indefinite article plays the role. They make a constituent, which serve as a modifier of the head noun weeks. Examples in (8) argue against this view, however. (8) a. Portis finished the year with a career-low one touchdown in eight games. (COCA1) b. After the two point cut in interest rates since the UK’s departure from the ERM, minimum lending rate was cut a further one point to 7% from 13 November 1992, the lowest level for nearly 15 years. (BYU-BNC) As singular countable nouns, touchdown and point in (8) require a specifier. However, if the indefinite article is a specifier of the numeral, as in (7), touchdown and point in (8) do not have a specifier that they require. In the analysis proposed by Ellsworth et al. (2008: 28), the adjective and the numeral make a constituent and the indefinite article is a specifier of the head noun 1 the Corpus of Contemporary American English An HPSG analysis of ‘a beautiful two weeks’ 411 (See also Honda (1984)). Approaches along these lines can be schematically depicted as in (9). (9) NP Det N' XP N | a beautiful two weeks For this analysis (8) is no problem: the indefinite article is a specifier of the head noun, and it can satisfy the requirement from the head noun if it is a singular countable noun. However, there is evidence that seems problematic to this approach. First, the examples in (10) are DMNNCs with the numeral and the following noun conjoined. (10) a. the long [2 hours and 14 minutes] b. an amazing [12 performances and 602 rehearsals] (Ionin and Matushansky 2004: 111) In (10) the adjectives long and amazing do not make a constituent with the following numerals. They clearly combine with the phrase in square brackets, in which two combinations of the numeral and the head noun are conjoined. It is difficult, if not impossible, to accommodate these examples with an analysis like (9). Second, the modifier in DMNNCs does not have to be an adjective: a relative clause can occur instead, as illustrated by the following attested examples.2 (11) a. By the end of the four days, my group and I were ready to leave, but it was [a four days [that we will all look back on with great memories]]. (http://wilsonlanguagegrants.blogspot.com/) 2 I would like to thank an anonymous reviewer for providing me with the data in (11). 412 Takafumi Maekawa b. Still, the internet is alive with pictures, comments, ideas, threads, and remembrances of [a three days [that we will never forget]]. (http://pulpmachine.blogspot.jp/2011/05/pulp-ark-day-three-as-doors- closed_19.html) c. [A three days [that will include the finest assortment of Italian wines ever to be offered in Hong Kong]], including the first ever direct consignment from the cellars of Gaja. (http://www.businesswire.com/news/home/20121119005886/en/Dom aine-de-la-Romanee-Conti-Dominates-Acker-Merrall) As relative clauses, the modifiers in (11) should occur post-nominally. The numeral and the relative clause are disconnected and hence impossible to make a constituent. Van Eynde (2006: 154, 169) assigns the following structure to the Dutch equivalent of a good forty pages, namely, een geode veertig pagina’s. (12) N N N | Pron N pagina’s | een Adj N | | geode veertig In this structure the adjective modifier and the numeral make a constituent. Therefore, the objections that we raised against the analysis along the lines of (9) are also applicable here. This analysis is incompatible with the facts shown in (10) and (11). Gawron (2002) and Ionin and Matushansky (2004: 110ff; 2006: 323ff) claim that the numeral and the following noun make a constituent. An HPSG analysis of ‘a beautiful two weeks’ 413 (13) NP Det N' | a Adj N' | beautiful N N | | two weeks This structure can capture (8), (10) and (11). First, the indefinite article is a determiner of the noun in (13), so there is no problem even if the head noun is a singular count noun. Second, the numeral and the head noun make a constituent, so the data in (10) can be accommodated easily. Finally, the modifier is adjoined to the numeral-noun combination in (13): the examples in (11) can be captured by this approach because both the adjective and the relative clause can be analysed as adjoined to the numeral-noun combination: the only difference is that the adjective is a pre-nominal adjunct while a relative clause is a post-nominal adjunct. Thus, the structure in (13) can capture the relevant data quite nicely and we will also assume this constituent structure. However, Gawron’s (2002) and Ionin and Matushansky’s (2004: 110ff; 2006: 323ff) analyses are problematic in details. Gawron (2002) argues that the numeral and the following noun make a measure phrase, which is unspecified for number and therefore allows a. For Ionin and Matushansky (2004: 110ff; 2006: 323ff) the numeral is the head noun and the following noun is its complement. Like other nouns, numerals allow adjectival modification. In their analysis the numeral is singular in number, so an indefinite article is allowed in this construction. For both of these analyses, it is not clear why the indefinite article occurs only in the presence of a modifier. Moreover, Ionin and Matushansky’s analysis cannot accommodate the cases where the determiner is plural.3 3 For Gawron (2002) this is no problem. Measure phrases can take a plural determiner, as the following examples from BYU-BNC show. (i) These two groups of children have quite different characteristics, care histories, prognoses and 414 Takafumi Maekawa (14) a. While this therapist has been working away to make things better for these lucky 30 people, in this country another 600,000 babies have been born. (http://www.orgonomyuk.org.uk/Self-Regulation.html) b. Those busy nine days of snow we had were beginning to stretch us to say the least. (http://norfolkwinter.com/2013/02/08/meet-the-gritters/) In addition, Ionin and Matushansky largely depend on the idea that the numeral is the head noun. This would be an exceptional case where a noun takes a nominal complement without a preposition. Other things being equal, it is preferable not to have such an exception. It seems, then, although right-branching structures like (13) can capture all the data observed in this section, the particular analyses by Gawron (2002) and Ionin and Matushansky (2004, 2006) contains some problems. In the following sections we will consider how the data should be analysed. We will first present the theoretical assumptions of HPSG which are relevant to the present study. Then we will see how they can deal with the rather peculiar properties of DMNNCs. 3. HPSG In HPSG linguistic expressions have a complex feature makeup encoding their syntactic, phonological and semantic properties. Within the version of HPSG assumed in this article, syntactic properties include the features HEAD, which encodes information shared between a phrase and its head, such as information about part of speech, and MARKING (MKG), which indicates whether the expression involves a determiner or a numeral, or whether it can stand alone without these elements (Van Eynde 2006, etc.). Within this framework numerals denoting numbers larger than 1, such as two, will have the partial lexical description in (15). needs for social work support. (ii) These two pieces of legislation provide interesting contrasts in approaches to the provision of social security. An HPSG analysis of ‘a beautiful two weeks’ 415 (15) Here the value of the feature HEAD identifies two as a noun. The MKG feature has a value whose type is marking. Much HPSG work assumes that linguistic objects are typed and organised into a hierarchy. We can propose that MKG values can be organised into the following type hierarchy. (16) marking unmarked marked non-numeral numeral incomplete bare The MKG value is marked if the expression contains a determiner or it itself is a determiner, and unmarked otherwise. The distinction between numeral and non-numeral is to differentiate between nominals with a numeral and those without. Numerals themselves are [MKG numeral], like two in (15). Plural nouns and abstract nouns are [MKG bare] because they can stand alone without a determiner. Singular countable nouns such as week and student have an incomplete value because they require a determiner.4 4 Here we are assuming a constraint like the following. (i) [ARG-ST nelist()] → [MKG ¬incomplete] The ARGUMENT-STRUCTURE (ARG-ST) feature contains information about the word’s subcategorisation. This constraint states that if the ARG-ST list is a non-empty list (indicated as nelist), then the MKG value of the members of the list should not be incomplete. This constraint will exclude sentences like those in (ii), where singular countable nouns girl and park appear without a determiner. (ii) a. *Paul met girl. 416 Takafumi Maekawa We further assume that elements such as determiners, numerals and adjectives are ‘functors’: non-heads that select their head (Allegranza 1998; Van Eynde 2006, 2007; etc.). SEL(ECT) is a part of the HEAD value, specifying what kind of word/phrase it selects: (15) indicates that two selects a bare plural nominal. Following Ellsworth et al. (2008: 32), we assume that the feature AGR(EEMENT) is a part of the MKG value. AGR represents morphosyntactic properties of the expression. The NUMBER (NUM) feature indicates whether a sign is singular or plural. The AGR|NUM values in (15) shows that the expressions that two selects are morphosyntactically plural. NP-internal agreement is based on the AGR value (Kathol 1999, Kim 2004, Wechsler and Zlatić 2000). The combination of the functor and the head is ensured by the constraint imposed on the phrase type head-functor-phrase (hd-funct-ph). The constraint on this phrase type is given below. (17) Phrases in general are composed of the head daughter and some non-head daughters. The HD-DTR (HEAD-DAUGHTER) feature refers to the head daughter of a phrase and the NON-HD-DTRS (NON-HEAD-DAUGHTERS) to the non-head daughters.5 The value of the SYNTAX-SEMANTICS (SYNSEM) feature contains the sign’s syntactic and semantic information. The constraint in (17) states that in phrases of type hd-funct-ph the SEL value of the non-head daughter is required to be token-identical to the SYNSEM value of the head daughter. This means that the non-head daughter selects the head daughter. Following Van Eynde (2006, 2007), we assume that the hd-funct-ph is a subtype of a head-adjunct-phrase (hd-adj-ph) type, which in turn is a subtype of a headed-phrase (hd-ph) type. The constraints for hd-ph is given below. b. *There are many children in park. 5 Following Sag (1997) the head daughter and the non-head daughters are listed separately. An HPSG analysis of ‘a beautiful two weeks’ 417 HEAD 1 - HD-DTR HEAD 1 hd ph é ù ê ú® é ùê ú ë ûë û (18) Constraint in (18) states that the HEAD value of a headed phrase (a phrase of the hd-ph type) is structure-shared with the HEAD value of the head daughter. (Pollard and Sag 1994: 34). This means that information about parts of speech is shared between the phrase and the head daughter. Head-adjunct phrases (phrases of the hd-adj-ph type) are subject to the constraint in (19). (19) This constraint states that in head-adjunct phrases the MKG value is shared between the mother and the non-head daughter. Given the above lexical item and constraints we have structures like (20) for two weeks. (20) The combination of two and weeks in (20) is a structure of a head-functor phrase (a phrase of the type hd-funct-ph), where the functor two selects and combines with 418 Takafumi Maekawa weeks via the feature SEL. Constraint (18) ensures that the HEAD value of a mother is inherited from a head daughter. The MKG value of a head-functor phrase comes from the functor daughter, as specified by constraint (19). Therefore, the MKG value of the mother node is numeral although that of the head daughter is bare. One point of note here is that (20) is an informal representation. As the constraints in (17) and (18) show, the structure of phrases is analysed in terms of the features HD-DTR and NON-HD-DTRS. In the rest of this article, however, constituent structures will be represented in the form of traditional syntactic trees. Adjectives normally select nominals which do not include a numeral expression. The following examples illustrate this. (21) a. beautiful [(*two) weeks] b. lucky [(*three) students] Given this fact, we can propose that adjectives have the following syntactic properties. (22) Here the value of SEL specifies that the nominal selected by an adjective is [MKG non-numeral], indicating that it does not include a numeral expression. Note also that the MKG value of the adjective comes from the noun it selects. This ensures that, for example, the MKG value of beautiful in beautiful weeks is bare because that of weeks is. We have structures like (24) for the NMNCs in (2). (23) a. four beautiful days b. three lucky students [= (2)] An HPSG analysis of ‘a beautiful two weeks’ 419 (24) The combination of beautiful and weeks in (24) is a structure of a head-functor phrase, where the functor beautiful selects and combines with weeks via the feature SEL. The MKG value of the plural noun weeks is bare, which is a subtype of non-numeral, conforming to the selection restriction of adjectives indicated in (22). The HEAD value of a mother is inherited from a head daughter. The MKG value of a head-functor phrase comes from the functor daughter. The MKG value of beautiful weeks originally comes from weeks, as specified in the description of beautiful given in (22). The numeral two selects and then combines with this phrase via SEL, again constituting a head-functor phrase. Having introduced relevant features and constraints of HPSG, we will now see how the apparently puzzling data observed above can be accounted for in this framework. 4. ‘A beautiful two weeks’ in HPSG In this section we will provide an analysis which can provide a satisfactory account of the data. However, we will first consider an analysis in which the 420 Takafumi Maekawa properties of a lexical item are important in determining the properties of DMNNCs. It will be argued that this analysis seems unsatisfactory. As indicated above, we are assuming a right-branching structure for DMNNCs (See (13)). The structure can be schematically represented in the following way. (25) [a [beautiful [two weeks]]] (25) shows that the modifier fist combines with the numeral-noun combination, and then the resulting constituent combines with the indefinite article. In structures other than DMNNCs, such as NMNCs, a modifier selects a nominal without a numeral expression, as shown in (26). (26) a. two beautiful [weeks] b. *beautiful [two weeks] Thus, it might be possible to say that the modifier in DMNNCs has a special syntactic property whereby it modifies, and therefore selects, an NP with a numeral expression (cf. Ellsworth et al. 2008: 28). From this perspective, it would be possible to say that the indefinite article is required by the modifier in DMNNCs: the indefinite article never occurs with the numeral-noun combination unless there is a modifier, and the indefinite article has to occur if there is a modifier. (27) a. a *(beautiful) two weeks b. a *(lucky) three students (28) a. *(a) beautiful two weeks b. *(a) lucky three students [= (4)] Thus, one might argue that the modifier is the most significant element in building up DMNNCs. On this approach, the lexical description of the modifier of DMNNCs would be something like the following. An HPSG analysis of ‘a beautiful two weeks’ 421 (29) (29) indicates that the adjective with this information selects a nominal with a numeral expression in it and its MKG value incomplete necessitates the appearance of the determiner. Given (29), we will have structures like (30). (30) An analysis of two weeks was given in the last section. The combination of beautiful and two weeks in (30) is a head-functor phrase, in which the adjective selects the NP with a numeral in it. Note that the MKG value of the adjective is specified as incomplete. This value is inherited to the phrase node and necessitates the occurrence of a determiner. There is an objection to this analysis. It is incompatible with the sound generalization that a determiner is something required by nominal elements: there are no other cases in English where a determiner is required by an adjective. One might suppose that the following phenomenon is such a case. The examples are cited from 422 Takafumi Maekawa Swan (2005: 132). (31) a. We need a secretary with *(a) first-class knowledge of German. b. My parents wanted me to have *(a) good education. Uncountable nouns knowledge and education are not normally used with a determiner, but in these examples, where they are modified by the adjectives first-class and good, respectively, the determiner is obligatory. Swan (2005: 132) states, however, that most uncountable nouns cannot be used with the indefinite article even when they have an adjective. The following examples are again Swan’s (2005: 132). (32) a. My father enjoys (*a) very good health. b. We’re having (*a) terrible weather. It is obvious, then, that these cases are entirely different from DMNNCs, where the determiner is always obligatory. It looks as if the DMNN were the only case in which the adjective requires the determiner. As stated above, approaches like this miss an important generalisation that a determiner is required by nominal elements. We conclude, then, that this is not a satisfactory analysis. We turn now to an analysis which does not miss any important generalisation. This is an analysis in which a special construction is proposed for DMNNCs. As stated above linguistic expressions are classified and organised into a hierarchy in HPSG. We can propose that the classification of phrase includes the following types.6 6 Only the portion of the hierarchy which will be relevant for our discussion is given here. Among other subtypes of hd-ph are head-subject-phrase and head-complement-phrase, which are responsible for combining a head with a subject and with some complements, respectively. An HPSG analysis of ‘a beautiful two weeks’ 423 (33) We introduced the hd-ph, hd-adj-ph and hd-funct-ph types in the last section: the hd-funct-ph type is a subtype of the hd-adj-ph type, which in turn is a subtype of the hd-ph type. Here we have a head-independent-phrase (hd-indep-ph) type, which is another subtype of the hd-ph type. Moreover, the hd-indep-adj-ph type is an immediate subtype of both the hd-adj-ph and the hd-indep-ph type. The Big Mess construction such as so big a mess, discussed by Van Eynde (2006. 2007), are subtypes of the hd-indep-adj-ph type. Finally, the modifier-numeral-phrase (mod -num-ph) type is a subtype of the hd-indep-ph type.7 As argued above, the modifiers in DMNNCs are unusual in that they combine with an NP with a numeral expression. We will assume that a type hd-indep-ph (Van Eynde 2006, 2007), which is subject to the following constraint, licenses such combination. (34) This is a subtype of hd-ph, but not of hd-funct-ph. Therefore the non-head daughter of this phrase type does not lexically select its head sister: The SEL value is none. The combination of the daughters is guaranteed merely by the identification of the indices. Especially for DMNNCs, we will propose a mod-num-ph type, which is a subtype of hd-indep-ph. 7 This type hierarchy is slightly different from Van Eynde’s (2006, 2007). In his model the hd-indep-ph is a subtype of the hd-adj-ph type, and asymmetric coordination and apposition in Dutch and the Big Mess construction in English are subtypes of the hd-indep-ph. 424 Takafumi Maekawa (35) Constraint (35) states that in a phrase of the mod-num-ph type a nominal with a numeral expression combines with the non-head which does not contain a numeral expression. Phrases of this type require a determiner because the constraint specifies that its MKG value should be incomplete. We assume that a lexical description of the indefinite article a(n) is something like (36). (36) This states that the indefinite article a(n) selects a nominal whose MKG value is non-numeral and the AGR|NUM value is sg. The combination with the head daughter is licensed by a constraint on the hd-funct-ph given in (17). The lexical description of an adjective was given in (22), but we will modify it slightly. (37) Here the value of SEL is amended so that it can be none. This means that the adjectives either select the nominal with the [MKG non-numeral] specification or do not select anything. The latter option enables it to appear in the phrases of the hd-indep-ph type. The above constructional constraints and lexical descriptions work together for An HPSG analysis of ‘a beautiful two weeks’ 425 characterising DMNNCs. Our analysis of a beautiful two weeks is given in (38). (38) An analysis of two weeks was given in the last section. The combination of beautiful and two weeks is an instance of a type mod-num-ph. As a subtype of hd-indep-ph it is constrained by the constraint in (34): the indices of the two daughters are identified. This phrase type is a subtype of hd-ph, so the mother node has the same HEAD value as its head daughter. Note also the MKG value is incomplete: unlike the hd-adj-ph type and its hd-funct-ph subtype, the MKG value is not inherited from the non-head daughter. Thus, a can combine with beautiful two weeks although the latter contains a plural nominal head. We will now look at how the above analysis can deal with the idiosyncratic properties of DMNNCs observed in the earlier sections. First of all let us consider the fact that the determiner is obligatory in DMNNCs. The data, observed in (4), are repeated in the following. (39) a. *(a) beautiful two weeks b. *(a) lucky three students [= (4)] 426 Takafumi Maekawa Phrases of the mod-num-ph type require a determiner because constraint (35) specifies that its MKG value is incomplete. Second, as we saw in (27) the modifier is also obligatory. (40) a. a *(beautiful) two weeks b. a *(lucky) three students [= (27)] This fact can also be accommodated by the constraint on the mod-num-ph type given in (35). This constraint specifies that the MKG value of the non-head daughter is non-numeral. Without an adjective, the non-head daughter of the modifier-numeral phrases in (40) would be the indefinite article. However, the indefinite article whose partial description was given in (36), does not conform to this combination. The description of the indefinite article is repeated in (41). (41) This shows that the MKG value of a(n) is marked. This means that it cannot fit in as a non-head daughter of numeral-modifier phrases. Note also that structures like *a two weeks and *a three students are not licensed as a hd-funct-ph type either. As the description in (41) states, a(n) selects a nominal whose MKG value is non-numeral. This means that the indefinite article cannot combine directly with two weeks or three students as in (40), whose MKG value is numeral. The ungrammaticality of the examples in (3) can be accounted for in the same way. The relevant data are repeated below. (42) a. (*a) two beautiful weeks b. (*a) three lucky students [= (3)] These examples show that NMNCs, unlike DMNNCs, do not require a determiner and exclude an indefinite article. As discussed in section 3, the MKG value of An HPSG analysis of ‘a beautiful two weeks’ 427 NMNCs is numeral. Therefore the indefinite article cannot select them. Third, this analysis can capture the fact that the indefinite article can occur with the plural head noun. In our analysis the indefinite article and the noun do not have to agree in number because the indefinite article is required by the phrase (mod-num-ph), not its head noun. A plural noun occurs simply because the numeral is larger than 1: if the numeral is one then a singular noun appears, as illustrated by the examples in (8). The relevant parts of the data are repeated here for convenience. (43) a. a career-low one touchdown in eight games (COCA) b. a further one point (BYU-BNC) In these examples the head noun needs to be singular because the numeral is one. Fourth, the constraint on mod-num-ph stated in (35) only licenses the combination of the adjective and the numeral-noun combination. Therefore, cases like (5), where the adjective intervenes between the numeral and the noun, are not licensed. The data are repeated below. (44) a. *a two beautiful weeks b. *a three lucky students [= (5)] Note also that the constraint for hd-funct-ph does not license this structure either, as stated above, Fifth, it is not difficult for this analysis to capture the instances of DMNNCs with a relative clause, observed in (11). (11a) is repeated here for convenience. (45) By the end of the four days, my group and I were ready to leave, but it was [a four days [that we will all look back on with great memories]]. [= (11a)] The order between the modifier and its head daughter is underspecified in the constraints of both hd-indep-ph in (34) and mod-num-ph in (35). Given the general constraint on the constituent order in English, the adjective modifier occurs pre-nominally while the relative clause appears post-nominally. Finally, it is possible to accommodate the DMNNCs with determiners other than 428 Takafumi Maekawa a, as in (14). (46) a. these lucky 30 people b. those busy nine days [= (14)] The constraint for mod-num-ph in (35) does not specify the AGR value. As stated above, AGR represents morphosyntactic properties of the expression and NP-internal agreement is based on the AGR value. The underspecification of the AGR value in constraint (35) enables DMNNCs to have either singular or plural agreement with the determiner. The following data show that DMNNCs can take all sorts of determiners, such as the, this, that, each, every and another. (47) a. As with all TV, there were scores of people across a number of departments who have participated in one way or another, both in the lead up to filming and during the busy two days the team were with us. (http://blog.ordnancesurvey.co.uk/2013/07/ordnance-survey-features- on-cbbcs-blue-peter/) b. Now these rare images, this scant 800 feet of film, rushed through the projector. (COCA) c. Forget that the budget for that extra fifty copies used to be spent on lower-profile movies. (COCA) d. At altitudes of more than 1,000 feet, add 1 minute to processing time for each additional 1,000 feet of altitude. (COCA) e. If your army includes sixteen or more Snotling bases then the possible number of units goes up by one for every extra five bases. (BYU-BNC) f. Vampire World 2: The Last Aerie (Roc original (Penguin), 5.99, 26th) Another staggering 784 pages of nerve-shattering horror from the author of the Necroscope series. (BYU-BNC) We have now provided an account of the DMNNC data which gets all the facts observed in the earlier sections right. An HPSG analysis of ‘a beautiful two weeks’ 429 5. Further data There are some further relevant data that should be considered. First, (48) illustrate that not only a determiner and a modifier, but a numeral is also obligatory in DMNNCs. (48) a. a beautiful *(two) weeks b. a lucky *(three) students The constraint on the mod-num-ph type given in (35) and repeated below states that in this type of phrase the head daughter is a nominal with the [MKG numeral] specification. (49) Without a numeral the head daughters of the modifiers, namely weeks and students in (48), do not have the numeral value for MKG. Moreover, the constraint for hd-funct-ph does not license this structure either. If beautiful and weeks and lucky and students formed a head-functor phrase, the MKG value of the head nouns (weeks and students) would be inherited to the mother node via the adjective (See (22)). Therefore, the AGR|NUM value, which we assume is a part of the MKG value, of beautiful weeks and lucky students would be pl, which is not compatible with the SEL value of the indefinite article (See (36)). Second, when a DMNNC is a subject, the verb can show either singular or plural agreement. In (50a) an estimated 3.3 million people has plural agreement with the verb whereas an estimated 43,000 people in (50b) shows singular agreement. (50) a. An estimated 3.3 million people have died as a result of the war making it the “tragedy of modern times”, according to a report issued by the International Rescue Committee aid agency. (http://news.bbc.co.uk/2/hi/africa/2928127.stm) 430 Takafumi Maekawa b. An estimated 43,000 people has already died since 1979 due to asbestos exposure and thousands more continue every year despite serious efforts from local and federal government to ban and or/limit the use of asbestos. (http://www.asbestos-attorney-center.com/asbestos-attorney-cancer-m esothelioma) Plural agreement in (50a) occurs simply because the head of an estimated 3.3 million people is a plural noun people. Following much HPSG work we assume that subject-verb agreement in number is dependent on the information encoded in the value of the CONTENT (CONT) feature of the subject (e.g., Kathol 1999, Kim 2004, Wechsler and Zlatić 2000). The CONT feature indicates what kind of semantic properties the linguistic expression has. The CONT value includes the INDEX feature, which represents what it refers to in the actual world. The INDEX value includes the NUMBER (NUM) feature, which indicates whether it is singular or plural in meaning. The partial description of people in (50a) is given in (51). (51) The CONT|INDEX|NUM value of the head noun people is pl (plural), so an estimated 3.3 million people is semantically plural and shows plural agreement with the verb. What about singular agreement in (50b), then? Let us observe the fact that in English plural nouns sometimes show singular agreement with the verb. (52) a. Five pounds is/*are a lot of money. (Hudson 1999: 174) b. Most of us can agree that 8 million people is too many to be receiving disability payments from the government. (http://www.startribune.com/printarticle/?id=177023831) An HPSG analysis of ‘a beautiful two weeks’ 431 In (52) five pounds and 8 million people refer to a group of people conceived as a whole rather than discrete entities (Kim 2003: 1117-1118). Let us assume that pounds and people in (52) have the following lexical description. (53) These nouns are normally countable nouns which are both morphologically and semantically plural, but in (52) they are morphologically plural but semantically singular. In (53) the MKG|AGR|NUM value indicates that the word is morphologically plural, while the CONT|INDEX|NUM value is singular (sg), indicating that it is semantically singular. Likewise, we can propose that the head noun of an estimated 43,000 people in (50b) is morphologically plural but semantically singular, and its semantic singularity causes singular agreement with the verb. In English even singular words such as family, team, and government can have either singular or plural agreement with the verbs, depending on the context. Verbs take a plural form when the group is seen as a collection of people and take a singular form when the group is seen as a unit (Swan 2005: 526). Therefore, we can conclude that the facts observed in (50) are nothing special: they just conform to the general patterns of subject-verb agreement in English. 6. Conclusion Let us summarise the discussion. In this article we have first looked at some analyses which have been proposed for DMNNCs and argued that they are not successful. We have then developed a fairly detailed analysis within HPSG in which the peculiarities of DMNNCs are attributed to the special constructional constraints. We have argued that our HPSG analysis can provide a satisfactory account of the properties of DMNNCs. The use of hierarchically organised network of phrasal types in (33) allows us to 432 Takafumi Maekawa have constraints of any level of generality. Our HPSG analysis accommodates not just the construction-specific properties of DMNNCs but also the regularities that they share with other constructions, such as NMNCs such as two beautiful weeks and the Big Mess construction such as so big a mess. This approach can thus capture the distinctive properties of DMNNCs without missing any generalisations. References Allegranza, V. 1998. Determiners as functors: NP structure in Italian. In S. Balari and L. Dini (eds.), Romance in HPSG, 55-108. Stanford: CSLI Publications. Davies, M. 2004. BYU-BNC. (Based on the British National Corpus from Oxford University Press). http://corpus.byu.edu/bnc/. Davies, M. 2008. The Corpus of Contemporary American English: 450 million words, 1990-present. http://corpus.byu.edu/coca/. Ellsworth, M., R. Lee-Goldman, and R. Rhodes. 2008. A paradox of English determination: the construction of complex number. Proceedings of the 38th Western Conference on Linguistics. Gawron, J. M. 2002. Two kinds of quantizers in DP. Handout given at the Linguistics Society of America. Honda, M. 1984. On the syntax and semantics of numerals in English. Journal of Osaka Jogakuin 2-year College 14: 97-115. Hudson, R. 1999. Subject-verb agreement in English. English Language and Linguistics 3: 173-207. Ionin, T. and O. Matushansky. 2004. A singular plural. Proceedings of WCCFL 23, 399-412. Ithaca, NY: Cascadilla Press. Ionin, T. and O. Matushansky. 2006. The composition of complex cardinals. Journal of Semantics 23: 315-360. Jackendoff, R. 1977. X-bar Syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press. Kathol, A. 1999. Agreement and the syntax-morphology interface in HPSG. In R. Levine and G. Green (eds.), Studies in Contemporary Phrase Structure Grammar, 223-274. Cambridge: Cambridge University Press. Kim, Jong-Bok. 2004. Hybrid agreement in English. Linguistics 42: 1105-1128. Ohna, T. 2003. A beautiful two weeks: its syntactic structure and the semantic relations of the adjective to the numeral and head noun. In S. Chiba, et al. (eds.), 23-40. Empirical and Theoretical Investigations into Language. Pollard, C. and Ivan A. Sag. 1989. Information-Based Syntax and Semantics. Stanford: CSLI An HPSG analysis of ‘a beautiful two weeks’ 433 Publications. Pollard, C. and Ivan A. Sag. 1994. Head-driven Phrase Structure Grammar. Chicago: University of Chicago Press. Sag, I. A. 1997. English relative clause constructions. Journal of Linguistics 33: 431-484. Swan, M. 2005. Practical English Usage. Oxford: Oxford University Press. Van Eynde, F. 2006. NP-internal agreement and the structure of the noun phrase. Journal of Linguistics 42: 139-186. Van Eynde, F. 2007. The Big Mess Construction. In S. Müller (ed.), Proceedings of the HPSG07 Conference, 415-433. Stanford: CSLI Publications. Wechsler, S. and L. Zlatić. 2000. A theory of agreement and its application to Serb-Croatian. Language 76: 799-832. Takafumi Maekawa Department of Clinical Social Welfare Ryukoku University E-mail: maekawa@soc.ryukoku.ac.jp Received: 2013. 07. 23 Revised: 2013. 11. 27 Accepted: 2013. 11. 27