<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-1041">
  <Title>GENERATION OF ACCENT IN NOMINALLY PREMODIFIED NOUN PHRASES</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
GENERATION OF ACCENT IN
NOMINALLY PREMODIFIED NOUN PHRASES
RON ZACHARSK1
</SectionTitle>
    <Paragraph position="0"> Tile primary purpose of this paper is to present a set of conditions that constrain accent placement in focused nominally ptemodified NPs. Selkirk (1984) argues that if the premodifier is an argument of the head, then the head can be deaccented.</Paragraph>
    <Paragraph position="1"> I agree with Selkirk's proposal and argue that what is essential is not whether the premodifier is a grammatical argument of the head noun, but rather, whether it is a 0-complement in lexical conceptual structure. This proposal is evaluated by testing it against a corpus of naturally occurring data.</Paragraph>
    <Paragraph position="2"> O. introduction It is generally agreed that an utterance can be divided into two parts which are related to the discourse function of the information represented by that utterance. The TOPIC is what the sentence is about and the FOCUS represents a new predication about the topic. This information structure constrains accent placement. For example, 'primary' accent must be within the constituent that represents the focused information. The ability of an accent on a single word to mark a larger phrase as focus is widely recognized. For example in (1) the accent on conservative can mark the phrase redneck conservative as the focused constituent (since (1) can be used to answer the question What was your town like? ). (Accent is indicated by small caps and focus by underlining.) 11 would especially like to thank Jeanette Gundel and Nancy Hedberg for discussions and comments.</Paragraph>
    <Paragraph position="3">  (1) My hometown was redneck CObISERY_ATI~. null (Lia Matera 1988 Smart Money) in (1) focus is represented by a single constituent, but this need not be the case as (2) illustrates. (2) A: Where's Karl?  B: ~ her ~.</Paragraph>
    <Paragraph position="4"> In (2B), Karl, the referent of her, is the topic and tire focus is thus discontinuous.</Paragraph>
    <Paragraph position="5"> Though there seems to be agreement about the importance of a theory that accounts for the accent-focus relation, there is little agreement about the exact nature of this relationship. For instance, different theories give different answers to the question of whether this relationship is syntactic, semantic, morphological, or pragmatic. There is also disagreement over how large a phrase can be brought into focus by a single accent.</Paragraph>
    <Paragraph position="6"> Even for simple constructions the relationship between accent and focus is unclear. In exam~ pies (3)-(7), a MODIFIER + NOUN constituent is focused. Note that in the (a) member of each pair, it is the modifer that receives the accent and in  the (b) member of the pair, it is the head noun. (3) a. Those are CRAWLING things.</Paragraph>
    <Paragraph position="7"> b. Those are crawling INSECTS.</Paragraph>
    <Paragraph position="8"> (Bolinger 1986.120) (4) a. He has HUNTINGTON's disease.</Paragraph>
    <Paragraph position="9"> b, He has Huntington's CHOREA.</Paragraph>
    <Paragraph position="10"> (Bolinger 1986.118) (5) a. 1~ do university RESEARCH.</Paragraph>
    <Paragraph position="11"> b. I do CETACEAN research.</Paragraph>
    <Paragraph position="12"> (6) a. (How do you know Fran72 What was  he to you?) He was my HISTORY teacher.</Paragraph>
    <Paragraph position="13"> AuIa/s DE COLING-92, NANTES, 23-28 AO1Yr 1992 2 5 3 I'ROC. OF COLING-92, NANTES, AUG. 23-28, 1992 b. (How do you know Fran72 What was he to you?) He was my first-grade TEACHER.</Paragraph>
    <Paragraph position="14"> (7) a. I work for the ROCKEFELLER Foundation.</Paragraph>
    <Paragraph position="15"> b. I work for the Carnegie ENDOWMENT.</Paragraph>
    <Paragraph position="16"> (Bolinger 1986.118) ~'his paper examines the association of accent and focus in nominally premodified NPs. 2 1.0 previous work.</Paragraph>
    <Paragraph position="17"> Pierrehumbert and Hirschberg (1990) state that some cases of the relationship between accent and focus seem dear. For example, when accent is on dress in the glrl with the red DRESS the focus can be information represented by the noun dress, the NP the red dress, the PP with the red dross, or the entire NP the glrl with the red dress. Whereas, accenting RED in the same phrase can serve only to focus information represented by the adjective red. However, regarding the relationship between accent and focus in general, they state that q'he question of how an accent becomes associated with certain material is not yet well understood.' (p309 n.4) Some researchers view this rdationship between focus and accent as essentially syntactic. In the computational literature, Lyons and Hirst (1990) present the following rule constraining the accent focus relation: (8) a. What is accented is necessarily in focus. null b. 'focus is optionally and nondererministically percolated up the syntax tree, to any node from its rightmost daughter (rightmost because stress manifests itself only at the end of the focused constituent).' (1990.57) Many theoretical linguists have proposed similar syntactic constraints relating phrasal accent  placement and focus (see, for example, Chomsky and Halle 1968, Chomsky 1971, Jackendoff 1972, and Culicover and Rochemont 1983). However, as noted by these researchers, a rule like the one in (8b) makes incorrect predications for most nominally premodified tips. For example, consider 21 use the more neutral term nominally premodified NPs imtead of mmp~n~.</Paragraph>
    <Paragraph position="18"> the NP the histor? teacher in (9B) which has the structure presented in (10): (9) A: What do you do? B: I'm a HISTORY teacher.</Paragraph>
    <Paragraph position="19"> (lO)</Paragraph>
    <Paragraph position="21"> Clearly the NP a h/nor? teacher is the focused constituent. 3 But by rule (8) the accent on h/nor? cannot serve to focus the entire NP since hirtor? is not the rightmust daughter of this phrase. 4 Thus, a phrasal rule such as (8b) cannot be operative in these cas~. Instead, history teacher is viewed as a compound (as a structure of category N 0) and accent is determined by a separate compound accent rule, which places stress on the leftmust element (see Chomsky and Halle 1968, and Selkirk 1984).</Paragraph>
    <Paragraph position="22"> Steedman (1991) claims that accent serves to divide an utterance into an optional constituent that represents the topic (what he, following Jackendoff 1972, calls the 'open-proposition') and a constituent that represents the focus. He argues that within each of these constituents, accent is put on the parr that represents what is 'emphasized or contrasted' with something in the discourse context--the 'interesting part'. This idea has also been proposed by a number of other researchers (see, for example, Schmerling 1976, Gundel 1978, Selkirk 1984, Bolinger 1986, 1989, and Rochemont 1986). Although there is wide disagreement about the formal definitions of 'topicfocus' and 'the interesting part' there is no doubt that these are essential pragmatic determinants that constrain accent placement, and it is equally 5By focused constituent I mean the phr~,e that represents focused information. 4Rule (7) also fails in a range of other cases including most intransitive aen~nces ( Whau h~ppenedI--BUSH resigned ). See Schmeding 1976, Gundel 1978, Bolinger 1986, and Lambr~ht 1992 for further discmalon.</Paragraph>
    <Paragraph position="23"> ACRES DE COLING-92, N^h'T~, 23-28 ^OfT 1992 2 5 4 PROC. OF COLING-92, NANTES, AUO. 23-28, 1992 clear that they play a major role determining accent in nominally premodified NPs. Consider {11), a dialogue between two linguistics professors  about the dissertation of a student in the department: null (11) A: What's the dissertation about? B: Something to do with language ACQUISITION.</Paragraph>
    <Paragraph position="25"> Ladd (1980) notes that when linguistics professors are talking about a dissertation of a student in the department, they assume that the dissertation is about some aspect of language and language can be deaccented. Ladd continues 'But the linguist's proud parents, who are not linguists, would tell their friends, who are also not linguists, that their son's dissertation was about LANGUAGE acquisition.' (Ladd 1980.90) It could be argued that the accent patterns in (3)-(5) are also determined by 'interestingness'. However, the role of 'interestingness' is less clear in phrases like those in (6) and (7). As Bolinger (1989.200) notes, accent placement in some prenominally modified NPs is more invariable than in others. For example, while in (12) the old information constrains the the location of accent, this is not the case in (13) and (14)  (12) a. My research is on language- { Particularly on language ACQUISITION.~ #Particularly on LANGUAGE acquisition.J b. My research is on acquisition#Particularly on language ACQUISITION.1 Particularly on LANGUAGE acquisition. J (13) June hold a singular place in the study of insects.</Paragraph>
    <Paragraph position="26"> JUNE bugs can be seen.</Paragraph>
    <Paragraph position="27"> ??June BUGS call be seen.J (14) As we were travelling along the road we I&amp;quot; ROADblock.</Paragraph>
    <Paragraph position="28"> came to a l#roadBLOCg.~ (= Bolinger 1989.216)  The fixed stress on the initial syllables serves to mark the phrase as a unitary concept rather than a compositional one. However, not all NPs representing unitary concepts receive leftward stress (for example, human Being}.</Paragraph>
    <Paragraph position="29"> 2. argument .trucmre and accent What then governs the location of accent in nominally premodified NPs? Why, for example, does He was my history TEACHER sound odd as an answer to the question How do you know Franz? What was he to you? Selkirk (1984) suggests a particularly compelling answer. She presents the two constraints in (15) and (16).</Paragraph>
    <Paragraph position="30">  (15) Basic Focm Rule A constituent to which a pitch accent is assigned is a focus. (1984.207) (16) Phrasal Focus Rule  A constituent may be in focus if (i) or (ii) (or both) is true: i. The constituent that is its head is a fOg~us.</Paragraph>
    <Paragraph position="31"> ii. A constituent contained within it that is an argument of the head is a</Paragraph>
    <Paragraph position="33"> The Basic Focus Rule in (15) is mandatory. This rule states that every word that contains an accent is necessarily focused. The phrasal focus rule given in (16) is optional. The rule (16i) states that if a head is a focused constituent, then any projection of that head can optionally be focused. Ifl say (17) Victoria visits BONEyards the N, boneyards, is necessarily a focused constituent by rule (15). Rule (16i) permits focus to percolate to the projections N' and N&amp;quot; as shown in (18):  focused, and if that focused constituent is contained within the maximal projection of that head, then the projection of that head can be focused. For example, if I say (19) Ann danced the taranTELla the N, tarantella, is necessarily a focused constituent. Since this N is the head of the N&amp;quot;, the tarantella, by rule (16i) the tarantella can be a focused contituent. This N&amp;quot; is an argument of the V, dancea~ and is contained within the maximal AcrES OF. COLING-92, NANTES, 2.3-28 AO~'r 1992 2 5 5 PROC. OF COLING-92, NANTES, AuG. 23-28, 1992 projection of that V. Thus, the VI' can be a focused constituent by rule (16ii). This leads to correct results since this utterance can be used to respond to the question What did Ann do at the party? (She \[V&amp;quot;:+focus danced the taranTELla.\]) Selkirk's theory of prosody-focus relation predicts the following accent patterns in focused  modifier-noun constructions: (20) Selkirk's predictions concerning raoeh'fier-noun pairs</Paragraph>
  </Section>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
ARGUMENT head adjtmct HEAD
ARGUMENT HEAD ADJUNCT HEAD
*argument HEAD *ADJUNCT head
</SectionTitle>
    <Paragraph position="0"> Thus, if a noun phrase represents focused information, and if that noun phrase consists of an argument followed by the head noun, then the prediction is that the argument is necessarily accented. This seems like an elegant way to characterize the difference between leftward and rightward accented prenominaUy modified noun phrases, which has been problematic for other approaches to accent placement. Consider, for example, the difference in accent pattern between  (21) which has an accent on its left constituent, and (22) which has the accent on its right constituent. (21) HISTORY teacher (22) student TEACHER  Selkirk's theory offers a straightforward explanation of this difference. History is an argument of teacher in (18) and thus by rule (13ii) representation of the entire noun phrase can be focused. However, in (19) student is an adjunct and accenting the head, teacher, is required to focus the representation of that noun phrase. A similar notion has been suggested by Rochemont (1986) as shown in (23). (See Rochemont 1986 for a discussion comparing his focus rules to those of Selkirk 1984.) (23) a. If0~ is \[+focus\] and (x is X 0, then X n is \[+focus\].</Paragraph>
    <Paragraph position="1"> b. If o~ is \[+focus\] and O~ is an argument of X 0contained in X n, then X degis \[+focus\].</Paragraph>
    <Paragraph position="2"> c. If X 0 is \[+focus\] and Ct is an adjunct ofX deg then O~ is \[+focus\].</Paragraph>
    <Paragraph position="3"> There are three possible relations between a prenominal NP and its head. The prenominal can be a grammatical argument, a complement in lexical conceptual structure, or an adjunct modifier. A distinguishing characteristic of an adjunct modifier is that it is licensed by predication.5 As a result, it can be separated from its head by a copula. For example, in the phrase the red car, red is an adjunct modifier and can be separated from its head as (24) illustrates.</Paragraph>
    <Paragraph position="4"> (24) The car is red.</Paragraph>
    <Paragraph position="5"> In (22) student is a modifier and can be separated from its head as in (25) The teacher is a student.</Paragraph>
    <Paragraph position="6"> However, in (21) history is not a modifier and it cannot be separated from its head as (24) shows. (26) ??The teacher is of history.</Paragraph>
    <Paragraph position="7"> There is one question that immediately comes to mind: is Sdkirk's notion of 'argument' a syntactic notion (that is, is it a grammatical argument licensed by A-structure), or is it a semantic notion involving 0-participants in lcs?</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2. arguments
</SectionTitle>
    <Paragraph position="0"> Grimshaw (1990) argues convincingly that only nouns that have an internal aspeetual analysis (nouns that refer to what she calls complex events) have argument structure (A-structure). She describes significant differences in the behavior of complex event nouns and other nouns to support this analysis. For example, complex event nominals have obligatory arguments as shown in (27)  (27) a The assignment is to be avoided.</Paragraph>
    <Paragraph position="1"> b. *The constant assignment is to be avoided.</Paragraph>
    <Paragraph position="2"> c. The constant assignment of unsolvable problems is to be avoided.</Paragraph>
    <Paragraph position="4"> Grimshaw considers assignment to be ambiguous between a complex event interpretation and a process interpretation. The addition of the modifier constant forces the complex event interpretation since constant can only be construed as a modifier of assignment on the complex event reading. Thus, it's A-structure must be satisfied as in (27c) just as  5That is, the meaning of modifier is predicated on the external argument of the head noun (Its Rargumen0. Aca-~ DE COLING-92, NANTEs, 23-28 ^Ot3T 1992 2 5 6 PROC. OF COLING-92, NANTES, AUG. 23-28. 1992 the A-structure of the verb assign must be satisfied as in (28).</Paragraph>
    <Paragraph position="5"> (28) a We constantly assign problems.</Paragraph>
    <Paragraph position="6"> c. *We constantly assign.</Paragraph>
    <Paragraph position="8"> If Grimshaw's analysis is correct, as I believe it is, then the argument referenced in Selkirk's Phrasal Focus Rule, is best not construed as an argument in A-structure. For while rule (16) would make correct predictions regarding complex event nominals such as those illustrated in (29), it fails on nominals that do not represent complex events as in (30)  (29) a. TREE felling b. COOKIE baking (30) a. HISTORY teacher b. BIT guzzler 1 believe the distinction is a semantic one involving 0-participants in lexical conceptual structure  (Ics). Every verb and noun (including deverbal nouns) has a lexical conceptual structure that includes the entities involved in the events or states described (see, for example, Dowty (1989), Fillmore (1968), and Jackendoff (1987, 1990)). Selkirk's intuitions expressed in rules (15) and (16) are essentially correct. Reformulating her rules as constraints between lexical conceptual structure and focused information offers a more precise characterization of her insights. A reformulation of (15) and (16) constraining the accent-focus relation of premodified NPs containing a de- null verbal element is given in (31) and (32): (31) The representation of a constituent is focused if that constituent receives a pitch accent.</Paragraph>
    <Paragraph position="9"> (32) A representation, R, may be focused if a representation that is a 0-complement of R is focused.</Paragraph>
    <Paragraph position="10"> Consider the pairs in (33) and (34).</Paragraph>
    <Paragraph position="11"> (33) a. PACKAGE delivery b. overnight DELIVERY (34) a. CETACEAN research h. university RESEARCH  According to Grimshaw, neither delivery nor research has an A-structure since neither has an internal aspectual analysis. However, since these are both deverbal nouns, they inherit their O-structure from the related verbs. Package, then, is a 0-complement of delivery, since package is the theme of deliver, (He delivers packages). Thus delivery can be deaccented as in (33a). Itowever, itt (33b), overnight is not a O-complement of delivery (*tie delivers overnight) and thus delivery must be accented. Likewise, in (34a) cetacean is the theme of research (He researches cetaceans) and thus research can be deaccented. However, in (34b) university is not a 0-complement of research and research must be accented.</Paragraph>
    <Paragraph position="12"> The rules presented in (31) and (32) were tested on data collected from multiple genres of natural discourse induding public radio news articles, nruhiple participant discussions, and academic lectures. The results are given in Table 1 modifier's relation to laead</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
NOUN MODIFIER + DEVERBAL NOUN
</SectionTitle>
    <Paragraph position="0"> consD'ucrions As shown in this table, the rule makes correct predictions in approximately 90% of the cases. The rule predicts that if the modifier's relation to ht~d is a 0-complement, then the accent should be on the left element--tire complement. This was indeed true for 115 instances. Some examples are given in (35) and (36).</Paragraph>
    <Paragraph position="1">  (35) Why should I buy one of those POTATO twaddlerJ (that can make potato romettes simply and easily and comes with a free set of Ginso knives?) (36) a. FEMINIST bashing a. LANGUAGE users b. PUB crawlers c. COMPUTER makers d. TEAK trade e. CONSISTENCY checkers  AcrEs DE COLING-92, NANTES. 23-28 nO~' 1992 2 5 7 PROC. OV COLING-92, NANTES. AUO. 23-28, 1992 The rule also predicts that if the modifier is not a 0-complement of the head then the head needs to be accented. That was the case for 18 of the 20 instances in the data. Examples of this are given in  (37) and (38) (37) She and her sister P..~PER5 never stop TELLING each other to be PROUD of what they ARE, to have RESPECT for themselves and the culture they COME from.</Paragraph>
    <Paragraph position="2"> (38) a. amateur WRITER b. woman HIPHOPPERS c. continuation COVERAGE e. last-minute FILER  There were 11 0-complement instances that were counterexamples to the rule presented in  (32). Some examples of this are presented in (39). (39) a. government ENCOURAGEMENT b. systems ANALYST c. tenant BLACKLISTING d. human INTERVENTION e. relationship DEPENDENCY  At present I have no explanation as to why these are accented the way they are. Why, for example, is one problem referred to as DRUG addiction and another as relationship DEPENDENCY?. However, 1 suggest that woman in (41a) is not a complement of swimmer. Woman SWIMMER is an appositional compound (similar to helicopter GUNSHIP). Appositional compounds are lists of propositions, and like all lists, the last element of the list typically receives the main accent. Some evidence to support this view that (41a) is appositional is that the prenominal can be separated from the head by a copula as in (42) (42) The swimmer is a woman.</Paragraph>
    <Paragraph position="3"> Following Booij and van Haffxen (1988) I bdieve that a semantic effect of -er afFzxation is to bind the agent or experiencer in lcs. Thus, the agent role is not available to the representation of woman and that representation must be linked by 0-identification. Other examples of this type include (36) and (37a&amp;b). In cases were the nonhead is the agent of the head the accent pattern varies--sometimes the head receives the accent (as (390, (41b), and (41c)) and at other times the nonhead receives the accent as in (43). 6  (43) a. DOG bite b. BEE sting c. COCK fighting 4. discussion 3. agents and experiencers  The accent characteristics of phrases where the nominal premodifier can be construed as a subject of the head is less dear. There is some controversy as to whether such constructions are possible. For example, Selkirk (1982.34) restricts subjects from occurring in these compounds by use of the rule presented in (40) (40) The SUBJ argument of a lexical head may not be satisfied in compound structure..</Paragraph>
    <Paragraph position="4"> Sproat and Liberman (1987) point out that subjects in compounds are not usually accented. The examples they give are presented in (41) (41) a. woman SWIMMER b. child DANCING c. student DEMONSTRATION Sproat and Liberman 1987.143 Deverbal nouns head 25% of the nominally premodified Nps in the corpus examined. 7 Since deverbal nouns are distinguished from other nouns in the lexicon, the generation system can correctly determine when to apply the rule in (32). As Dowry (1989), Jackendoff (1987, 1990) and others have noted, lexical conceptual structure is needed for correct semantic interpretation. (For computational approaches see Charniak (1981), Dorr (1989), and Sowa (1991).) Thus, the rule requires only information that has independent motivation for being in the lexicon.</Paragraph>
    <Paragraph position="5"> As Selkirk noted, the same factors that govern accent placement in these constructions also constrain accent in verb phrases and sentences. In both cases the semantic interpretation of a head can be focused if its 0-complement is focused. 6This variation in accent is ~ seen in agents realized u verb. They dn nnt nePSeum'ily have A-structure (other than the external role for nouns).</Paragraph>
    <Paragraph position="6"> ACRES DE COLING-92, NANTES, 23-28 AO~r 1992 2 5 8 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 The same rules ((31) and (32)) are operative. Thus accenting potatoes in John twaddles POTATOES can serve to focus the representation (x twaddles potatoes) and similarly, accenting potato in POTATO twaddler focuses the representation (the x such that (x twaddles potatoes)).</Paragraph>
    <Paragraph position="7"> A central question in the study of intonation is what factors govern accent placement. I have argued here that argument structure plays no role in this determination at least as to prenominally modified noun phrases and have shown how a theory of focus like the one presented in Selkirk 1984 can be refined to account for semantic constraints for accent placement.</Paragraph>
  </Section>
class="xml-element"></Paper>