Glue and information structure Mary Dalrymple, University of Oxford LFG Meeting, SOAS, 6 Feb 2010
1
Overview
Information structure, and early work in LFG
Glue and the representation of information structure: based on Chapter 4 of Objects and information structure, Mary Dalrymple and Irina Nikolaeva, Cambridge University Press, to appear (soon!). Glue and information structure – 2 / 53
Information structure Information structure is the level of sentence organisation which represents how the speaker structures the utterance in context in order to facilitate information exchange. Specifically, it indicates how the propositional content of an utterance fits the addressee’s state of knowledge at the time of utterance. (Dalrymple & Nikolaeva, 2010, Chapter 3) Glue and information structure – 3 / 53
Content of information structure topic:
the entity or entities that the proposition is about (Reinhart, 1982; Gundel, 1988; Lambrecht, 1994). An entity, E, is the topic of a sentence, S, iff in using S the speaker intends to increase the addressee’s knowledge about, request information about, or otherwise get the addressee to act with respect to E. (Gundel, 1988)
focus:
informative, newsy, contrary-to-expectation part of the sentence (Vallduv´ı, 1992; Vallduv´ı & Engdahl, 1996); the semantic component of a pragmatically structured proposition whereby the assertion differs from the presupposition (Lambrecht, 1994) Glue and information structure – 4 / 53
Content of information structure background:
old/presupposed information specifying detailed knowledge that may be necessary for a complete understanding of new (focused) information — different from topic, which is a pointer to the relevant information to be accessed by the addressee (Butt & King, 2000)
completive:
new to the addressee but, unlike focus, not associated with the difference between pragmatic assertion and pragmatic presupposition (Butt & King, 2000)
What is Bill eating? He is eating topic
background
pizza
in the kitchen.
focus
completive
Glue and information structure – 5 / 53 2
Signalling information structure relations In an LFG setting:
Agreement and casemarking as signals of topicality: Dalrymple & Nikolaeva (2010)
Word order as a signal of information structure role: Choi (1999) for Korean and German, Butt & King (1996, 2000) for Hindi-Urdu
Prosody as a signal of information structure role: Mycock (2006)
These signals may partially or completely specify information structure role, and must be consistent with the context of utterance. Glue and information structure – 6 / 53
Prosodic encoding/ambiguity
Argument-focus structure: I heard your motorcycle broke down? My car broke down. (Lambrecht, 1994, 223) focus
background
Sentence-focus structure: What happened? My car broke down. (Lambrecht, 1994, 223)
focus
Inappropriate prosody: What happened to your car? *My car broke down. topic
focus
Glue and information structure – 7 / 53
Early work in LFG Prosodically focused final constituent: prislal muˇz den’gi sent husband money ‘My husband sent (me) the money.’ (King, 1995) : ⎡
focus
pred
‘money’
⎤
⎢ ⎥ ⎢ pred ‘sendsubj,obj’ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ subj pred ‘husband’ ⎥ ⎢ ⎥ ⎣ ⎦ obj
Glue and information structure – 8 / 53 3
Early work in LFG Choi (1999):
+new, +prominent
Butt & King (2000) topic is [−new] and [+prominent], focus is [+new] and [+prominent], completive information is [+new] and [−prominent], and background information is [−new] and [−prominent]. Glue and information structure – 9 / 53
Early work in LFG Hans das Buch dem Sch¨ uler gegeben hat Hans the book.Acc the student.Dat given has ‘Hans gave the student the book.’ F-structure (Choi, 1999): ⎡ pred ⎢ ⎢ ⎢ ⎢ ⎢ subj ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ i.obj ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ d.obj
‘gebensubj, d.obj, i.obj’
⎡
⎤ pred ‘Hans’ ⎣ new − ⎦ prom + ⎡ ⎤ pred ‘student’ ⎣ new − ⎦ prom − ⎡ ⎤ pred ‘book’ ⎣ new + ⎦ prom −
⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦
Glue and information structure – 10 / 53
Problems Goes against LFG view of modularity: do not represent very different kinds of information in the same structure. Glue and information structure – 11 / 53
Problems King (1997): f-structure constituents inappropriate for representation of information structure. ˇitala knigu. Ona proc she read book ‘She read the book.’ ⎡
pred
‘readsubj, obj’
⎢ ⎢ focus ⎢ ⎢
⎢ pred ‘she’ ⎢ subj ⎣
pred ‘book’ obj
⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦
Focusing the verb seems (incorrectly) to focus its arguments as well: this is the granularity problem. Glue and information structure – 12 / 53 4
Information structure as a separate structure King (1997): Information structure is separate from f-structure. [Was it the ex-convict with the red shirt that he was warned to look out for?] No, it was an ex-convict with a red [tie] that he was warned to look out for. (King 1997, 8, citing Jackendoff 1972, 232) F-structure: ⎤
⎡
pred
‘ex-convict’ ⎧⎡ ⎢ ⎢ pred ‘withobj’ ⎪ ⎪ ⎢ ⎨⎢ ⎢ ⎢ ⎢ adjunct pred ‘tie’ ⎢ ⎣ obj ⎪ ⎪ ⎣ ⎩ adjunct pred
⎤⎫ ⎥ ⎪⎥ ⎬⎥ ⎥⎪ ⎥ ⎥ ⎥ ⎥
⎦⎪ ⎪ ⎭⎦ ‘red’
I-structure: ⎡
foc
⎢ ⎢ ⎢ bck ⎣
⎤ tie ⎧ ⎫⎥ ⎨ ex-convict ⎬ ⎥ ⎥ with ⎩ ⎭⎦ red
Glue and information structure – 13 / 53
Information structure and glue Mycock (2009): the traditional semantic structure of LFG, an important component of the “glue” approach to the syntax-semantics interface (Dalrymple, 1999, 2001; Asudeh, 2004), plays an important role in representing information structure relations. Glue and information structure – 14 / 53
Glue and the syntax-semantics interface John married Rosa. IP NP
I
N
VP
⎡
⎤ ‘marrysubj,obj’ ⎢
⎥ ⎢ subj ⎥ pred ‘John’ ⎢ ⎥ ⎣ ⎦
pred ‘Rosa’ obj pred
V
N
John
V
NP
married
N N
Rosa marry (john, rosa) Glue and information structure – 15 / 53 5
Meaning contributions
⎤ ‘marrysubj,obj’ ⎥ ⎢ ⎢ subj pred ‘John’ ⎥ ⎥ ⎢ ⎦ ⎣
pred ‘Rosa’ obj ⎡
pred
The word John contributes the meaning john.
The word Rosa contributes the meaning rosa.
The word married contributes meaning assembly instructions of the following form: When given a meaning x for my subject and a meaning y for my object, I produce a meaning marry (x, y ) for my sentence. Glue and information structure – 16 / 53
Contribution of ‘John’ NP N
john:[ ]
pred
‘John’
N
John
Every f-structure has a corresponding semantic structure, related to it by the projection function σ.
john:[ ] is a meaning constructor.
In the lexicon: john:↑σ Glue and information structure – 17 / 53
Meaning assembly: Gluing meanings together ⎡
pred
‘marrysubj,obj’
⎢ ⎢ subj [ ] ⎣ obj [ ]
⎤ ⎥ ⎥ ⎦
λx.λy .marry (x, y ):sσ −◦(oσ −◦mσ )
λx.λy .marry (x, y): a relation between two individuals x and y that holds if x marries y sσ −◦(oσ −◦mσ ): If I am provided with the semantic structure of my subject and then the semantic structure of my object, I produce the semantic structure of the sentence. In the lexicon: λx.λy .marry (x, y ):(↑ subj)σ −◦((↑ obj)σ −◦↑σ ) Glue and information structure – 18 / 53 6
Proof rules X : fσ P : fσ −◦ gσ P (X) : gσ Glue and information structure – 19 / 53
Meaning proof for ‘John married’ X : fσ P : fσ −◦ gσ P (X) : gσ john:sσ
λx.λy .marry (x, y ):sσ −◦(oσ −◦mσ ) λy .marry (john, y ):oσ −◦mσ Glue and information structure – 20 / 53
Meaning proof for ‘John married Rosa’ X : fσ P : fσ −◦ gσ P (X) : gσ john:sσ
λx.λy .marry (x, y ):sσ −◦(oσ −◦mσ ) λy .marry (john, y ):oσ −◦mσ marry (john, rosa): mσ
rosa:oσ Glue and information structure – 21 / 53
Another meaning proof for ‘John married Rosa’ X : fσ P : fσ −◦ gσ P (X) : gσ rosa:oσ
λx.λy .marry (x, y ):sσ −◦(oσ −◦mσ ) λx.marry (x, rosa):sσ −◦mσ marry (john, rosa): mσ
john:sσ Glue and information structure – 22 / 53 7
Proposal: Information structure features and glue New idea: Categorise meaning constructors according to their information structure role. John
married Rosa.
topic ⎡
focus
⎤ ‘marrysubj,obj’ ⎢
⎥ ⎢ ⎥ m :⎢ subj j : pred ‘John’ ⎥ ⎣
⎦ obj r : pred ‘Rosa’ pred
⎡
⎢
mσι ⎢ ⎣
⎤
{john:jσ }
topic focus
⎥
λx.λy .marry (x, y ):jσ −◦(rσ −◦mσ ) ⎥ ⎦ rosa:rσ Glue and information structure – 23 / 53
Abbreviations for meaning constructors John
john:jσ
married
λx.λy .marry (x, y ):jσ −◦(rσ −◦mσ )
Rosa
rosa:rσ
married-Rosa
λy .marry (x, rosa):rσ −◦mσ Glue and information structure – 24 / 53
Simpler proofs married
Rosa married-Rosa marry (john, rosa): mσ
John Glue and information structure – 25 / 53
Simpler representations IP NP
I
N
VP
N
V
John
⎡
⎤ ‘marrysubj,obj’ ⎢
⎥ ⎢ ⎥ m :⎢ subj j : pred ‘John’ ⎥ ⎣
⎦ obj r : pred ‘Rosa’ pred
V
NP
married
N N
⎡ ⎢
mσι ⎢ ⎣
⎤
topic focus
{John } ⎥ married ⎥ ⎦ Rosa
Rosa Glue and information structure – 26 / 53 8
Simpler representations married-Rosa can be derived from the premises married and Rosa: IP NP
I
N
VP
N
V
John
V
married
⎡
⎤ ‘marrysubj,obj’ ⎢
⎥ ⎢ ⎥ m :⎢ subj j : pred ‘John’ ⎥ ⎣
⎦ obj r : pred ‘Rosa’ pred
NP N
⎡
mσι ⎣
⎤
topic
{John }
focus
{married-Rosa}
⎦
N
Rosa Glue and information structure – 27 / 53
Summary A meaning constructor is assigned an information structure role by virtue of its appearance in the topic, focus, background, or completive set. Meaning constructors contributed by the various parts of an utterance are categorised according to their information structure contribution, and appear in the relevant information role category. Glue and information structure – 28 / 53
How can we achieve this?
Each meaning constructor is required to take on some information structure role (i.e., to appear in some set at information structure).
Which role it takes on can be specified in various ways: agreement, casemarking, word order, intonation, ...
Solution: Specify the information structure role of a meaning constructor as the value of the attribute df in its semantic structure. This allows for specification of information structure roles by various modules of the grammar.
We can then use the value of that attribute to assign the appropriate role at information structure. Glue and information structure – 29 / 53 9
Our architecture c-structure φ f-structure σ s-structure ι i-structure Glue and information structure – 30 / 53
Revised lexical entries Each meaning contribution must bear some information structure role, specified as a feature at semantic structure: John
(↑ pred) = ‘John’ john ∈ (↑σι (↑σ df))
N
The meaning constructor john (defined as john:↑σ ) is a member of the set value of the discourse function signified by (↓σ df) within the information structure ↑σι . (This is similar to the treatment of PCASE in early treatments of obliques.) Glue and information structure – 31 / 53
Phrase structure rules Subject as the default topic: IP
−→
NP (↑ subj)=↓ ↑σι =↓σι ((↓σ df)=topic)
I ↑ =↓
Glue and information structure – 32 / 53
Relations between structures IP NP (m subj)=j mσι =jσι ((jσ df)=topic) N
I
m : subj
j : pred
‘John’
N
John (j pred) = ‘John’ john ∈ (jσι (jσ df)) Glue and information structure – 33 / 53 10
Functional description (m subj)=j mσι =jσι ((jσ df)=topic) (j pred) = ‘John’ john ∈ (jσι (jσ df))
m : subj
j : pred
‘John’
Glue and information structure – 34 / 53
Functional description
mσι =jσι : requires the information structure corresponding to m and j to be the same. We assume that all members of a clause share the same information structure, and that all phrase structure rules of a language bear specifications like this one to ensure this. The result is that specifying a particular information structure role for a meaning constructor means that it bears that information structure role within the entire clause.
((jσ df)=topic) provides an optional, default discourse function topic for the subject.
john ∈ (jσι (jσ df)): the meaning constructor john must bear the role specified by (jσ df) at information structure. Glue and information structure – 35 / 53
Functional description ((jσ df)=topic) john ∈ (jσι (jσ df))
≡
john ∈ (jσι topic)
the semantic structure jσ corresponding to j has the feature df with value topic
the value of jσ ’s df appears as the feature topic in the information structure for the clause, mσι Glue and information structure – 36 / 53
Rules IP
−→
I
−→
NP (↑ subj)=↓ ↑σι =↓σι ((↑σ df)=topic) I VP ↑ =↓ ↑ =↓
VP −→
V ↑ =↓
−→
V ↑ =↓
V
I ↑ =↓
⎞ NP ⎝(↑ obj)=↓⎠ ↑σι =↓σι ⎛
Glue and information structure – 37 / 53 11
Lexical entries married
V
(↑ pred) = ‘marrysubj,obj’ marry ∈ (↑σι (↑σ df))
Rosa
N
(↑ pred) = ‘Rosa’ rosa ∈ (↑σι (↑σ df)) Glue and information structure – 38 / 53
C-structure and f-structure
⎤ ‘marrysubj,obj’ ⎢
⎥ ⎥ ⎢ m :⎢ subj j : pred ‘John’ ⎥ ⎣
⎦ obj r : pred ‘Rosa’
IP NP (m subj)=j mσι =jσι ((jσ df)=topic)
I
N
VP
N
V
John (j pred) = ‘John’ john ∈ (jσι (jσ df))
V
⎡
pred
NP (m obj)=r mσι =rσι
married (m pred) = ‘marrysubj,obj’ marry ∈ (mσι (mσ df))
N N
Rosa (r pred) = ‘Rosa’ rosa ∈ (rσι (rσ df))
Glue and information structure – 39 / 53
Contribution from linguistic and pragmatic context (jσ df) = topic (mσ df) = focus (rσ df) = focus This information comes from agreement, casemarking, prosody, word order, context... Glue and information structure – 40 / 53 12
John married Rosa (A)
⎤ (m pred) = ‘marrysubj,obj’ ⎡ pred ‘marrysubj,obj’ (m subj)=j ⎢
⎥ ⎢ ⎥ m :⎢ subj j : pred ‘John’ ⎥ (j pred) = ‘John’ ⎣
⎦ (m obj)=r obj r : pred ‘Rosa’ (r pred) = ‘Rosa’
(B)
(jσ df)=topic
jσ :[ df topic ]
(mσ df) = focus
mσ :[ df focus ]
(rσ df) = focus
rσ :[ df
(C)
focus ]
john ∈ (jσι (jσ df)) marry ∈ (mσι (mσ df)) rosa ∈ (rσι (rσ df)) mσι =jσι mσι =rσι Glue and information structure – 41 / 53
John married Rosa (C)
john ∈ (mσι topic) marry ∈ (mσι focus) rosa ∈ (mσι focus) ⎡ topic
⎢ ⎢ ⎣ focus
mσι :⎢
{john}
marry rosa
⎤ ⎥ ⎥ ⎥ ⎦
Glue and information structure – 42 / 53
Our architecture c-structure φ f-structure σ s-structure ι i-structure Glue and information structure – 43 / 53 13
King (1997)’s proposal c-structure f-structure
i-structure s-structure semantics
Similar to Mycock (2006). Glue and information structure – 44 / 53
Advantages
Easy to specify constraints on f-structure/i-structure relation.
Solves granularity problem: Meaning constructors are the right size to be assigned an information structure role
All meaning constructors bear some information structure role Glue and information structure – 45 / 53 14
15
References
*
47 / 53
References Asudeh, Ash. 2004. Resumption as Resource Management. Ph.D. thesis, Stanford University. Butt, Miriam & Tracy Holloway King. 1996. Structural topic and focus without movement. In Miriam Butt & Tracy Holloway King (editors), On-line Proceedings of the LFG96 Conference. URL http://csli-publications.stanford.edu/LFG/1/lfg1.html. Butt, Miriam & Tracy Holloway King. 2000. Null elements in discourse structure. In K. V. Subbarao (editor), Papers from the NULLS Seminar . Delhi: Motilal Banarsidass. Choi, Hye-Won. 1999. Optimizing Structure in Context: Scrambling and Information Structure. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised and corrected version of 1996 Stanford University dissertation. Dalrymple, Mary (editor). 1999. Semantics and Syntax in Lexical Functional Grammar: The Resource Logic Approach. Cambridge, MA: The MIT Press. Dalrymple, Mary. 2001. Lexical Functional Grammar , volume 34 of Syntax and Semantics. New York, NY: Academic Press. Dalrymple, Mary & Irina Nikolaeva. 2010. Objects and information structure. Cambridge, UK: Cambridge University Press. In preparation. Gundel, Jeanette K. 1988. The role of topic and comment in linguistic theory . New York: Garland. Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar . Cambridge, MA: The MIT Press. King, Tracy Holloway. 1995. Configuring Topic and Focus in Russian. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised and corrected version of 1993 Stanford University dissertation. King, Tracy Holloway. 1997. Focus domains and information structure. In Miriam Butt & Tracy Holloway King (editors), On-line Proceedings of the LFG97 Conference. URL http://csli-publications.stanford.edu/LFG/2/lfg97.html. Klein, Ewan & Ivan A. Sag. 1985. Type-driven translation. Linguistics and Philosophy 8, pp. 163–201. Lambrecht, Knud. 1994. Information structure and sentence form: Topic, focus, and the mental representation of discourse referents. Cambridge, UK: Cambridge University Press. Mycock, Louise. 2006. The Typology of Constituent Questions: A Lexical Functional Grammar analysis of ‘wh’-questions. Ph.D. thesis, University of Manchester. Mycock, Louise. 2009. “What do you do?”: Variation in interrogative predicates. Paper presented at the workshop “Blurring Component Boundaries: levels of analysis or growth of information?”, LFG09, Cambridge, July 2009. Reinhart, Tanya. 1982. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27, pp. 53–94. Vallduv´ı, Enric. 1992. The Informational Component. New York: Garland. Vallduv´ı, Enric & Elisabet Engdahl. 1996. The linguistic realization of information packaging. 16 Linguistics 34, pp. 459–519. Glue and information structure – 47 / 53
Meanings Meanings are expressions like David , yawn(David) , yawn ... Function: when applied to an argument, yields a unique value. yawn :
applied to David , yields “true”. applied to Fred , yields “true”. applied to George , yields “false”. ...
Function application: yawn applied to David = yawn(David)
Glue and information structure – 48 / 53
Application and abstraction Lambda abstraction: λX.P represents a function from entities represented by X to entities represented by P . Usually, the expression P contains at least one occurrence of the variable X, and we say that these occurrences are bound by the λ lambda operator. Function application: [λX.P ](a) The function λX.P is applied to the argument a. Equivalent to the expression that results from replacing all occurrences of X in P with a. Example: [λX.yawn (X)](David ) ≡ yawn (David ) Glue and information structure – 49 / 53
Resource logic Klein & Sag (1985, page 172): Translation rules in Montague semantics have the property that the translation of each component of a complex expression occurs exactly once in the translation of the whole. . . . That is to say, we do not want the set S [of semantic representations of a phrase] to contain all meaningful expressions of IL which can be built up from the elements of S, but only those which use each element exactly once. Glue and information structure – 50 / 53 17
Linear logic Linear logic: a resource logic without rules of weakening and contraction.
Weakening: we can include additional hypotheses in forming a conclusion. If it is raining, you might get wet. It is raining. Therefore you might get wet. If it is raining you might get wet. It is raining. It is Wednesday. Therefore, you might get wet.
Contraction: a hypothesis can be used any number of times. If it is raining, you might get wet. If it is raining, you might catch a cold. It is raining. Therefore, you might get wet, and you might catch a cold. Glue and information structure – 51 / 53
Linear logic and language Weakening unwanted: *Bill yawned Fred = Bill yawned. Contraction unwanted: *He thought that would win = He thought that he would win. Glue and information structure – 52 / 53
Linear logic and language multiplicative conjunction: ⊗ (“and”) linear implication: −◦ (“implies”) INCORRECT:
A
(A ⊗ A)
INCORRECT:
(A ⊗ B)
A
(A ⊗ (A−◦B))
B
CORRECT:
INCORRECT: (A ⊗ (A−◦B))
(A ⊗ B)
INCORRECT: (A ⊗ (A−◦B)) (A−◦B) ⊗ B Glue and information structure – 53 / 53
18