Accenting and Given/New Status

Presentation on theme: "Accenting and Given/New Status"— Presentation transcript:

1 Accenting and Given/New Status
Julia Hirschberg CS 4706 11/26/2018

2 Issues What discourse structures are important in processing?
What are the intonational cues to these structures? Can listeners readily attend to these cues? On-line? How do they incorporate intonational information in their interpretations? What other information do they use? 11/26/2018

3 Defining Given/New Halliday ‘67:
Given: Recoverable from some form of context New: Not recoverable Chafe ’74 ’76: Given: what S believes is in H’s consciousness New: what S believes is not… “Chafe-givenness” Yesterday I had my class disrupted by a bulldog/dog. I’m beginning to dislike dogs/bulldogs. But not vice versa…. 11/26/2018

4 Why do we care about the given/new distinction?
Building a model of the discourse What do S and H believe to be true? What is in their consciousness now? What is ‘grounded’? Speech technologies TTS: Given information is often deaccented while new information is usually accented … but ASR... 11/26/2018

5 Prince ’81: A Given/New Taxonomy
Text as set of instructions from S to H on how to construct a discourse model Model includes discourse entities, attributes, and links between entities Discourse entities: individuals, classes, exemplars, substances, concepts (NPs) Entities as ‘hooks’ on which to hang attributes (Webber ’78) Entities when first introduced are new 11/26/2018

6 Brand-new (H must create a new entity)
I saw a unicorn today. Unused (H already knows of this entity) I saw your roommate today. Evoked entities are old -- already in the discourse Textually evoked The unicorn was eating a muffin. Situationally evoked That bus was going too fast! Inferables Containing 11/26/2018

7 Non-containing I bought a pair of socks. One of them had a hole in it.
I bought a new car. One of the tires is already flat. 11/26/2018

8 Prince ‘92 Definiteness: subject NPs tend to be syntactically definite and old Indefiniteness: object NPs tend to be indefinite and new I saw a black cat yesterday. The cat looked hungry. Definite articles, demonstratives, possessives, personal pronouns, proper nouns, quantifiers like all, every signal definiteness Indefinite articles, quantifiers like some, any, one signal indefiniteness 11/26/2018

9 This guy came into the room. There were the usual subjects at the bar.
But…. This guy came into the room. There were the usual subjects at the bar. 11/26/2018

10 What’s wrong with a Hearer-centric model of given/new?
Hearer-centric information status: Given: what S believes H has in his/her consciousness New: what S believes H does not have in his/her consciousness But discourse entities may also be given and new wrt the current discourse Discourse-old: already evoked in the discourse Discourse-new: not evoked 11/26/2018

11 Hearer-new --> Discourse-new (unless…?) Discourse-new --> ?
Hearer-old --> ? Discourse-old --> Hearer-old (unless….?) What does this new distinction buy us? A way to explain definiteness/indefiniteness in terms of Hearer status only More fine-grained distinctions to explain why items are accented or deaccented 11/26/2018

12 Inferrables: H-old, H-new, D-old,D-new
Entities S may believe H can infer based upon a previously evoked discourse entity A bus plowed into a store on Broadway. The driver emerged without a scratch. ?The bumper sticker read “How’s my driving?”. *The coins spilled everywhere. 11/26/2018

13 Containing Inferrables
The entity that evokes the inferable is mentioned at the same time as the inferable The tires of my car were all flat. The fingers on her right hand turned blue. 11/26/2018

14 Empirical Analysis (Prince ‘92)
What regularities can be observed for subjects? Definite/indefinite Hearer-old? Discourse-old? Results: D-old NPs tend to be subjects D-old pronouns more likely to be subjects than other D-old NPs D-old non-pronouns+Inferrables seem to pattern together 11/26/2018

15 Discourse-old predicts subjecthood better than Hearer-old status or definiteness
11/26/2018

16 How does this help us in analyzing/generating a discourse?
Analysis: How can we identify sentence topics? How can we decide what is in S and H’s discourse models? What would this tell us about what they know? Believe? Generation: How do we produce natural-sounding sentences? How easy is this to do? For humans? Machines? 11/26/2018

17 Karl Rove has called the state Ground Zero in the presidential race, and as soon as the general election contest began, Challenging President Bush and Democrat Democrat John Kerry took their campaigns to the Sunshine State. Asked whether economic conditions or terrorism would be more important to their vote this fall, 65% in the survey said economic conditions, and only 26% said terrorism. Fifty-seven percent of Americans say it's time for a "new direction" in the nation's leadership, fewer than at this point in his father's presidency but a majority nonetheless. Arguing that he is happy to make Sept. 11 an election issue, Kerry said Sunday that the Bush administration is "stonewalling" a commission investigation into security failures before the terrorist attacks. Kerry on Friday said job losses "rip the heart out of our economy" and criticized Bush's ability to make up for the millions of jobs that have disappeared during his administration. Kerry, giving the Democrats' weekly radio address, said Bush had misplaced priorities in Iraq, spending billions of dollars on contracts to Halliburton Co. but not providing troops enough body armor and other protective equipment. 11/26/2018

18 Intonational marking of given/new
Pitch accents can cue given/new: Halliday (1967), Brown (1983), Terken (1984) Hirschberg & Pierrehumbert (1986) relate given/new and accentuation to G&S’s model of global salience, accounting for ‘exceptions’: deacc of entities in non-immed global focus deacc of entities in sister DS, see Davis & Hirschberg (1988), Nakatani (1997) accentuation of re-introduced given entities 11/26/2018

19 Brown ‘83: How does accent related to given/new?
Speech elicitation in lab Scottish English Prince ‘81 derived categories Results: Brand-new information accented Note: new entity/old expression Inferable information accented Evoked information deaccented 11/26/2018

20 Bard’99: Givenness, deaccenting and intelligibility
Speech elicited in lab Glasgow English Map Task Given: repeated mention (within dialogue vs. across dialogue) Does structural similarity --> deaccenting --> less intelligibility (Terken & Hirschberg ‘94) Weird definition of deaccenting (includes difference in boundary tone) 11/26/2018

21 So, do lab experiments generalize?
Results: Deaccenting rare in repeated mentions (within and across dialogues) Second mentions less intelligible whether deaccented or not (tho more change when deaccented) Structural difference doesn’t account for deaccenting But same structure repetitions rare 11/26/2018

22 Intonation in processing information status
Are listeners sensitive to intonational correlates of information status? Evidence that ‘appropriate’ accentuation facilitates comprehension: Birch & Clifton (1995): appropriateness speeds makes-sense judgments in Q&A pairs Bock & Mazzella (1983): comprehension time of denial-counterassertion pairs Davidson (2001): phoneme-monitoring in denial-counterassertion pairs Terken & Nooteboom (1987), Nooteboom & Kruyt (1987): verification latencies of target words 11/26/2018

23 Intonational cues in on-line processing
Dahan et al. (2002): accentuation effects referential interpretation even at very early stages of processing Used eye-tracking to monitor listeners’ fixations on pictured entities as they heard instructions to manipulate these entities on computer screen Examined moment-by-moment recognition of accented vs. unaccented words which share a primary-stressed initial syllable (e.g. candy/candle). 11/26/2018

24 Dahan et al. (2002) Example discourse:
“Put the candy below the triangle. Now put the CANDLE above the square.” pred. fixation utt 1 utt 2 <candle> Now put the [kæn] ... <candy> Now put the [kæn] ... = candle = candy 11/26/2018

25 Competition from “candy” upon hearing accented [kæn].
candle-CANDLE condition candle candy [From Dahan et al. (2002)] Competition from “candy” upon hearing accented [kæn]. Accentuation is used by listeners to process discourse representations on-line, as a word is unfolding. 11/26/2018

26 Accent, Given/New, and Grammatical Function
Given/new interacts with property-sharing constraints to determine the distribution of accents in discourse. Terken & Hirschberg (1994): prior mention and grammatical function --> deaccenting Dahan et al ‘02 also examine conditions in which the antecedent and target did not share grammatical role: “Put the necklace below the candle. Now put the CANDLE above the square.” NO competitor effects (i.e. looks to “candy” upon hearing accented [kæn]) 11/26/2018

27 “Put the necklace below the candle.
Now put the CANDLE above the square.” Prediction: “candle” is given  competition from “candy” upon hearing accented [kæn]. [From Dahan et al. (2002)] No competitor effects in this condition! Here, accent serves to cue shift in “focus”. [See also Terken & Hirschberg (1994)] 11/26/2018

28 Grammatical Role or Syntactic parallelism?
Dahan et al.’s and Terken & Hirschberg’s data confound grammatical role with syntactic parallelism Venditti et al ‘02,’03: syntactic parallelism (NOT just persistence of grammatical role) affects interpretation of nuclear-accented pronouns John hit Bill and then HE ... hit George. (N2 pref) ... ran away (less N2 pref) N2 pref when syntatically parallel clauses 11/26/2018

29 Future directions Processing in dynamic models of attention and discourse salience Property-sharing, parallelism, and other interacting constraints on accentuation Time course of integration of intonational information Cross-linguistic perspectives on intonation and discourse processing Role of ambiguity ‘awareness’ in experimental designs 11/26/2018

30 Next Week Read ASR overview
Go to Sameer’s lab on March 24 to get help building a TTS system or using a speech recognizer 11/26/2018