Introduction Discourse (narrow definition): A sequence of sentences. Discourse (wide definition): A sequence of sentences in language use. > stays close to a purely linguistic analysis > opens up for cross-disciplinary research in linguistics, retorics, anthropology, psychology,...
Discourse in Utrecht (selection) Discourse (in the narrow sense) is at the heart of much of the research of the Weak Referentiality Project (Henriette de Swart). Discourse (in the wider sense) is at the heart of the research projects on Subjectivity and Causality (Ted Sanders) and Emotion and Perspective (Jos van Berkum) > We’ll get back to this next week. > We’ll have a look at it right now.
Discourse in Utrecht: Ted Sanders De auto van de buren is weg. Dus ze zijn niet thuis. Waarom kun je niet zeggen De auto van de buren is weg. Daardoor zijn ze niet thuis? Wat zijn de regels achter het taalgebruik? We onderzoeken hoe verschillende talen zulke oorzakelijke verbanden uitdrukken in connectieven als daardoor, daarom en dus, hoe lezers ze interpreteren en kinderen ze leren. Het project vertrekt vanuit de gedachte dat de keuzes die sprekers maken uit linguïstisch verwante vormen, informatie verschaffen over de categorieën van menselijk denken. We bestuderen taalgebruiksgegevens, in geschreven en gesproken taal. De hypothesen zijn dat Causaliteit een universeel principe is en dat talen hun causaliteitstypen opdelen in termen van Subjectiviteit, zij het elk op een eigen wijze. Daardoor is dan een zeer objectief connectief, dus een zeer subjectief.
Discourse in Utrecht: Ted Sanders Omdat Causaliteit en Subjectiviteit twee fundamentele cognitieve principes zijn, verwachten we dat ze ook conceptuele complexiteit bepalen. Met andere woorden: verondersteld wordt dat de mate waarin Causaliteit en Subjectiviteit (prominent) aanwezig zijn én de wijze waarop ze gecombineerd worden, bepalend zijn voor de mate van complexiteit van het tekstverband. Daarom onderzoeken we in welke volgorde kinderen tekstverbanden leren uitdrukken. Dat gebeurt [...] in corpora en in experimenten. Daarbij is de hypothese dat causale en subjectieve relaties relatief complex zijn en dus later zullen worden verworven dan connectieven die minder complexe relaties uitdrukken. Dit zijn de objectieve en niet-causale relaties, zoals additieve (en) en temporele opeenvolging (toen) relaties. De complexiteitshypothese wordt bovendien vergeleken met een andere mogelijke verklaring, namelijk het taalaanbod dat kinderen krijgen van hun ouders: is dat aanbod niet méér bepalend voor de verwervingsvolgorde?
Discourse in Utrecht: Ted Sanders Ten slotte onderzoeken we de wijze waarop lezers causale relaties interpreteren. De relatieve complexiteit van causale relaties zou kunnen leiden tot de hypothese dat ze lastiger zijn om te verwerken dan niet-causale relaties als additie en temporele opeenvolging. Toch is bekend dat lezers vaak snel en gemakkelijk causale relaties leggen. Een sequentie als Jan kwam binnen. Ida rende weg wordt veelal causaal geïnterpreteerd. De vraag is, of mensen altijd causale relaties leggen als ze de kans krijgen, en zo ja, onder welke voorwaarden dit het geval is. Wat is de rol van de connectieven hierbij? Deze vragen worden experimenteel onderzocht, in leestijd- en oogbewegingsregistratie-onderzoek.. Het onderzoek is sterk interdisciplinair georiënteerd en maakt gebruik van de methodologie van convergerende evidentie – een combinatie van linguïstische theorie, geschreven en gesproken corpora uit verschillende talen, kindertaaldata en leesexperimenten – om de fundamentele rol van Causaliteit en Subjectiviteit in taal én in menselijke cognitie te verduidelijken. Zo maakt de studie van kleine woordjes (dus, want) het mogelijk om grote kwesties van taal, cognitie en communicatie te onderzoeken.
Discourse in Utrecht: Jos van Berkum The relation between discourse and emotion is at the heart of several questions raised within the humanities. In literary theory, people have wondered about the paradox of fiction, about how a non-existing world created with words can evoke real emotion. Researchers in language and communication ask how we can use text to change people’s affective evaluations (attitudes), and how we can get readers to be interested in text in the first place. Linguists, psycholinguists, and philosophers ask about the fundamental goals and mechanisms behind linguistic communication, drawing attention to the fact that such communication is inherently social and argumentative, with speakers and authors always conveying their own affective or moral perspective and ideology (author stance). And cultural/cognitive anthropologists ask about the mechanisms by which interlocutors regulate emotion in a conversational exchange, e.g., via politeness and maintaining face.
Discourse in Utrecht: Jos van Berkum To make progress on these matters, we need an account of how readers are moved by language, at the level of actual processing. The aim of this proposal is to develop such an account. Using methods borrowed from cognitive neuroscience, we ask four questions about how language interfaces with affect. First, how do linguistic and affective representations interact as people read a simple sentence? Second, how do readers deal with the affective stance of the author, implied by the choice of words and expressions? Third, how do readers balance their own feelings, and the value systems beneath them, with those of protagonists, as narrative develops? And finally, where in the processing of language do involvement and perceived personal relevance make a difference? Results will shed new light on old questions, and can help us understand how affective factors relate to ‘cold cognition’. Also, with affect controlling so many of our day-to-day choices, the findings should have implications for the design of societally important messages.
Discourse in this course In this course and especially in this and the following class, we’ll focus primarily on discourse on its narrow definition.
Discourse Analysis 2011 DRT: preliminaries
Discourse Analysis 2011 Why do we do DRT?
DRT “[In order] to understand what information is added by the next sentence of a discourse [...] the interpreter must relate that sentence to the information structure he has already obtained from those preceding sentences” Kamp & Reyle (1993) > Discourse Representation Theory > Major insight:
DRT “[In order] to understand what information is added by the next sentence of a discourse [...] the interpreter must relate that sentence to the information structure he has already obtained from those preceding sentences” Information structure A Information structure B sentence X
Sentence order is not random [about the climbing of a mountain] We climbed till cabin H. We spent the night there. The next day we started on the north side. 12 hours later we reached the top. [about the climbing of a mountain] The next day we started on the north side. We climbed till cabin H. 12 hours later we reached the top. We spent the night there. > The order in which we put sentences matters! A different order leads to semantic differences!
Sentence order is not random > Try to come up with a four sentence story in the third person with at least two possible orders. > Is this difficult? Languages come equipped with explicit ways to indicate that a sentence b is to be linked to a preceding sentence a. It’s not always easy to overrule these links. >John came in. He sat down. He came in. John sat down. >Immediately after that our friend fell. Mary pushed Tintin. Mary pushed Tintin. Immediately after that our friend fell. >A man and a woman are walking in the park. The man is whistling. The man is whistling. A man and a woman are walking in the park.
Sentence order is not random > Pronouns and definite descriptions signal that they refer back to something that has been introduced before. > Proper names and indefinite descriptions don’t come with this constraint...
Sentence order is not random >I saw a pig running around in the garden and a few minutes later I also discovered a pig running around in the house. >I saw John running around in the garden and a few minutes later I also discovered John running around in the house. > Indefinite descriptions typically introduce new referents into the discourse. The referents of proper names are typically stable across the discourse.
Discourse Analysis 2011 DRT: first formal steps
Formalizing simple discourses > The semantics: We create a mini-world (an information structure) on the basis of all the information we get. This mini-world consists of individuals and lists all the information we have on them. The representation of this mini-world is called ‘Discourse Representation Structure’ or simply ‘box’.
A toy example Miffy loves Boris. She’s very happy. the individuals in our mini-world:the information we have on them: Miffy Boris
A formalized toy analysis Miffy loves Boris. She’s very happy. the individuals in our mini-world:the information we have on them: Miffy Boris x y (x) (y) x loves y x is happy universe conditions box or DRS
Now the procedure... > Starting point:... we’ll keep it simple though. > Examples: Miffy loves Boris. [Miffy [loves [Boris]]] She is happy. [She [is happy]] Given that we’re dealing with a semantic theory and given that semantic theories attribute meanings to forms we have to start from a syntactic analysis of sentences...
Now the procedure... [Miffy [loves [Boris]]] Construction Rule for Proper Names 1. Introduce a new discourse referent into the universe. 2. Introduce into the condition set a condition formed by placing the discourse referent in parentheses behind the proper name which, in the syntactic structure from which the triggering configuration is drawn, is inserted in the Proper Name slot of the configuration. 3. Introduce into the condition set a condition obtained by replacing, in the syntactic structure referred to (under 2), the NP-constituent by the new discourse referent. 4. Delete the syntactic structure containing the triggering configuration from the DRS. x Miffy (x) [x [loves [Boris]]] y Boris (y) [x [loves [y]]]x loves y
Now the procedure... x Miffy (x) y Boris (y) x loves y [She [is happy]] z [z [is happy]] z =x z is happy Construction Rule for Pronouns 1. Introduce a new discourse referent into the universe. 2. Introduce a condition obtained by substituting this referent for the NP slot of the local configuration that triggers the rule application in the syntactic structure containing this configuration and delete that syntactic structure. 3. Add a condition of the form = where is a new discourse referent and is a suitable discourse referent chosen from the universe of the DRS.
Formal details... x Miffy (x) y Boris (y) x loves y z [z [is happy]] z =x z is happy x Miffy (x) y Boris (y) x loves y z [z [is happy]] z =x x is happy
What to do with a DRS? > In semantics we in general formalize the conditions that have to be met for a sentence to be true. > DRT is not really different: A DRS is true provided we can find individuals for each of the discourse referents in its universe in such a way that the conditions which the DRS contains for particular discourse referents are satisfied by the corresponding individuals. > When is this DRS true? We need to find two individuals that are called Miffy and Boris respectively and of which the first loves the second and is happy.
One more construction rule... Construction Rule for Indefinites (slightly modified) 1. Introduce a new discourse referent. 2. Introduce into the condition set a condition formed by placing the discourse referent in parentheses behind the N which, in the syntactic structure from which the triggering configuration is drawn, is inserted in the N slot of the configuration. 3. Introduce the result of substituting this discourse referent for the NP-constituent in the syntactic structure to which the rule is being applied. 4. Delete the syntactic structure containing the triggering configuration from the DRS. [Miffy [has [a [dog]]]] x Miffy(x) [x [has [a [dog]]]] y [x [has [y]]] dog(y) x has y
Discourse Analysis 2011 Exercises
Exercise 1 Describe the DRS-constructions for the examples below. Some of the discourses are intuitively not acceptable. They are, however, analysable by the Kamp/Reyle construction algorithm, which assigns all of them a well-formed, completed DRS. How might the algorithm be modified, so that the construction of these DRSs be blocked? John loves Mary. She likes him. Oliver Twist loves Anna Karenina. It fascinates it. Oliver Twist loves Anna Karenina. She fascinates it. Mary bought a novel. She adores it. Jones admires a woman. He likes her.
Exercise 2 Imagine we have a sentence with several NPs. To which one should we apply the construction rules first? Base your answer on an intuitive analysis of the following pair: A stockbroker who knows him likes Bill. A stockbroker who knows Bill likes him.
Exercise 3 For the two discourses below answer the following questions: Does a DRS-derivation exist for the given discourse – and if yes, which? Does the resulting DRS match the intuitive meaning of the discourse? Is the discourse ambiguous? Anna loves Carl. Betty abhors Carl. Anna loves a man. Betty abhors a man.
Discourse Analysis 2011 A few more rules
An intuitive account of definites [The weather is really bad and lightning is striking everywhere] A man is walking on the beach. The fool is carrying an antenna. > Intuitively definite descriptions work like pronouns in the sense that they pick up a referent that has been introduced before. They will therefore introduce a condition of the type = . > They differ from pronouns though in having lexical content. They will therefore furthermore introduce a condition of the type N(x). x man (x) y x walking_on_the_beach y is_carrying z z antenna(z) y =x fool(y)
Every Every human being is mortal. x human being (x) x is_mortal x every human being (x) x is_mortal
Every Every human being is mortal. *He will die sooner or later. x... (x) x is_mortal x will die sooner or later > Intuition: we create a hypothetical mini-mini-world within our mini-world. We furthermore assume that hypothetical mini-mini-worlds in principle don’t extend across sentence boundaries.
Every Every human being is mortal. *He will die sooner or later. x human_being(x)x is mortal y y=??? [he will die sooner or later]
If...then If a tax agent calls, then he wants money. x tax agent(x) y wants_money If a farmer owns a donkey, he beats it. x y farmer(x) z=x donkey(y) x owns y y y=x z w w=y z beats w
Discourse Analysis 2011 Exercises
> Construct a DRS for the following sentence. What do examples like this one tell you about proper names in if...then constructions? If John sees a book, he buys it. He’s possessive. > Construct a DRS for the following sentence. Is this discourse supposed to be felicitous? (you can analyze one fireman as a fireman) Every fireman received a silver medal. One fireman sold it right away.
Discourse Analysis 2011 Why do we do DRT?
> DRT is a formalization of the way a discourse is built up. As such it’s a valuable tool for those who want to give precise analyses of discourse phenomena. > Discourse phenomena are not to be underestimated. They include: - pronouns - definite descriptions - presupposition projection (see last week) - tense/aspect (see TK4 in the French department) -...
Why do we do DRT? > DRT offers an interesting way of looking at indefinites that solves some thorny problems that standard first order predicate logic cannot cope with. x y farmer(x) z=x donkey(y) x owns y z w w=y z beats w If a farmer owns a donkey, he beats it. > The fact that this sentence is about all farmer/donkey pairs is due to the conditional. > A similar effect cannot easily be obtained in standard first order predicate logic.
Why do we do DRT? > A farmer owns a donkey. x y(farmer(x)&donkey(y)&own(x,y)) > If a farmer owns a donkey, he beats it. x y(farmer(x)&donkey(y)&own(x,y)) beat(x,y) x y(farmer(x)&donkey(y)&own(x,y) beat(x,y)) x y(farmer(x)&donkey(y)&own(x,y) beat(x,y))
Discourse Analysis 2011 A glimpse of what lies beyond DRT
Question 6 forward looking center Cf backward looking center preferred center Cb Cp
Question 9 the backward-looking center of sentence A is not the same as the backward- looking center of sentence B (doesn’t hold when the backward-looking center of A is unknown) the backward-looking center of sentence A is the same as the backward- looking center of sentence B (also holds when the backward-looking center of A is unknown) the backward-looking center and the preferred center of B aren’t the same the backward-looking center and the preferred center of B are the same rough shift smooth shift retain continue
Why do we do DRT? > Use your knowledge about Centering to explain why in the following discourse speaker B uses the full name Hillary instead of the pronoun her, despite the fact that her would refer unambiguously to Hillary. AWell, I have to go. I have to do a lot of studying. And Hillary said she’d call me if she was going to go to the library with me. But I don’t think she will. Anyway, I’m going to have these papers xeroxed and I’ll come back in a little bit. BOkay, say hi to Hillary for me. AOkay, I will.
Why do we do DRT? > Use your knowledge about Centering and your intuitions about discourse relations to explain why in the following fragment B uses a pronoun in her last utterance to refer to her grandmother, despite the fact that DRT would predict massive ambiguity.
Why do we do DRT? AMy mother wanted to know how your grandmother is doing. BI don’t know, I guess she is all right. She went to the hospital again today. AMm-hm? BI guess today was the day she’s supposed to find out if she goes in there for an operation or not. AMm hm. BSo I don’t know, she wasn’t home by the time when I left for school today. Well, my aunt went with her anyway this time. My mother didn’t go. AMm hm. BBut I don’t know, she will probably have to go in soon, though.