- 1 Hannah Rohde
- 2 Vera Demberg:
- 3 Laura Kertz:
- 4 Jennifer Spenader:
- 5 Oliver Bott & Torgrim Solstad
- 6 Elsi Kaiser & Boutaina Cherqaoui
- 7 Joshua Hartshorne
- 8 Arnout Koornneef
- 9 Roger van Gompel
- 10 Nathaniel J. Smith & Roger Levy
- 11 J. P. de Ruiter & Chris Cummins
- 12 Katharina Spalek, Nicole Gotzner & Isabell Wartenburger
- 13 Juliane Burmester, Katharina Spalek & Isabell Wartenburger
(did I hear this before? AMLaP? Maybe CogSci)
Speakers use language to give people a snapshot. Do they? Oh yes, I think I remember this. strictly sentences.
Cues from context guiding coherence -
last q: IS highly predictable discourse what comprehenders are expecting (should ref this)
Story continuation experiments (Stevenson, Crawley and Kleinman, 1994) ambiguous pronoun. I'm not surprised it's 50/50. Apparently about event structure (end-state). That makes sense too - given that
we are interpreting stuff incrementally and we are bound by time.
Okay, so this is all about how on earth do we know what anybody is going to say in any given situation.
Implicit causality verbs (scolded vs annoyed). Pronouns as indicators of coherence. Well, yes.
Structural ambiguity. Relative clause attachment. But do these things ever end up ambiguous in general dialogue?
Online expectations. Yeah, go on then. Oh. Anticipatory looking (your Altman and Kamide schtick)
Predictable versus useful (if it was all totally predictable, then nobody would need to say anything) Is predictable stuff less informative? Or easier to process.
Expectations of informativity (sounds Gricean)
Oh dear. Intended message. Bollocks and fuck. This question is totally gricean. Not really sure what he means by a genertive model of the speaker (sounds meta... DS would throw that out for a start).
Do we actually process dialogue and the experimental sentence stuff in the same way?
Good question - which goes to my stuff (really predictable stuff you don't need to say). But we do say stuff that is totally mundane and predictable. All the time. So...
It's all context. We bring so much baggage to interpreting anything. And we try to interpret everything.
Discourse connectors (because, so, etc etc. Still sentences).
Are they processed incrementally (well, yeah?) and can people make predictions based on them.
Visual world paradigm (Altman and Kamide-esque again)
Concessives as negative causals (although vs because). They are harder. Okay. And?
Pupillometry is the new thing, isn't it? Pupil dilates more with increased cognitive load. Okay, I'll buy that. Why though?
Index of cognitive activity (changes in pupil size).
Blimey. The same boring study as visual world, pupillometry, reading times and EEG. FFS.
I fucking hate EEG studies. So meaningless. What do N400s and P600s even mean? And why do they standardly put negative upwards. Ponces.
Coordination. But only in the sense of 'and'.
General priming effect. Ha ha. Syntactic priming is STILL BOLLOCKS.
Is it about structure or semantic integration? If DS is true, that's the same thing (actions)
Choice of coordinator matters (and/but etc etc)
Coherence-based account (connections among ideas)
VPE information structure.
Sentence completion task.
Argument ordering phenomenon (works for us, I think)
Discourse connectives again (the emo connective 'but'. Nice)
What are the semantics of discourse connectives (if we don't need them for the coherence relationship) - do they also have a focusing function.
We try to find an interpretation. This is now a running theme.
Nonsense verbs to factor out verb semantics - I like it. But people interpret them as something close.
Finding the opposite in the corpus and experiment (subject preference in corpus, object in experiment).
Oliver Bott & Torgrim Solstad
More implicit causality
I have no idea what is being said or why. I think I may have reached my limit.
Okay it's another sentence completion task. This should be referred to by me.
Underspecification. Am I going to say anything about that?
Hot hot hot. Where did that nice breeze go?
Elsi Kaiser & Boutaina Cherqaoui
Pronouns and demonstratives in French - influenced by referents predictability?
Anticipation/prediction sentence completion again (open-ended prompt versus pronoun)
Oh that's not a good model of speaker and listener. That's basically Shannon and Weaver i.e. WRONG
2 kinds of expectations: likelihood of mentioned; choice of referring expression
Connected or not (she claim not; does the fact you choose a pronoun rely on how predictable the referent is?)
How specific are expectations?
Pronouns used for more salient referents --> sliding scale allegedly.
What happens when a pronoun is potentially ambiguous? How on earth do you code coherence relations? There's an annotation task I would not like.
After 'so' (alors), object is more predictive and after 'then' (et apres) subject is more predictable.
Pronouns when subject, not when predictable (but still more when predictable? So there could be a bit of both going on)
Demonstratives only in object ones
Exp 1 looks at speaker, exp 2 at listener (well not really as it happens) allegedly.
Oh please don't give her more time. Cripes.
Question about syntactic force? No idea what he means by that. But still. Hans Kamp question 2 (centering theory?)
My topic isn't much like their topic. Oh dear. Loose talk costs lives.
Re-mention biases. All about who a pronoun refers to again.
I really wonder about corpus stuff
- rapid effects on pronoun interpretation
- not predictable from linguistic structure
- largely derived from world knowledge (Pickering and Majid)
All false, allegedly
Semantics of the verbs matters for implicit causality/consequentiality biases
- attempts to do taxonomies of semantic classes fail:
- stimulus-experiencer (frighten)
- agent-patient (call)
- experiencer-stimulus (like)
- agent-evocator (blame)
This doesn't surprise me in the slightest (cf ditransitive verbs) as it's another way to misunderstand how words work.
He (with Snedeker): Syntactic verb classes - syntactic frames give you semantic classes (cf auxiliary ones) S hit M S is hitting M
Verbnet. These syntactic categorizations seem to pattern into groups that are weighted object/subject way M hit S because she is a dax. Who is a dax?
Seems to be about event structure, so DS deals with it (also cope with question from audience about sentences being available to update event structure - i.e. link to e_w node)?
Disproving things - and he got a paper out of it (where is our syntactic priming paper?)
I think this is interesting. It's all in the lexicon. We can do this.
Invited speaker. Final talk day one.
Looks like implicit causality and pronoun resolution rides again.
My talk will be a complete change of direction (though maybe JPs too)
Reading. Self-paced and eye-tracking.
Whoopdedoo it's incremental. Ha ha, funny, he's trying to carry on with his talk as was when the previous one said it's all a myth... Nice guy though.
Anticipation vs immediate integration?
Oh EEG. Bloody hell. Visual world. SO it's an anticipatory thing apparently (and then a mismatch causes longer reading times)
What is the source of implicit causality? lexical/discourse/connective
Different connectives (because/and/but) - oh there was a piping up of "I tried to replicate Ehrlich" (but it didn't work?) - his didn't either.
OK, I've switched off now.
Old people, different styles of reading who uses what information. So different people have different ways of integrating the information and updating their interpretation. That makes sense.
Nice. Bet it has an analogue for hearers - how proactive is your parsing strategy?
Different processing strategies available. Any route that gets you there works.
Maybe you'd get a dfference with depressed people. Or schizophrenics. Hmmmm.
Is lexical stuff different from world knowledge. Well, yes, I think so. Is the lexicon where we store world knowledge?
Roger van Gompel
Invited speaker. Didn't say much at dinner last night (maybe more of an indication of my ability to hijack anything) Oh. He did his PhD with Martin Pickering. So maybe my slagging off of structural priming was a bad thing.
Referring expressions. How do you choose what you choose?
Adults overspecify more than children. Colour is more often a redundant property.
Computer generated referring expressions based on conceptualisation; psycholinguistic stuff less so.
Incremental algorithm (Dale and Reiter) - deterministic. People are not.
Frank and Goodman (2012, Science) Bayesian model of reference. Not really production.
What, a psycholinguistic theory that's not really about language. Who'd'a thunk it. WTF. I am tired. Think I got away with it though. Hmmmm.Maybe not.
Underspecification rarely happens. Over, more so. PRO. computational linguists love an acronym.
Ooh repairs. Okay, pay attention now.
Errors. No no no no no
Underspecification repairs. Okay. But...
Nathaniel J. Smith & Roger Levy
Edinburgh guy. Knows Greg.
Surprisal. Okay. So I should mention this.
People exploit many cues for expectancies (hang on - isn't that my conclusion!)
Good question - how are expectancies represented.
Syntax and words. Words are read faster if they are more predictable.
Nice pitch. Speech recognition is hard. true fact.
Probabilities. Hmmm. Okay. Is the brain doing something quite clever...?
Trigram model of predictability estimation. Probabilities in the brain. Nice lead in to me. Local plus global context. BNC.
It's like a key word search
Featherstone and Sturt (2010)
J. P. de Ruiter & Chris Cummins
(Jan Peter is at TiGeR)
Dialogue act recognition. Hmmm. I sort of disagree with the principle.
Bayesian computational model.
Humans recognise dialogue acts (hmmm) resolve ambiguities. Imposed (interpreted) rather than inherent? Identification. Hmmm.
Ooh, he's quotng Rosa! Gisladottir 2012
'what the current speaker is trying to achieve with their utterance'.
expectations of dialogue acts (cf greetings) adjancency pairs.
scripts etc etc.
Context available as a cue. Yes, but what does this mean?
That's entropy, man.
Katharina Spalek, Nicole Gotzner & Isabell Wartenburger
Focus particles (e.g. only)
Are alternative sets cognitively real
Language processing as a spreading activation process.
Comprehension involves construction of discourse model.
Corrections provide contrastive focus.
Juliane Burmester, Katharina Spalek & Isabell Wartenburger
Last one apart from me. Oh dear - it's going to be an erp one (looks better in lower case!)
Topic - aboutness topic (link to me)
context type - topic vs neutral
I have no idea how to interpret erp stuff. It's just squiggles. Oh god I really am going to fall asleep. In my own talk. Crap.