Speech Transcript - When do we laugh?

0:00:16	my name you hear
0:00:18	it's joint work we can make the quality and driving is well that's are
0:00:24	i australia this that's george running manner
0:00:27	you can see where very active group and we are linguists
0:00:31	from a university pairs of it
0:00:35	and we're going to talk about
0:00:37	when do we love
0:00:43	in previous work has been found that laughter is very frequency compensation
0:00:48	it happens they are different speakers but
0:00:52	frequency between
0:00:53	six fifty times per ten minutes
0:00:55	so
0:00:57	and
0:00:58	laughter can takes
0:00:59	very useful in terms of its
0:01:02	four eight six
0:01:03	and occur in a variety of context
0:01:06	but if i are you
0:01:09	when do we have
0:01:10	i think we maybe not your body find someone the street
0:01:14	it probably like
0:01:15	say
0:01:16	that's we live when someone tell us the joke or we'll off when we have
0:01:20	been tackled but actually those two occurrences are not the most frequent context for laughter
0:01:27	and i'll show you a few examples i don't so we know that laughter is
0:01:31	contagious meaning that we more likely to laugh in social settings we more likely to
0:01:36	love
0:01:37	what we're in the presence of others
0:01:39	so just to show you
0:01:41	a different type of context that laughter can
0:01:44	can't occur i have got from
0:01:49	i
0:01:50	so i've got this one i'll just show your view
0:01:54	okay
0:01:58	the numbers your
0:02:01	is used for all their
0:02:06	model for portable my school seems to
0:02:13	or
0:02:17	also
0:02:21	okay
0:02:27	okay
0:02:31	i
0:02:36	i
0:02:42	i
0:02:46	we think that i i'm glad you guys a lot and so here is a
0:02:51	common may need for example after you know goes failure i mean does not to
0:02:56	get if at a guys is a failure videos are you to that people like
0:02:59	to watch a lot of about it
0:03:01	and but even he's laughing about it
0:03:04	but if he broke is provided
0:03:06	i probably okay he wouldn't and his friends probably would not either
0:03:10	that's not that only happened in happy moments okay
0:03:15	so i'll show you ever very unlikely context for laughter
0:03:19	and show all of you know about the terrorist attack in paris
0:03:24	when so this bands that you go that's
0:03:28	you don't of death metal
0:03:30	and four groups they were drafting the in the battle close the a fiesta with
0:03:36	a lot of people
0:03:38	and you knew about the terrible story so this is the video
0:03:42	of an interview
0:03:43	of the people from the band
0:03:45	you know this something under raising the are talking about the most horrible experiencing their
0:03:50	lives that you wouldn't think they would laugh right but actually a little bit of
0:03:54	a little bit
0:03:56	the describing what happen
0:04:02	or you know you have just so as to the gunfire
0:04:06	actually instinctively from my perspective it seems you see
0:04:12	we see that
0:04:13	and it will have a
0:04:20	okay that more usable or knowing that
0:04:25	where
0:04:27	but i don't use also his friends patted him on the on the shoulder and
0:04:32	he laughed as a little off a little one
0:04:36	and here's a another
0:04:38	there you know everyone's really more than once we don't
0:04:42	our knowledge this reason for you know you have just so this case the gun
0:04:48	actually instinctively from my perspective sushi sue
0:04:54	so that little or no
0:04:56	and here when he's talking about again they were hiding the one of the rooms
0:05:01	of the back
0:05:03	is reduced since we don't very few people would be shot
0:05:08	we people's which were ranchers the door
0:05:13	years ago maybe for each
0:05:15	as someone have led to
0:05:18	is a little elf is a overlapping with speech
0:05:22	it was trying to say someone has left able or something
0:05:26	the bottles you mean room
0:05:30	so
0:05:31	these examples of course they feel laughter is very different from the from the from
0:05:36	the first example
0:05:37	so
0:05:38	it can really locally in the very different context
0:05:41	and although it's that laughter most likely occur in the presence of others
0:05:46	will be interaction you can also happen we just by ourselves
0:05:51	so here is in another example when
0:05:53	the grandmother is laughing all by myself
0:05:56	they see section the sky on the one and thanks to the i
0:06:17	that a lot about their going on there
0:06:21	so i showed you that laughter can really take very different forms as you can
0:06:25	see the plastic one you goals of that metal example is very different from the
0:06:29	other two
0:06:30	and that's and i can occur in very different contexts
0:06:35	and also i was trying to say that laughter is contagious
0:06:39	so here's a very interesting example talking about laughter and dialogue
0:06:43	so this is a core are also
0:06:46	the question was
0:06:47	what's your best random conversation with a stranger
0:06:50	and this example is a also size
0:06:54	it wasn't a traditional confirmation process anybody when likeness
0:06:57	i mean having to mt anyway to start laughing to myself it funny takes no
0:07:01	phone i will then maybe standing opposite me smiles at me
0:07:05	me grin still giggling slightly men grumpy face
0:07:09	and then me come stop eating
0:07:11	is beginning to wonder woman catches my right laughing
0:07:15	slightly i mean me for problems for the laughing gasping for breath now
0:07:19	women burst out laughing at me magical is that women laughing
0:07:22	add to me and when the elevator continues to the bottom or in absolutely as
0:07:27	running down phase hysterics as we exit the elevator
0:07:31	so here's the conversation of or interaction of just laughter
0:07:35	okay so let's do some scientific research about their the current state of research and
0:07:42	after is that most studies this working psychology working social still social linguistics in conversation
0:07:51	analysis
0:07:52	then we have looked at its laughter
0:07:55	a lot about
0:07:57	on it is also be used laughter
0:08:00	independent context
0:08:02	as stimuli
0:08:03	and all they study laughter as a response to humorous
0:08:07	stimuli
0:08:08	but there are very few there are some but not very many studies about laughter
0:08:12	in conversations
0:08:14	if two we had dialogue you want to study laughter in dialogue what can we
0:08:19	study
0:08:20	if we think about this we can study we can approach it from
0:08:24	of three direction
0:08:25	we can study the precondition of laughter
0:08:28	many we can think about
0:08:30	what triggers laughter
0:08:31	we can we can look at the context of laughter
0:08:35	so what is graphical kernel and in what in what kind of order
0:08:40	we kind of what study laughter features of laughter
0:08:43	of the of it so we can study the
0:08:46	the for meaning the phonetic the phonology of laughter we can study all
0:08:51	for propose a semantic meaning of laughter
0:08:55	and we can of course look at laughter
0:08:58	each of laughter in you know
0:08:59	why the that features that are universal
0:09:01	or whether the other language and cultural specific
0:09:05	i is we can study the post condition of laughter meaning
0:09:08	the effects of laughter on dialogue no interaction
0:09:12	so we can study
0:09:13	what can expect a lot of the have some discourse in terms of
0:09:16	for the interaction
0:09:18	we can study the effect on the speaker the addressee and you're interaction in terms
0:09:22	of affiliate even those right
0:09:25	and let's look at two issues
0:09:27	one in the preconditioner whining the post condition
0:09:31	so
0:09:32	in precondition we have this question
0:09:34	which is
0:09:35	what does not have to happen
0:09:37	and there is a very commonly assumed to while slide widely cited
0:09:42	idea is that laughter
0:09:44	one laughter is about is what it follows
0:09:47	so the idea is that's we say something
0:09:50	and we laugh about its task or data it "'cause"
0:09:54	i if is really right so studies as far as we know try to resolve
0:09:58	what laughters about by looking at one laughters adjacent to
0:10:02	and
0:10:04	and
0:10:04	and there's a lot of conclusions based on this assumption
0:10:08	that's most laughter is about something been i'll
0:10:11	this is a very widely cited idea from providing
0:10:15	because they look is what laughter formal
0:10:18	and that's will have more is what we see ourselves
0:10:22	so this is one issue is it true that one laughter is about is what
0:10:25	it follows
0:10:26	this is an issue with a question about the post condition
0:10:30	so that we would like to know what kind of you fact
0:10:33	laughter has on dialogue no interaction so can we have a meaningful
0:10:39	functional taxonomy of laughter and currently doubt thousands of them systems
0:10:43	in the market and there there's little agreement
0:10:47	so we would like to ask what functions do not at all of them have
0:10:51	there have and how well
0:10:53	laughter can have so many different functions
0:10:56	let's look at the first issue in terms of current state so
0:10:59	is it true that's what laughter follows is what is about
0:11:03	and if in a study by from one i twenty three they observed natural conversations
0:11:08	and their methods was like this
0:11:10	when an and when an observed or heard laughter should recorded in a notebook the
0:11:14	moment immediately preceding the laughter the sorry the comments immediately preceding laughter
0:11:19	and if the speaker or the audience laughed
0:11:23	and they know that only ten twenty percent of the previous comments was humorous
0:11:29	and that's they concluded that laughter is for the most part not related to humour
0:11:33	but is about social interaction
0:11:36	i did this conclusion from this study was that nothing never interrupts the each but
0:11:40	punctuated
0:11:42	and
0:11:45	also there's a study by a patent or two thousand four
0:11:49	we also or milking et note that only timing
0:11:53	parameters to decide what laughter is about
0:11:57	so i think of missing one slide
0:11:59	so but
0:12:00	these studies assume an adjacency relationship between laughter and laughable so what the laughter is
0:12:06	referring to
0:12:08	and
0:12:09	we would like to question this assumption because this assumption hasn't been studied
0:12:14	and
0:12:15	then what the preceding comments
0:12:17	he's the reference
0:12:19	and it is not amusing itself image in which to refer to when using event
0:12:25	so and if we mention is dialogue a says
0:12:28	to remember that time and then both in time we're laughing
0:12:33	that the batteries do you remember that time is not humorous in itself
0:12:37	but do not able to all the reference it will be for this interaction is
0:12:41	has been reached
0:12:43	to
0:12:44	reference you are in hd notation of that event
0:12:47	every that right
0:12:50	that the water okay so second issue
0:12:54	how many times sub-block how many types of laughter other if we want to have
0:12:58	a good taxonomy functional taxonomy of laughter
0:13:01	i'm currently there are
0:13:02	so many
0:13:04	so for example
0:13:05	this a suggestion that's we can distinguish
0:13:10	we can have a binary distinction between physical
0:13:13	and the emotional after
0:13:14	and the pirate was ninety nine three proposed at least eight
0:13:19	different social functions including affiliation
0:13:22	aggressions torture anxiety fear joy
0:13:25	source of all
0:13:27	as you need i don't want nineteen ninety four proposed three types of laughter
0:13:31	laughter due to pleasant feeling social after laughter for each intention
0:13:37	higher a lot of the two thousand three oppose not different three types of laughter
0:13:41	approach when you release intention or is the called euler
0:13:45	and control i don't when two thousand five proposed
0:13:48	what i
0:13:50	hearty laughter muse laughter
0:13:52	that's recall after and social actors so on so forth
0:13:55	and so as you can see they really this very little overlap
0:14:00	and we believe that one we have the lack of agreements in terms of functional
0:14:03	taxonomy of laughter
0:14:05	is that there are several layers relates relevant to the analysis of laughter
0:14:09	and different classification system and even types within the system all related different layers of
0:14:15	analysis so just as an example
0:14:18	in spite of ninety three taxonomy
0:14:21	affiliation
0:14:23	meaning to agree is roughly be
0:14:25	location to react performed by laughter
0:14:27	well actually is the feature of the emotional trigger severe really relating to different levels
0:14:33	so we need to have a linguistic approach
0:14:36	to study that are so here's our proposal we propose to look at laughter linguistically
0:14:42	at any events map and a full
0:14:45	and so there's a again common assumption that laughter has only emotional content and no
0:14:51	propositional content
0:14:53	for example and two thousand thirteen
0:14:56	and we argue that laughter needs to be integrated with linguistic input
0:15:01	and for the following reasons first of all
0:15:03	second be was the aspects other than the communicative emotion
0:15:08	so
0:15:10	very often
0:15:11	we ask about clarification questions in terms of what is funny
0:15:15	so when you have
0:15:17	we
0:15:18	understands the emotion that you're communicating what we don't understand
0:15:22	is the reference
0:15:24	that's this laughter is referring to
0:15:27	i'm not
0:15:28	right okay
0:15:31	really
0:15:31	okay
0:15:33	and i spent so much time on the video
0:15:37	okay
0:15:37	so
0:15:38	we so we propose a multi layered approach to study laughter where we distinguish for
0:15:44	meeting and function
0:15:46	so in terms of
0:15:49	we know cats
0:15:51	mostly now in our study
0:15:54	contextual features
0:15:55	an instance of meeting
0:15:57	we look is we are proposing
0:16:00	one semantic meaning a unified semantic meaning with two dimensions
0:16:04	an instance of function
0:16:06	we look at things like feature and prosecution reacts
0:16:10	the semantic meaning we propose that too much
0:16:14	but not
0:16:15	one is the laughable and the are the other is the arousal
0:16:21	okay
0:16:21	so
0:16:23	i'll
0:16:24	to quickly about the so here's a problem proposal in terms of the semantics of
0:16:28	laughter
0:16:29	and we say that the meaning of laughter is guys the appraisal the laughable
0:16:34	l
0:16:35	triggered by triggered a positive psychological shift with the times and that you
0:16:40	of delta
0:16:41	and meaning of laughter is
0:16:43	how do you context dependent
0:16:45	and this meeting well aligned with context reasoning can generate a wide range of functions
0:16:52	go quickly about the arousal dimension with a that's not after a trick as a
0:16:57	possible positives like logical shift
0:17:00	and that's
0:17:02	the arousal dimension signals the amplitude of this shift
0:17:06	this is a continuous dimension
0:17:08	and that it doesn't signal be overall emotional states
0:17:12	so if someone laughing doesn't mean that overall this person has a positive emotional states
0:17:17	but
0:17:18	that's there is the positive shift
0:17:20	and in terms of what we propose at the moment three a degrees
0:17:26	it can run you can be triggered by the laughable can be of type
0:17:31	a playful enjoyment like the prime are running the a rollercoaster
0:17:35	can be about in contrast
0:17:38	for example
0:17:39	when the band member found able to champagne
0:17:43	i about in groupness
0:17:46	okay so
0:17:47	i'm really ready on time
0:17:49	i was a so we studied
0:17:52	i nine in natural dialogue
0:17:55	this is
0:17:56	from our own corpus
0:17:58	our jump to
0:18:00	so we call it is
0:18:01	several levels things oneself
0:18:04	at the four level we had look at
0:18:06	whether laughter overlapped with speech there's temporal sequence where the laughter following of the laughter
0:18:11	we had no cats in some solo laughter not available
0:18:14	what incurs before during all have the laughable
0:18:17	in terms of semantic meaning we have quotas
0:18:20	the perceived arousal and the type of laughable
0:18:26	so in some way that we think there's in congress you or not
0:18:29	and in functions we are we have a roughly binary distinction between cooperative function
0:18:36	and non-cooperative function but i'll show you
0:18:38	a more detailed functions later
0:18:42	our job all of course i'm sorry about that's
0:18:45	so all we i don't i six hundred and around six hundred sixty laughter events
0:18:50	we found that it's very frequent the average
0:18:54	duration it's about
0:18:57	under two seconds
0:19:00	and we found that since a novel the most frequent have above is described events
0:19:06	followed by extra for a conventional something happening the physical context rather than the linguistic
0:19:11	context
0:19:12	followed by metalinguistic divan sounds for example if i is if a mispronounced a word
0:19:19	and we found that
0:19:21	there are more self produced laughable then part to produce laughable meaning is true that
0:19:27	will have more often about what we set ourselves
0:19:30	i still
0:19:31	that's laughter immediately for the laughable if the true that what laughter follows is what
0:19:36	laughters about well we actually found that there is the rubber free alignments between the
0:19:41	laughable i'm the laughter
0:19:42	so here is the other graph
0:19:44	channel plotting the distance between the end of the laughable and the beginning of the
0:19:49	laughter
0:19:49	and it's you can say they in the time peaks at zero
0:19:53	that's the average number actually below zero and there's the wide range
0:19:57	so laughter can really ago called long before the laughable
0:20:01	during the lovable and i and after all
0:20:05	we find that only thirty percent of laughter
0:20:09	happens immediately after the laughable
0:20:12	whereas but the majority of laughter the most frequently laughter happens during the laughable
0:20:20	okay in terms of context of form-related his we find that
0:20:25	dyadic laughter is very frequent meaning that was that one is very frequently your partner
0:20:30	withdrawing laughter
0:20:32	forty percent of laughter encoding immediately after we had the same time as partners laughter
0:20:37	and have to very frequently overlap with speech around forty to fifty percent of laughter
0:20:42	occur at the same time as speech
0:20:44	mean i we want to have an actual dialog system we need to be able
0:20:48	to generate speech
0:20:49	that's has like overnight laughter
0:20:54	i we found that laughter fast interrupt utterance of was very frequently we laugh with
0:20:59	someone else is speaking in the middle sentences
0:21:01	but sometimes we put a bit of laughter
0:21:05	we now on utterance
0:21:08	okay in terms of mean e
0:21:10	we found the majority of laughter actually have low arousal so they have low intensity
0:21:15	initial duration and arousal correlates with duration
0:21:19	and that the majority of the laughable each other to communicate recognition of incongruity
0:21:26	there's something concordance so for example nothing have to saying someone has found a bottle
0:21:32	of champagne that's signals that's finding a bottle of champagne this context is incompetent
0:21:41	so this is domain specific because these functions
0:21:45	are the most frequent function in our corpus because our proposed is natural
0:21:49	cooperative dialogue and we re well it's definitely not the case that this is of
0:21:55	full
0:21:56	functional taxonomy full or laughter in all kinds of dialogue but you know you know
0:22:01	that you know what was
0:22:03	the most frequent function for laughter is to show enjoyment
0:22:06	followed by
0:22:07	and there's a function called smoothing and soft ending so this is a well as
0:22:12	social function of laughter and
0:22:15	and there is show agreement
0:22:17	to mark funding this
0:22:18	to mark that will you about the same normally is funny and something called been
0:22:23	have anything to action so this function this a traumatised being
0:22:28	hence being it's being used in that you're already
0:22:31	and
0:22:32	i mean is
0:22:33	so a very common example been everything the action is rising something like could you
0:22:38	give me a couple of a coffee
0:22:40	data
0:22:41	that laughter i used to trigger happy not its meaning that i want you to
0:22:46	be close to me something like that okay
0:22:49	so
0:22:50	very quickly one minute
0:22:51	and
0:22:52	so we want to
0:22:54	we want to see
0:22:56	if we can use context one form related
0:23:01	features to predict all kinds characterize function
0:23:04	basically there is no single form
0:23:07	a related features that can characterize
0:23:10	and distinguishable kinda function
0:23:12	but we
0:23:14	and that specifically one thing is that french and training have very similar distributions and
0:23:20	it's very similar
0:23:22	signature in terms of these form-related yes
0:23:24	and that's
0:23:25	well as different as well
0:23:27	to show enjoyment is one of those frequent after that has normally
0:23:32	a wide range of duration but tend to be known duration
0:23:36	i been addressing the action and smoothing is often means that meaning that politeness related
0:23:41	laughter
0:23:42	time and
0:23:44	with low is
0:23:46	this is slightly less duration
0:23:49	and that happen it's was the end of the laughable
0:23:54	okay i think a rerun of running around trying to the detailed a please refer
0:23:58	to our favourite so i don't we can conclude
0:24:00	that's the we propose a semantic pragmatic accounts and thing which laughter is treated as
0:24:05	the gestural events and of all and that's in terms of data we found that
0:24:10	laughter
0:24:11	frequent speech floppies frequent
0:24:14	dyadic laughters the joint laughter is frequent we found that the distribution of laughter not
0:24:18	always rather free and only about thirty percent of laughter happening immediately after laughable
0:24:24	i is the majority of laughter is about incongruent stimuli
0:24:28	with low arousal
0:24:30	and that's a we found in a group was for frequent functions characterized by a
0:24:34	class of layers
0:24:36	curve form-related layers rather than a single layer
0:24:40	and guidepal frequency most patterns are very similar between in french and english french in
0:24:45	chinese
0:24:46	suggesting that have to behavior might be largely language and culture independent
0:24:51	a topics or stop here
0:24:54	okay
0:25:23	that's it happens already
0:25:25	most likely so i think two possibilities one is to do with incremental processing and
0:25:30	predictive processing right
0:25:32	so very of the we can predict what's
0:25:35	the end of the sensors
0:25:36	so very of the we know what exactly someone else to say
0:25:39	is about to say that's why i could be why we could laugh solely before
0:25:43	well the reference it of laughter is even finished
0:25:46	and the other one yes i think most definitely their ability gesture cues and also
0:25:52	on the visual expression account find that could indicates
0:25:57	that i'm about level you should love
0:25:59	we did capture body movement of fisher expression data using connect but we have analyzed
0:26:04	that yet
0:26:05	but as a good suggestion
0:26:14	and it wasn't the cases when laughing laughter happened right quite already is because you
0:26:20	can we can predict what about k
0:26:22	roughly all that someone has said mid sentence and we basically no the rest and
0:26:27	we start laughing
0:26:28	and that's you know like language every other aspects of language processing that is incremental
0:26:32	predictive
0:26:39	okay
0:26:41	no one is to study laughter acquisitions of the we found that and these the
0:26:46	optimal a three year old
0:26:48	the laughter pattern of a child is nowhere near
0:26:51	i don't so
0:26:53	and so we were thinking one thing is to link laughter behaviour is the only
0:26:58	indication of autistic
0:27:00	spectrum because
0:27:00	laughter although it seems to develop it develops lonely it is one of the only
0:27:05	is things but they used to it around three months
0:27:09	and we like to integrate a emotions we information state you know semantics would like
0:27:14	to study what the smile after differ only in scale or do they have different
0:27:18	functions and of course in laughter generation
0:27:44	it's true and also a lot of time people jointly complete the laughable so when
0:27:49	like someone is that how sent to the other one jumping in to complete this
0:27:53	and this will them while laughing
0:27:57	if you're image

When do we laugh?

Oral Session 6: Conversational phenomena and strategies

Ye Tian, Chiara Mazzocconi and Jonathan Ginzburg