WEBVTT 00:00:00.000 --> 00:00:13.280 So, yeah, that was a nice introduction. I just had a bit more context. So I'm a linguist 00:00:13.280 --> 00:00:18.720 and specifically I'm a discourse analyst. So I'm interested in how language, primarily, 00:00:18.720 --> 00:00:25.220 but also other modes of communication like imagery and visuals, help to really construct 00:00:25.220 --> 00:00:29.520 and shape our social realities. And today I'm going to talk about some work that I've 00:00:29.520 --> 00:00:34.880 been doing looking at the language and images that we use when we talk about and communicate 00:00:34.880 --> 00:00:40.280 about dementia. I'm going to talk about this work because I think it provides a nice demonstration 00:00:40.280 --> 00:00:46.840 of the digitalized method but also the digitalized data. So I'll get back to that idea shortly. 00:00:46.840 --> 00:00:54.000 Okay, it works. So I want to start with a bit of context about dementia itself. So dementia 00:00:54.000 --> 00:00:58.960 is a syndrome characterized by progressive deterioration in the brain and this leads 00:00:58.960 --> 00:01:04.260 to symptoms like memory loss and problems with thinking and coordination. The term 00:01:04.260 --> 00:01:09.360 dementia doesn't refer to one specific disease. Many people would use it in that way but it's 00:01:09.360 --> 00:01:17.560 actually an umbrella term as sort of demonstrated by this visual from Kate Swaffer. So it's 00:01:17.560 --> 00:01:22.400 an umbrella term and there are actually over 100 different types of dementia. And the most 00:01:22.400 --> 00:01:27.200 common type is Alzheimer's disease which I think most people have heard of but there 00:01:27.440 --> 00:01:32.560 are lots of other types too. Dementia is a major global public health challenge. So in 00:01:32.560 --> 00:01:38.080 the UK which is the context that I live and work in, dementia currently affects around 00:01:38.080 --> 00:01:44.520 900,000 people but that figure is estimated to rise to 1.6 million people by the year 00:01:44.520 --> 00:01:51.120 2040. There's presently no known cure for dementia. So it's a chronic condition and 00:01:51.120 --> 00:01:54.640 that means that people who develop it and are diagnosed with it will eventually die 00:01:54.800 --> 00:02:00.360 with it as well. However, and importantly for a point I'll make a bit later, while 00:02:00.360 --> 00:02:05.800 people die with dementia they don't actually die of it. The most common cause of death 00:02:05.800 --> 00:02:12.400 in people with dementia is actually pneumonia. So the work I'm presenting today is from the 00:02:12.400 --> 00:02:18.480 public project called Public Discourses of Dementia. So the project's funded by UKRI 00:02:18.480 --> 00:02:24.280 which is the main research funder in the UK and it's based at Lancaster University. 00:02:24.520 --> 00:02:30.920 There are two of us working on the project. I'm the one on the left obviously. So I'm 00:02:30.920 --> 00:02:37.640 the PI and I'm working with the brilliant Emma Putland who's the Senior Research Associate 00:02:37.640 --> 00:02:43.040 on the project. And in this work we're exploring how dementia is represented across a wide 00:02:43.040 --> 00:02:48.240 range of contexts of public communication focusing as I've mentioned on language and 00:02:48.240 --> 00:02:53.400 imagery. So we're looking at things like the news, social media, online support groups, 00:02:53.400 --> 00:02:58.240 public health campaigns, the language that charities use and most recently we've been 00:02:58.240 --> 00:03:02.080 looking at generative AI and that's something I'll talk about later in this session. 00:03:03.720 --> 00:03:09.160 Importantly we're working with stakeholders from the media and from charities and also 00:03:09.160 --> 00:03:15.120 people from public health all with the aim of developing linguistically informed context 00:03:15.120 --> 00:03:20.120 sensitive guidance and training that we can provide to those different stakeholders to 00:03:20.120 --> 00:03:24.600 really help them to communicate about dementia in ways that reduce stigma and reduce the 00:03:24.600 --> 00:03:28.480 shame around it but also promote some genuine awareness. 00:03:30.480 --> 00:03:34.800 So the first question I want to consider today to demonstrate what some of the work we're 00:03:34.800 --> 00:03:40.160 doing looks like is how is dementia represented in UK newspapers? And here I want to focus 00:03:40.160 --> 00:03:46.160 in particular on how newspapers use metaphorical language to talk about dementia. And in this 00:03:46.200 --> 00:03:53.800 work I've drawn on I guess the digital as method and specifically corpus linguistics. 00:03:55.320 --> 00:04:00.880 So corpus linguistics for those, by a show of hands of people, are people aware of corpus 00:04:00.880 --> 00:04:05.800 linguistics? Do people know? I'm curious who's heard of corpus linguistics? Oh great. 00:04:05.880 --> 00:04:11.040 OK I can do this slide pretty quickly. There are still some that aren't initiated but yeah. 00:04:11.200 --> 00:04:15.680 So corpus linguistics is essentially a collection of methods for studying linguistic patterns in 00:04:15.680 --> 00:04:22.040 very large collections of naturally occurring language use or corpora. This language data 00:04:22.040 --> 00:04:26.440 is stored in a digital format which means we can examine it using computer software 00:04:26.440 --> 00:04:31.240 and because it relies on computer software in this way corpus linguistics methods are 00:04:31.240 --> 00:04:36.120 quite well equipped to deal with very large data sets, digitised data sets, typically 00:04:36.120 --> 00:04:42.960 millions of words, sometimes billions of words. So corpus is an electronic collection of naturally 00:04:43.000 --> 00:04:47.240 occurring text which have been selected to be representative of a particular language, 00:04:47.240 --> 00:04:54.640 a language variety and the words I've highlighted here signal some of the main features of a 00:04:54.640 --> 00:05:00.960 corpus. I won't go into lots of detail about what corpora are now but I can provide some 00:05:00.960 --> 00:05:05.840 further reading for those interested but this is probably enough for you to understand the 00:05:05.840 --> 00:05:09.960 approach that we're taking. But essentially when we're looking at corpora and we're using 00:05:09.960 --> 00:05:14.240 corpus linguistics techniques we're interested in patterns, we're interested in frequency 00:05:14.240 --> 00:05:19.640 but we're interested in patterning so things that happen a lot or sometimes things that 00:05:19.640 --> 00:05:24.520 don't happen a lot that run contrary to a pattern across a corpus, across a collection 00:05:24.520 --> 00:05:31.520 of texts. So for the analysis that I'm talking about here I compiled a corpus containing 00:05:31.760 --> 00:05:38.760 all UK national tabloid news articles mentioning either the word dementia or dementia as once 00:05:40.760 --> 00:05:46.560 in their headline and or three or more times throughout the article. I sampled articles 00:05:46.560 --> 00:05:53.560 over a period of ten years between 2010 and 2019 so that was 6,751 articles which came 00:05:54.360 --> 00:06:00.560 to just under four million words of data. And the analytical approach that I took to 00:06:00.600 --> 00:06:06.360 identifying and interpreting the use of metaphor in this data was sort of a mixed methodology 00:06:06.360 --> 00:06:10.600 which drew together techniques from corpus linguistics as I've mentioned but also frame 00:06:10.600 --> 00:06:16.520 analysis and critical discourse studies. So I started by using the corpus linguistic technique 00:06:16.520 --> 00:06:22.600 of collocation analysis to identify metaphorical patterns. Now collocation is essentially the 00:06:22.600 --> 00:06:29.080 phenomenon where two words tend to occur together in very close textual proximity and do so 00:06:29.080 --> 00:06:34.920 very often or with some frequency. And in this process what happens is that words can 00:06:34.920 --> 00:06:40.120 become associated with each other in language users' minds and these associations develop 00:06:40.120 --> 00:06:46.680 through exposure over time. And corpus linguistics techniques can help us to identify these kinds 00:06:46.680 --> 00:06:52.040 of patterns, these collocations by searching for words that frequently occur together across 00:06:52.120 --> 00:06:59.120 large collections of text across large corpora like my news corpus, corpus of news articles 00:07:00.160 --> 00:07:05.800 about dementia. And so in this study I used collocation analysis to look at words that 00:07:05.800 --> 00:07:12.200 frequently occurred alongside the word dementia in the articles in order to identify frequent 00:07:12.200 --> 00:07:17.680 uses of metaphorical language. And there's a bit of technical information there at the 00:07:17.680 --> 00:07:24.680 bottom about how I did that. It's not so important but for those who are interested that's there. 00:07:25.200 --> 00:07:31.660 So having analysed those collocation patterns manually to identify metaphorical patterns 00:07:31.660 --> 00:07:36.880 I then interpreted those collocates which did tend to be used metaphorically in terms 00:07:36.880 --> 00:07:41.760 of the frames that they contributed towards. So here I draw on Goffman's notion of frames 00:07:41.760 --> 00:07:47.120 as organising principles which help us to make sense of a thing and to represent it 00:07:47.120 --> 00:07:51.040 by foregrounding some elements and backgrounding others. So I was interested in the framing 00:07:51.040 --> 00:07:56.240 function that these metaphors performed. Through the framing analysis I was able to group the 00:07:56.240 --> 00:08:01.320 metaphors according to the particular aspects of dementia that they represented. And these 00:08:01.320 --> 00:08:06.840 were dementia's prevalence, its causes, its symptoms and prognosis, the lived experience 00:08:06.840 --> 00:08:13.840 of dementia and then responses to it. So how as a society we should deal with this challenge. 00:08:14.120 --> 00:08:19.260 And then my analysis importantly was also critical. So I wasn't interested in just describing 00:08:19.260 --> 00:08:25.640 these patterns but I also wanted to interpret the metaphor and their framing functions as 00:08:25.640 --> 00:08:31.640 being both constitutive of and constituted by their contexts. So I was interested in 00:08:31.640 --> 00:08:36.680 contexts of production for example. So how the metaphorical framings might serve the 00:08:36.680 --> 00:08:42.000 commercial or ideological aims of the newspapers I was investigating. In terms of the broader 00:08:42.000 --> 00:08:47.080 socio-cultural context I wanted to consider how the framings might reflect but also shape 00:08:47.080 --> 00:08:52.840 societal experiences, understandings and indeed actions in relation to dementia. I think a 00:08:52.840 --> 00:08:59.840 ghost has come in. Oh it's left. Okay. So onto some analysis. Beginning with dementia's 00:09:00.160 --> 00:09:04.800 prevalence this was framed using metaphors belonging to a diverse range of domains but 00:09:04.800 --> 00:09:09.960 all of them conveyed a sense in which the prevalence of dementia was high and also rising. 00:09:09.960 --> 00:09:14.280 So most frequently the articles drew on the domain of sports and competition to frame 00:09:14.280 --> 00:09:19.960 dementia as the leading cause of death in the country. This is indicated in uses of 00:09:19.960 --> 00:09:25.240 collocates like leading, leader, overtaken and overtook which frequently occurred alongside 00:09:25.240 --> 00:09:30.880 the word dementia. For example in this extract at the top we see dementia is described as 00:09:30.880 --> 00:09:36.260 overtaking heart disease to become the leading cause of death and it also knocks heart disease 00:09:36.260 --> 00:09:42.060 off the top of the table like the premier league of diseases or something. Violence 00:09:42.060 --> 00:09:47.540 metaphors were also used to frame dementia's prevalence where its incidence could be quantified 00:09:47.540 --> 00:09:52.900 in terms of the number of people who had been hit by it or what its death toll was. Related 00:09:52.900 --> 00:09:57.660 to the violence metaphor dementia was also construed as a weapon so a bomb or a time 00:09:57.660 --> 00:10:03.100 bomb which I think foregrounds dementia's not only its rising prevalence but also implies 00:10:03.100 --> 00:10:09.340 that this would continue to grow to disastrous effect. And dementia's growing prevalence 00:10:09.340 --> 00:10:14.620 was also rendered as an epidemic and then around a fifth of cases as a silent epidemic. 00:10:14.620 --> 00:10:19.580 Now at first glance the use of that kind of language might not appear metaphorical because 00:10:19.580 --> 00:10:25.980 we talk about health problems as epidemics but sort of belongs to the domain of illness. 00:10:25.980 --> 00:10:30.340 However strictly speaking the term epidemic actually denotes the spread of infectious diseases 00:10:30.540 --> 00:10:35.460 which dementia certainly is not. And when we look at general language use like Michaela 00:10:35.460 --> 00:10:40.220 did this morning I consulted the British National Corpus and found that when we talk about epidemics 00:10:40.220 --> 00:10:46.820 we usually do this to talk about some kind of negatively evaluated phenomenon so things 00:10:46.820 --> 00:10:53.820 like crime or knife crime in particular and the illegal use of drugs things of that nature. 00:10:53.820 --> 00:10:57.980 So like the ticking time bomb metaphor then the epidemic metaphor helps to provide that 00:10:58.020 --> 00:11:04.380 kind of negative evaluation of dementia prevalence. In terms of its causes these could be framed 00:11:04.380 --> 00:11:09.000 using a domain of movement with certain behaviours and risk factors being construed as speeding 00:11:09.000 --> 00:11:16.000 up the development of dementia so we get collocates like accelerate, accelerated and speeds. As 00:11:16.940 --> 00:11:21.380 well as focusing on the environmental risks associated with dementia the newspapers also 00:11:21.380 --> 00:11:27.020 reported on certain biological factors particularly protein variants and parts of the brain to 00:11:27.060 --> 00:11:32.900 which dementia onset had been attributed. So these risks could be framed in violent 00:11:32.900 --> 00:11:38.580 terms where parts of the brain attack other parts of the brain. So a striking example 00:11:38.580 --> 00:11:44.540 of this is the notion of rogue cells or the dementia dagger cell which is a part of the 00:11:44.540 --> 00:11:49.740 brain which then attacks other parts of the brain's cause dementia. And as this bottom 00:11:49.740 --> 00:11:56.180 example shows these kinds of creative uses of language aren't necessarily invented by 00:11:56.180 --> 00:12:00.900 the newspapers themselves but sometimes they're intertextual or interdiscursive they borrow 00:12:00.900 --> 00:12:05.620 them from other sources in this case it's from actually from an academic journal article 00:12:05.620 --> 00:12:11.260 so from the genre of science writing. So we're beginning to see a prevalence of violence 00:12:11.260 --> 00:12:16.300 and militaristic tropes and this was particularly the case for how the newspapers framed dementia 00:12:16.300 --> 00:12:22.060 symptoms and its prognosis. In these cases dementia is anthropomorphised so it's constructed 00:12:22.420 --> 00:12:28.820 as another social actor and one which inflicts forms of violence on those who are diagnosed 00:12:28.820 --> 00:12:34.420 with it so dementia could be construed as a thief which robs and steals people of their 00:12:34.420 --> 00:12:41.140 memories their cognitive function and even their personhood. Other symptoms are framed 00:12:41.140 --> 00:12:46.220 as forms of violence in more generic terms so we get dementia described as something 00:12:46.220 --> 00:12:51.580 which ravages and strikes. In terms of the prognosis the newspapers frequently foreground 00:12:51.580 --> 00:12:56.420 a fatalistic view of dementia so they frame it as a murderer that kills people diagnosed 00:12:56.420 --> 00:13:02.740 with it so essentially if you get dementia it's a death sentence it will kill. And then 00:13:02.740 --> 00:13:08.300 this fatalistic framing is supported again by some kind of anthropomorphic characterisation 00:13:08.300 --> 00:13:13.060 of dementia which gives it qualities that imply it to be not only a violent actor but 00:13:13.060 --> 00:13:20.060 one that inflicts violence willfully and deliberately so we get words like brutally cruel, cruelest, 00:13:20.060 --> 00:13:27.060 heartless and savage. Consistent with the metaphorical framing of dementia as a cruel 00:13:27.140 --> 00:13:32.580 and brutal actor that inflicts violence the lived experience of dementia could also be 00:13:32.580 --> 00:13:37.500 characterised in terms of conflict so people with the condition are framed as battling, 00:13:37.500 --> 00:13:43.580 facing and struggling with dementia. Other constructions were less agentive though on 00:13:43.580 --> 00:13:48.300 the part of people with dementia so they don't even put up a struggle or a fight. In those 00:13:48.300 --> 00:13:54.060 cases people could be described as victims or a victim of dementia or being described 00:13:54.060 --> 00:14:01.060 as being in dementia's grip or hold without much resistance or being struck by it passively. 00:14:01.260 --> 00:14:07.140 Meanwhile the framing of people with dementia as succumbing or having succumbed to dementia 00:14:07.140 --> 00:14:13.180 arguably implies an initial struggle or resistance but an eventual failure or even surrender 00:14:13.300 --> 00:14:19.500 as we see in this example here. Another set of metaphors which similarly frames people 00:14:19.500 --> 00:14:25.500 with dementia as victims are found in stories which focus in particular on cases where they've 00:14:25.500 --> 00:14:32.500 been abandoned by their relatives so in this case we get people with dementia equated to 00:14:32.820 --> 00:14:37.940 kind of waste that's being disposed of so we get people with dementia being described 00:14:37.940 --> 00:14:44.940 as being dumped, dumped in care homes, dumped in hospitals, in some cases dumped on the 00:14:44.940 --> 00:14:50.540 street. And the negative appraisal of life with dementia is also conveyed through metaphors 00:14:50.540 --> 00:14:55.500 which frame the experience of it in horrific or even apocalyptic terms so we get life with 00:14:55.500 --> 00:15:02.500 dementia described as hell, a horror movie, a nightmare. And we also get more subtle kind 00:15:03.500 --> 00:15:10.180 of constructions in that same way through the collocate descent which is a kind of spatial 00:15:10.180 --> 00:15:16.100 metaphor which constructs life with dementia as downward movement into the condition and 00:15:16.100 --> 00:15:21.420 draws on that broader association that we have of things going down being bad or getting 00:15:21.420 --> 00:15:28.420 worse. Not all metaphors necessarily imply this kind of negative outlook on life with 00:15:28.580 --> 00:15:34.220 dementia though or even foreground the victimhood of people diagnosed with it. And a particularly 00:15:34.220 --> 00:15:41.020 frequent kind of counter example to that trend is the characterisation of the relationship 00:15:41.020 --> 00:15:45.820 between dementia and the person with it as one of companionship. So this is indicated 00:15:45.820 --> 00:15:50.860 in uses of the collocates live and living. And this companionship metaphor I think is 00:15:50.860 --> 00:15:55.640 more flexible than those that I've described so far because it could be used to describe 00:15:55.680 --> 00:15:59.920 the relationship that a person has with dementia in more or less harmonious terms as these 00:15:59.920 --> 00:16:04.600 two examples show. So sometimes life with dementia is very difficult but sometimes it's 00:16:04.600 --> 00:16:08.800 not that difficult and dementia isn't the thing that defines a person's life and the 00:16:08.800 --> 00:16:15.800 uses of these metaphors can help to convey that. And likewise the metaphorical framing 00:16:17.800 --> 00:16:21.420 of dementia as a journey could be used to appraise life with the syndrome in more or 00:16:21.420 --> 00:16:26.180 less positive terms as well as to foreground mixed experiences in dementia's progressive 00:16:26.180 --> 00:16:29.860 nature so it's actually quite helpful from an awareness raising perspective because the 00:16:29.860 --> 00:16:36.860 experience of the syndrome changes over time. Finally now, last set of metaphors I promise, 00:16:38.660 --> 00:16:43.540 how do we respond to dementia in society? What must we do? Well the most pervasive metaphorical 00:16:43.540 --> 00:16:49.440 domain in this category was again violence and here we see the response to dementia framed 00:16:49.480 --> 00:16:54.240 as a war or a battle or a mission against the syndrome and this is one usually that 00:16:54.240 --> 00:17:00.320 the whole of society is involved in. And in the majority of cases the key to victory in 00:17:00.320 --> 00:17:06.040 this war is the development of pharmacological interventions so essentially a cure or some 00:17:06.040 --> 00:17:12.840 kind of drug. Relatedly the promise of studies and pharmacological trials could be framed 00:17:12.840 --> 00:17:18.160 as having the potential to beat dementia in a race or to slow, stop or reverse dementia's 00:17:18.160 --> 00:17:23.680 movement. While other metaphors imply that there is a cure for dementia somewhere out 00:17:23.680 --> 00:17:29.800 there in the world and that it exists, it just has yet to be identified by scientists 00:17:29.800 --> 00:17:34.880 so these scientists are undertaking a quest or a hunt or looking for clues to find the 00:17:34.880 --> 00:17:41.880 key to curing the syndrome. So although the vast majority of metaphors focused on biomedical 00:17:42.360 --> 00:17:47.280 response to dementia, so the development of drugs and curative treatments, in a small 00:17:47.320 --> 00:17:53.560 minority of cases the articles focused instead on social care so how we support people living 00:17:53.560 --> 00:17:59.480 with dementia in the here and now. And this was particularly the case in the use of gambling 00:17:59.480 --> 00:18:05.240 metaphors so we get the quality of social care critiqued in newspapers through gambling 00:18:05.240 --> 00:18:11.120 metaphors so social care is a postcode lottery so it varies a lot where you live but it's 00:18:11.120 --> 00:18:15.760 a lottery you don't know whether or not you'll get good care or even a Russian roulette 00:18:15.800 --> 00:18:20.760 regarding how well prepared your local hospitals are to deal with you if you have dementia 00:18:20.760 --> 00:18:27.160 and again that's also a violence metaphor because Russian roulette can be quite violent. 00:18:27.160 --> 00:18:32.200 So this has been a bit of a whistle stop tour of a wider programme of work but I hope I've 00:18:32.200 --> 00:18:38.580 been able to illustrate some of the more prominent patterns in dementia press representation. 00:18:38.580 --> 00:18:43.080 So there are a wide range of tropes with the particular prominence of anthropomorphising 00:18:43.120 --> 00:18:48.280 metaphors which equate dementia variously to a companion, a competitor but most commonly 00:18:48.280 --> 00:18:55.280 actually to a violent killer. Almost all metaphors foreground the agency of dementia, this is 00:18:55.800 --> 00:18:59.080 something that came up earlier today this question of agency and I think it's particularly 00:18:59.080 --> 00:19:05.160 important here. In many cases we see dementia anthropomorphised as a social actor, the disease 00:19:05.160 --> 00:19:12.080 as an actor and those same metaphors background or preclude altogether the possibility that 00:19:12.080 --> 00:19:17.680 the person living with the syndrome can have agency in this process. As an agentive force 00:19:17.680 --> 00:19:24.080 dementia moves and invades, it races against and hides from those seeking to cure it and 00:19:24.080 --> 00:19:29.280 it competes with other conditions to become the leading cause of death in the country. 00:19:29.280 --> 00:19:33.320 People with dementia meanwhile are largely denied agency, they tend instead to be rendered 00:19:33.320 --> 00:19:38.760 as the target of violent actions such as murder and theft. 00:19:38.760 --> 00:19:44.280 Now a lot has been written about violent metaphors and broadly they have the potential 00:19:44.280 --> 00:19:51.280 to stigmatise a disease or an illness or any health topic by framing those affected by 00:19:51.280 --> 00:19:58.780 it as the enemy of some kind of war and their bodies are then rendered as metaphorical battlefields 00:19:58.780 --> 00:20:02.960 and we saw this in how the press could frame the parts of the brain as attacking other 00:20:02.960 --> 00:20:07.880 parts of the brain for example. A further issue with violent metaphors is that if the 00:20:07.880 --> 00:20:12.400 disease prevails and the person fails to recover then those affected are often viewed 00:20:12.400 --> 00:20:16.560 as having lost that metaphorical battle and in the case of dementia as we don't have a 00:20:16.560 --> 00:20:22.760 cure this is at present a battle that people will inevitably lose. 00:20:22.760 --> 00:20:28.360 But dementia stigma is not propagated just through metaphors of violence, we also saw 00:20:28.360 --> 00:20:35.240 the nightmarish scenarios, the apocalyptic scenarios and lots of other patterns which 00:20:35.280 --> 00:20:41.960 broadly foreground dementia's most disastrous and fear inducing aspects while backgrounding 00:20:41.960 --> 00:20:47.240 if not shutting down altogether the possibility of living well with dementia which actually 00:20:47.240 --> 00:20:53.240 many people do as long as they have the right support around them. 00:20:53.240 --> 00:20:56.720 It's not all bad news though, I know it's a bit doom and gloom and the weather seems 00:20:56.720 --> 00:21:00.440 to have gotten darker since I started talking about this as well, maybe it will get brighter 00:21:00.440 --> 00:21:06.160 when Marianna starts talking. But some of the metaphors might actually be more helpful 00:21:06.160 --> 00:21:11.240 in terms of reducing stigma and raising awareness so those which frame the experience of dementia 00:21:11.240 --> 00:21:16.800 as a journey or which characterise it as companionship respectively articulate the progressive and 00:21:16.800 --> 00:21:22.160 chronic nature of the syndrome but they could also be used more flexibly in foreground 00:21:22.160 --> 00:21:28.120 not only negative but also positive aspects of life for somebody experiencing dementia 00:21:28.240 --> 00:21:33.000 I think this is important because people living with this syndrome experience it differently 00:21:33.000 --> 00:21:38.640 and no two days are the same. 00:21:38.640 --> 00:21:43.920 So the second part of my talk which I promise is a lot shorter don't we? 00:21:43.920 --> 00:21:46.960 Sabina getting scared on the front row! 00:21:46.960 --> 00:21:51.320 Yeah it's much shorter. So having used digital methods to look at how newspapers talk about 00:21:51.320 --> 00:21:55.640 dementia and I want to switch my focus to the digitalised data and here I want to look 00:21:55.640 --> 00:22:01.120 at how AI generated imagery might represent dementia visually so I'm also flipping from 00:22:01.120 --> 00:22:05.600 the textual to the visual or the linguistic I should say to the visual. And this is something 00:22:05.600 --> 00:22:10.440 I got interested in as some dementia charities and advocacy groups that I've been collaborating 00:22:10.440 --> 00:22:16.080 with had indicated that generative AI tools could actually be really useful for them in 00:22:16.080 --> 00:22:21.600 terms of crafting and creating text and so my instinct was to think about okay that's 00:22:21.600 --> 00:22:25.800 great for them in a practical sense but how could this be used critically and what as 00:22:25.800 --> 00:22:31.360 a linguist and a discourse analyst could I do to help them to get to that point? 00:22:31.360 --> 00:22:36.860 So for a bit of context just as the invention of the printing press revolutionised the sharing 00:22:36.860 --> 00:22:43.540 of information through printed media today generative AI is increasingly influencing 00:22:43.540 --> 00:22:50.360 the ways in which we communicate and here what happens is that a user gives a textual 00:22:50.360 --> 00:22:57.840 prompt to an AI tool and then that tool can generate text and also in many cases images 00:22:57.840 --> 00:23:04.240 in response and it does this having been trained on a huge database of language and or in this 00:23:04.240 --> 00:23:11.800 case captioned images. Now text to image generation is an increasingly 00:23:11.800 --> 00:23:17.600 accessible and popular way of creating visuals for professional text producers. For instance 00:23:17.600 --> 00:23:22.920 stock image banks like Getty and Shutterstock are now offering users the option to generate 00:23:22.920 --> 00:23:27.120 their own images as well as to buy pre-made ones so you don't just have to buy an image 00:23:27.120 --> 00:23:32.240 from Getty you can say these are all rubbish I want one that does this so get an AI tool 00:23:32.240 --> 00:23:38.360 to make you one that does just that and this is significant as image banks are a leading 00:23:38.360 --> 00:23:43.960 force in producing and disseminating visual representations of all kinds of social phenomena 00:23:44.280 --> 00:23:48.240 including the kinds of images actually that we find in newspapers and other mass consumption 00:23:48.240 --> 00:23:55.320 texts. Now many ethical debates currently surround AI and AI generated imagery and greater 00:23:55.320 --> 00:24:01.320 critical literacy is definitely needed at all levels. Increasingly realistic AI generated 00:24:01.320 --> 00:24:06.800 images and videos risk worsening misinformation with potentially world altering health, social 00:24:06.800 --> 00:24:13.640 and political ramifications and to exemplify that I provided here perhaps infamous example 00:24:13.960 --> 00:24:19.640 of a false image that did the rounds a couple of years ago of Donald Trump being arrested 00:24:19.640 --> 00:24:26.640 and this was generated using an AI image generation tool but it did the rounds on social media 00:24:26.640 --> 00:24:30.600 of course because of the High Court judgment we know that this isn't actually possible 00:24:30.600 --> 00:24:37.600 so it must be a false image. AI bias is another central concern so research has shown that 00:24:37.600 --> 00:24:44.440 AI can not only reflect but actually amplify pre-existing human stereotypes in terms of 00:24:44.440 --> 00:24:49.680 how it represents individuals from particular backgrounds. So at the bottom of this slide 00:24:49.680 --> 00:24:54.840 here I've included a couple of examples from a study carried out by Bianchi and colleagues 00:24:54.840 --> 00:25:01.840 where what they did was they typed in words like poor into an AI image generation tool 00:25:02.040 --> 00:25:09.040 and they found that the word poor was particularly likely to elicit images of black people while 00:25:09.040 --> 00:25:15.040 the word attractive tended to elicit images of white people. Meanwhile occupations they 00:25:15.040 --> 00:25:20.040 also found were gendered with roles like software engineer being associated with men and housekeeper 00:25:20.040 --> 00:25:27.040 being associated with women. And the same can be said for these for other kinds of bias 00:25:27.040 --> 00:25:34.040 as well. So Bianchi and colleagues found persistent visual stereotypes when they searched across 00:25:34.040 --> 00:25:40.040 different tools. They found even that altering the text prompt to generate less oppressive 00:25:40.040 --> 00:25:46.040 imagery often didn't mitigate racist, sexist and ableist biases and again these biases 00:25:46.040 --> 00:25:52.040 often intersected with one another. So for example the authors found that typing in a 00:25:52.040 --> 00:25:57.040 blonde woman leading a meeting worked. Okay great it wasn't just men leading a meeting 00:25:57.040 --> 00:26:03.040 and even blonde women could lead a meeting in an AI image tool. Fantastic. But if they 00:26:03.040 --> 00:26:09.040 replace the word blonde with the word disabled then the visually disabled woman in the image 00:26:09.040 --> 00:26:13.040 is shown instead listening to another woman leading the meeting rather than actually 00:26:13.040 --> 00:26:19.040 leading that meeting herself. So there's something happening algorithmically which means that 00:26:20.040 --> 00:26:26.040 these tools essentially can't generate an image of a woman in a wheelchair leading a 00:26:26.040 --> 00:26:31.040 meeting and that's because the online data that these tools are trained on simply don't 00:26:31.040 --> 00:26:36.040 have enough of those kinds of representations out there in the world for these tools to 00:26:36.040 --> 00:26:42.040 draw from and use when they're creating imagery. Okay so what we wanted to do in our analysis 00:26:42.040 --> 00:26:47.040 was to see whether certain kinds of bias or stigma around dementia would also emerge if 00:26:48.040 --> 00:26:53.040 we used this kind of tool to create images of people with dementia. And for this we used 00:26:53.040 --> 00:26:59.040 a tool called stable diffusion and using this tool we created 171 images using the simple 00:26:59.040 --> 00:27:04.040 prompt dementia we kept it very simple. I've also displayed some of the settings that we 00:27:04.040 --> 00:27:09.040 used to create the images here in case people are interested. This isn't really my bag, 00:27:09.040 --> 00:27:14.040 that's something that one of my collaborators did, he's into this and he understands it 00:27:14.040 --> 00:27:20.040 in industry practices. I don't really, I'm coming at it as a fairly naive discourse analyst. 00:27:20.040 --> 00:27:25.040 But looking at the images generated we were first of all interested in who was shown as 00:27:25.040 --> 00:27:32.040 having dementia and there were clear stereotypes in this regard. Notably the person was shown 00:27:32.040 --> 00:27:38.040 to have dementia tends overwhelmingly to be white with clear signs of older age, that's 00:27:39.040 --> 00:27:45.040 very visible. For example having visible wrinkles and white hair. But this sort of ignores the 00:27:45.040 --> 00:27:50.040 reality of the many people living with early onset dementia which can actually start in 00:27:50.040 --> 00:27:56.040 your 40s and sometimes even in your 30s. There's a similar lack of diversity in the characteristics 00:27:56.040 --> 00:28:02.040 attributed to those individuals who tended overwhelmingly to be shown alone in isolation 00:28:03.040 --> 00:28:09.040 and in passive and immobile poses with expressions or gestures conveying sadness, suffering or 00:28:09.040 --> 00:28:15.040 a sense even of being vacant. These figures also tend to be set against empty backgrounds 00:28:15.040 --> 00:28:21.040 that provide little to no context about the individuals and they also tend to look off 00:28:21.040 --> 00:28:27.040 camera or have their eyes closed even and I've indicated in these cases using arrows and 00:28:27.040 --> 00:28:32.040 so arrows show the direction of the eye gaze and a cross shows where people have their 00:28:32.040 --> 00:28:38.040 eyes closed. Now I'd argue that this distinct lack of eye contact mimics real life in that 00:28:38.040 --> 00:28:43.040 not looking towards the viewer encourages viewers to observe these individuals rather 00:28:43.040 --> 00:28:49.040 than to seek to connect with them as individuals or even on an emotional level. So these kinds 00:28:49.040 --> 00:28:55.040 of averted gazes help instead to create a sense of social distance between us as the 00:28:55.040 --> 00:29:03.040 viewers and the represented participants i.e. the people with dementia. The main way that 00:29:03.040 --> 00:29:07.040 people living with dementia are contextualised when that happens is through their brains 00:29:07.040 --> 00:29:13.040 and certain neural processes which are presented in these cases as being abnormal. So in 00:29:13.040 --> 00:29:19.040 foregrounding this biological even neurological focus the images again necessarily background 00:29:19.040 --> 00:29:26.040 the individual. There's no possibility for an even simulated life or kind of life world 00:29:26.040 --> 00:29:33.040 beyond the defective brain and any non-medical aspects of life with dementia for that matter. 00:29:33.040 --> 00:29:38.040 Instead the people are reduced to their brains and to neurological processes and faulty ones 00:29:38.040 --> 00:29:44.040 at that. A number of the images show dementia metaphorically, don't worry I won't bang on 00:29:44.040 --> 00:29:50.040 about metaphor for too much longer, but using visual metaphors of a tree. Trees are widely 00:29:50.040 --> 00:29:56.040 regarded as a symbol of life and an image of seasonal change. Thank you Sabina, that's 00:29:56.040 --> 00:30:02.040 perfect. Here then it's notable that these trees are consistently bare when they're shown 00:30:02.040 --> 00:30:09.040 in context of dementia. Elsewhere the lost leaves for these bare trees have been associated 00:30:09.040 --> 00:30:14.040 with lost brain cells, lost memories or even a loss of personal qualities. This notion 00:30:14.040 --> 00:30:19.040 of dementia as a tree is quite a common trope, it's interesting to see these tools reproduce 00:30:19.040 --> 00:30:26.040 it here. So if the life cycle can be regarded as a metaphorical seasonal cycle with spring 00:30:26.040 --> 00:30:32.040 as birth and then these bare trees signal a personal winter for the person living with 00:30:32.040 --> 00:30:38.040 dementia and of course that's the season widely associated with bleakness and death. And this 00:30:38.040 --> 00:30:44.040 living death connotation I'd argue is enhanced by the dullness and blandness and lifelessness 00:30:44.040 --> 00:30:49.040 of the colour palette here. So we tend to see that it favours blues, beiges, greys and blacks 00:30:49.040 --> 00:30:57.040 over more vibrant colours. Okay so in summary the AI generated images reproduce the kinds 00:30:57.040 --> 00:31:03.040 of stereotypes of people living with dementia that we've observed early including in the 00:31:03.040 --> 00:31:07.040 kinds of news text that I showed you earlier in this talk. And I'd argue that these kinds 00:31:07.040 --> 00:31:12.040 of depictions can therefore draw on and reinforce the stigma and shame that currently surrounds 00:31:12.040 --> 00:31:19.040 dementia. So there's clear evidence that the images generated by tools like stable diffusion 00:31:19.040 --> 00:31:25.040 reproduce existing discourses and existing stereotypes. In this case they emphasise old 00:31:25.040 --> 00:31:31.040 age, the faultiness and fragmentation of people's minds and brains and they focus on loss of 00:31:31.040 --> 00:31:36.040 abilities and loss of self. Suffering is also emphasised and people with dementia are distanced 00:31:36.040 --> 00:31:42.040 from others throughout. Taken together these patterns contribute to a sense of deathliness 00:31:42.040 --> 00:31:47.040 and reproduce the metaphorical choke that people with dementia and people with dementia 00:31:47.040 --> 00:31:53.040 are the living dead. And importantly I've highlighted the word reproduce here as these 00:31:53.040 --> 00:31:59.040 images don't constrict new visual discourses but simply reproduce or even recycle precisely 00:31:59.040 --> 00:32:05.040 the kinds of representations that surround dementia on the data on which these tools 00:32:05.040 --> 00:32:11.040 are trained. Such representations are of course partial. They offer just one way of looking 00:32:11.040 --> 00:32:15.040 at dementia. And on the other hand there's a distinct lack of consideration here for 00:32:15.040 --> 00:32:21.040 individuals' unique personalities and the focus on deficit and what is lost leads to 00:32:21.040 --> 00:32:26.040 a lack of attention on what kinds of abilities might remain following a dementia diagnosis. 00:32:26.040 --> 00:32:31.040 There are other perhaps more productive and life-affirming ways of looking at dementia 00:32:31.040 --> 00:32:35.040 or discourses around it that aren't really visible at all in AI generated data and that's 00:32:35.040 --> 00:32:41.040 also true for linguistic textual data that I've generated using these tools in other projects. 00:32:41.040 --> 00:32:48.040 Instead what could be possible is foregrounding the diversity of people who have dementia 00:32:48.040 --> 00:32:53.040 or fostering a sense of connectedness or the sense that life can continue after a dementia 00:32:54.040 --> 00:32:59.040 diagnosis. Thank you. That people aren't reduced to their suffering and that dementia can help 00:32:59.040 --> 00:33:05.040 them to critically reflect on existing social norms and allow us to do that as a society 00:33:05.040 --> 00:33:11.040 and to reflect on existing discourses. These opportunities are lost. Now at this point 00:33:11.040 --> 00:33:15.040 it's important to recognise that there were some exceptions to the overall trends I've 00:33:15.040 --> 00:33:21.040 talked about. For instance a few of the AI generated images, three of which I've put 00:33:21.040 --> 00:33:26.040 on this slide, provide alternative portrayals that instead emphasise connection and activity 00:33:26.040 --> 00:33:33.040 and so overall suggest the potential for more diverse representation. And to end on a more 00:33:33.040 --> 00:33:38.040 hopeful note, I realise this has been such a bleak talk, I'm so sorry everyone, goodness, 00:33:38.040 --> 00:33:44.040 as academics and I think as a society as a whole there's scope for us to explore how 00:33:44.040 --> 00:33:49.040 tools like generative AI tools like stable diffusion can be used for human good in the 00:33:50.040 --> 00:33:55.040 future. However I think right now this needs some working out and it certainly requires 00:33:55.040 --> 00:34:01.040 some change. Indeed generative AI is continuously developing and the kind of work I've presented 00:34:01.040 --> 00:34:07.040 here aims to be part of a broader critical conversation that highlights the socio-political 00:34:07.040 --> 00:34:12.040 effects of such emerging technologies and the kind of emerging genres that they bring 00:34:12.040 --> 00:34:18.040 with them. Whether we like it or not, it seems like generative AI certainly isn't going 00:34:18.040 --> 00:34:23.040 anywhere, so as a society I think it's imperative that we better understand the implications 00:34:23.040 --> 00:34:29.040 of using it. And as researchers in humanities disciplines and as socially oriented researchers 00:34:29.040 --> 00:34:36.040 in general, we have a key role to play in this. As well in general as working towards 00:34:36.040 --> 00:34:43.040 the development of more inclusive, ethical and importantly critical approaches to envisioning, 00:34:43.040 --> 00:34:50.040 and using these kinds of tools in the future. And here I can just about stop. So there are 00:34:50.040 --> 00:34:55.040 just some references from the work that I've talked about today and if you're interested 00:34:55.040 --> 00:35:00.040 more in corpus linguistics and how it can be used to look at health topics and the health 00:35:00.040 --> 00:35:06.040 humanities broadly, this book I recently authored with Luke Collins provides a lot of context 00:35:06.040 --> 00:35:11.040 on the approach. Yeah, and so there I can stop. Thank you for listening. 00:35:13.040 --> 00:35:21.040 Music