Negative Measure

Built to solve problems.  List deficiencies.  Map crises.  The ICTs for conflict management aggregate the negative and forget to leave a space for the positive.

In the survey I devised to collect descriptive information about a video scene my experiment participants watched, I followed the models of several other ICTs for conflict management and collected information about the individuals perpetrating the actions, the location, the level of damage inflicted, and the level of insecurity participants observed.  However, when I compared the structured answers with oral descriptions, participants often spent time detailing the involvement of bystanders.  Did people offer help?  There seemed to be an expectation of community intervention to calm a situation.  Also, participants were measured in their consideration of the guilt or innocence of the perpetrator.  They offered more than one explanation for the scenario they watched so as to place the motivations, culpability, or even the justification for involvement in mild violence into doubt.

The categories of perpetrator and victim, villain and target were not delineated in the same way as I expected.  The core conceptualization of the event as a 'problem' may in fact be the problem.  This is the initial premise for taking the report.  We want to learn more about it (the problem), about its components, its actors, locations, moving parts, so we can design a solution and prevent its re-occurrence.  What if the local population doesn't perceive a problem?  Or what if they understand the maladjusted components in a manner that is undetectable, or conceptually invisible, with the current ICT approach?  I think it's a matter of the wrong model, not the wrong impulse to improve.

From my experiment, I asked individuals to identify 'the attacker,' the person hitting another man about the head and chasing him through the scene.  Most took this to mean 'who is causing the problem?'  And they identified the man I would have called 'the victim.'  Moreover, several individuals told me they came to this conclusion because this problem-causer/victim was not fighting back but being chased and hit while offering no defense.  This meant he was guilty.  For me, this meant he was in need of help.  This model for recognizing justice, cause-effect, and culpability was foreign to me.  It would be worth doing more experiments around just this concept (perhaps conceptual transfer experiments) and sampling more than just me.

Is it just a matter of thinking up better names for categories?  Better questions to ask on surveys?  It's more than an issue of gathering information quantitatively or qualitatively.

Take for example the issue of the egg, the spear, and the egg-water.  Context is key for meaning.  This is true for any language.  In Luo, tong means spear and egg depending on tone, depending on context.  (The phrase tong pii means 'clean water' in Ethiopian Anuak, but 'egg water' in Kenyan Luo, two closely related Nilotic languages.  Although with negotiation, the phrases could mean water for eggs in either language.  So that's funny.)  The thing about tone and context is that they are reliant on a speaker-listener interaction, a volley, an exchange, a non-solo communication act.  Not like writing.

"Hope" is the thing with feathers—

That perches in the soul—

And sings the tune without the words—

And never stops—at all—.... E. Dickinson, (not a Luo).  

Much of our communication tech has evolved to capture, facilitate, speed and streamline writing, an individualistic form of expression.  It simply can't convey a type of communication with an essential, reverberative quality in which semantic content is as much (or more?) tied to the speaker-listener relationship as it is to anything that can be captured with text. 

Yes, these tools are meant to increase our ability to go into the field and gather information from 1000 individuals instead of 30.  This is great for researchers and polling and participatory governance and all sorts of reasons, but only if the tool is a good tool, that is, it assists us in doing a task we are already doing, makes it simpler, faster, easier in some way.... but if instead it brings us speedily to the wrong results, then what good is it?

The problem comes from the the fact that the tools being used now were built to capture western narratives (or logical constructs, conceptualizations of events) and communicate among NGO staff after disasters.  These same tools have been only slightly modified and then redeployed for conflict use such as post-reconstruction governance surveys, violence reports for election monitoring, etc.  All the while not recognizing the new users have new needs.  New conceptualizations of the events they are describing (and new ways of linking them) may be the key to empowering locally driven solutions and disengaging externally mandated ones.

Original and Extra...Complicated

Field experience as a cultural mediator led me to the hypothesis about how ICT software was capturing narratives.  In the experiment, participants watched a video and described it three times:

stage 1 out loud in Acholi
stage 2 written on a mobile device in Acholi
stage 3 out loud in English

I predicted a pattern of similarities between the the written Acholi version and the oral English version.  I thought that because the software had been developed from a western cultural perspective, that even when the interface was translated for users, they would have to get into an English mindset to engage with the logical organization of the software.  And because of this effort at a cognitive level, the narrative they would produce via the ICT would be more similar to the oral English version and less similar to the oral Acholi version.  In this way it could be said that the Acholi narrative was disrupted.

hypothesis: that narrative given via ICT in Acholi would be more similar to performance in English than to oral Acholi.  [stated as a null hypothesis: The oral Acholi version will resemble the ICT Acholi version]

The apples and oranges bit that makes statistical modeling a challenge is two-fold.  First, the experiments I adapted were comparing only speech (similar to my stage 1 and stage 3).... more like apples to apples.  And secondly, each of their participants provided similar narrative content, at least, similar enough to compare... so  here, oranges to oranges.  In my case, even though I controlled the topic, the video, one participant remembered seeing a dispute instigated by a woman while another remembered seeing a group of government thugs enforcing curfew.  Furthermore, some participants provided only a few phrases while others talked for several minutes, many paragraphs of data.  Their descriptions of the video clip are barely in the same galaxy. 

The plan was to compare each narrative stage by analyzing frames.  A frame refers almost to a fragment or phrase constructing the narrative of, in this case, an event.  Take for example the frame, I saw fighting  and I saw two men fighting.  While similar, these would be considered two different ways of framing because of the added detail of the number of actors, the two men. 

The frames, such as giving a general overview followed by reference to who was in the frame-- I saw fighting. I saw people fighting-- would be considered two frames.  If the details in these frames are not markedly different between any two stages, then they are counted as the same even if the syntax is different such as I saw a fight vs. They are fighting. 

When differences occur, they are not only counted but ranked for importance.  Based on the narrative structure in Acholi (used as a baseline/norm and derived from this study as well as literature), the general category, then the participants, then details of actions and location are given hierarchical importance.  So for example, if two versions match until a detail of location, if one mentions a location and the other does not, they may still be considered a match if the general scene category and description of participants and actions match.  If the locations differ entirely, one event is recalled as being in a market and one at a taxi stage, then they will likely be considered different narratives.  For shorter narratives, most if not all frames and details must match.  

Another consideration is tone.  Throughout stage 1 and 3, the space that an oral modality gives for adding extra words, words which convey the (un)certainty of the speaker about his/her memory play a role in strength of the witness statement.  However, both stage 2a and b do not afford this same linguistic wiggle room.  There were built-in points to express doubt in stage 2b, and they are a point of focus in analysis.  Stage 2a was an open SMS format, so it was up to the participant to inject the intended level of (un)certainty.  Divergence in tone, in expressed level of doubt, constitutes a means for differentiating narratives as well. 
 
For stage 2 (a and b), it gets a bit trickier to compare written results with spoken... and sometimes only a ticked box for yes or no.  But this is the meat of the question.  Does the ICT format adequately capture the categories and concepts for the event frames? (Many participants in my experiment were reluctant to hold the device, type themselves, and often spoke the answers aloud which I recorded with an audio device and typed for them.  The data collected via the mobile device in stage 2 was not, therefore, strictly written.  Much of it was still spoken.)

For stage 2b to be judged a match with another stage, the answers should convey similar information to that given in one of the oral narratives without contradicting or adding new information.  If the reader had only the information from stage 2b, could s/he reasonably imagine a scene as told in one of the other narrative versions?  If one of the oral narrative versions has several other frames or details not captured by stage 2b, then no, but if the sense of the event framing such as ‘theft with violence’ was captured by the answers to stage 2b’s question series, then it can be reasonable concluded that stage 2b adequately captured that narrative.  (If it sounds complicated, you're not wrong.  As I have said before, this was a first attempt at bringing a problem to the surface so it can be better understood, studied, and addressed... so far, lots of good indicators plus heaps of ideas to improve future trials.)

Because of the ordering of the stages, stage 2a often repeated much of what was said in stage 1.  For psychology experiments, the effects of ordering are considerable and are usually countered by mixing up the stages; however, in this case, the role of priming was anticipated.  The structure of the experiment captures the real-life field experience where an individual recalls an event orally and then creates a written record of that event with ICT software, often with the help of a trained information collection specialist.  Priming effects would counter my hypothesis, in other words, I almost stacked the deck in favor of my own hypothesis failing.  So if the results show my hypothesis still has value, there must be a very strong force at work.

Of the instances that were considered 'ICT success,' most came from participants who followed the stage 2a path.  This may have been because it was a quick experiment and easy to simply repeat in written form what you had just said out loud.  It is fresh in your mind.  The differences came in the form of tone as well as more obvious detail and frame shifts.  The tone change is interesting because in the oral account, the speaker may have said, I saw some fighting like maybe there was this guy who caused a problem or maybe it's a taxi conductor wanting change....  There was doubt and alternatives expressed.  In stage 2a, the version was clear and direct.  I saw one guy hit another guy in the street.  The versions often make it sound much more like the participant is sure of the identity and ready to accuse someone to the police; whereas the oral version is not accusatory and gives a couple of possible scenarios for what prompted the chaos.  If I only read the ICT version I would believe there was an actionable threat, a report of violence that needed a response.  If I heard the oral version, I would not be as concerned.  The aggregation of this type of report, stripped of the measured tone of the speaker can heighten perceived threat levels unnecessarily.  

I will spend more time on this topic in further posts as well as talk about the results of my frame analysis.  (Two other colleagues are reviewing my results. Independent evaluation.)  But preliminary conclusions are that my hypothesis was correct.  So that's cool.  Other results point towards the strong division between oral and written narrative structures, something we all kinda know intuitively and there is a ton of research on how speaking is different than writing, so it's reassuring that results stayed with this pattern... the interesting parts come in looking at