Edworthy, J., Hellier, E., Titchener, K., Naweed, A., & Roels, R. (2011). Heterogeneity in auditory alarm sets makes them easier to learn. International Journal of Industrial Ergonomics, 41(2), 136-146.
After all the reading I have done one thing appears to be a fact. Abstract alarms are harder to learn than everyday sounds, the sounds usually called “auditory icons”. But why? According to the authors, for a starter, everyday sounds are more heterogeneous, and so than can be easily learned not for semantic reasonds, but for psychoacoutic reasons. These sounds however – unlike mentioned on the previous post – can have problems in being faced as “alarms”, as people tend to be conservative as to what is an acceptable sound alarm.
In 1956 Miller suggested that the people’s ability to learn and retain information was positively correlated with the number of dimensions in which the item differed. Could this logic be applied to alarm sounds? If the number of dimensions that differ among sounds more sounds can be learned and retained? This was what the authors proposed to analyze in four different studies. The context was the rail industry.
The following table shows all seventeen existent alarm sounds. Some have different timbres, all have different meanings.
Participants were presented a pair of alarm sounds from the pool using loudspeakers. Their task was to rate how dissimilar s/he felt the two alarms were (confusability experiment) using a 10 item Likert-scale.
A hierarquical cluster analysis was made and the results demonstrated that most alarms are highly clustered, meaning that they are understood as being very similar.
The following study analyzed if the similarities between alarms found in #1 affected people’s ability to learn them. The procedure had a familiarisation and a learning phase. In the first phase, participants read the names of the alarms in cards. The experimenter showed a card and then played the corresponding alarm, and participants had to know the names of the alarms. Feedback was given to participants – correct/incorrect – correct response.
The results were not very positive: two alerts were recognised correctly 80% or more of the time; 9 alerts were recognised correctly more than 50% of the time and the remainder were recognised correctly less than 50% of the time, with some being recognised correctly less than 20% of the time.
For the third experiment a new set of sounds was designed including different classes of sound, including distinct temporal patterns in all alarms and including some auditory icons.
This experiment was similar to #1, but the results achieved were far better as alarms were less homogenous.
Again, experiment #2 was replicated with the new set of sounds.
The learnability of these new alarms was considerably higher than the set tested in Experiments 1 and 2. 7 were recognised correctly >80% of the time (three more than 90% of the time), and all but one were recognised >50% of the time, with the least recognised alarm falling just under the 50% mark.
The general discussion is very rish and should be closely read. What caught my attention was that auditory icons did not outperform abstract sounds, although they have obtained good results. The authors refer that one important area for future research is to determine whether the apparently easier learnability of auditory icons stems from their object-sound associations or from their acoustic heterogeneity. It is likely that the source is some combination from the two, but for design purposes it is important to knowas greater heterogeneity in abstract alarms seems also to enhance their learnability.