I think you could use a mapping approach but you will need some degree of manual work because you need to define the final mapping-results. This means for each possible search-value must exists a valid return-value.
It's not mandatory necessary to create such mapping manually else the search-values might be derived from the final values, for example:
temp:
load *, mid(F, 1, L - iterno() + 1) as R while iterno() <= L;
load *, len(F) as L;
load 'TOMATO' as F, autogenerate 1;
which applies per internal while-loop multiple value-parts in extra records. Such logic might be applied several times to run it backwards and forwards and/or to pick values in between and might be also combined with additionally conditions in regard to the min/max lengths (returning just a single char or twice won't be sensible).
With a similar logic you may also consider any typos by reversing chars against each other and/or replacing them with similar chars. Of course this won't be trivial but you may start it with just a few iterations.
Very important to create such listing is to add recno(), rowno() and iterno() as additionally fields because they are needed to sort the mapping-table. A mapping returned always the first matching and therefore you will look at first for: TOMATO before you look for TOM. Further for the above you couldn't use applymap() else mapsubstring() which won't directly return you final value else it will place the final value within calling-string and needs to be afterwards extracted, for example with textbetween() on additionally wrapped chars around your final values.
Mapsubstring() has further the benefit that it could replace multiple values within a string which might be also helpful to rank and sort the matches.
Beside this it might be useful not to apply everything within a single mapping else you may use multiple ones - again to increase the quality of the matching and the ranking. Another point which may minimize the needed efforts might be not to do this against a concatenated string of 3 fields else against the single fields.
If it looked like a quite heavy work - it is. But I think there aren't many alternatives and if they won't be really easier at least not to derive a valid final value. Nevertheless you may look on features like:
LevenshteinDist - script and chart function | Qlik Cloud Help
