(1) Indicator Functions -- here you have one text file per condition/task, including the Fix (Rest) condition. E.g., in the IR1 paradigm, there are three conditions (Fix, RealWord, FalseFont), so there will be three text files. In each file, there is one row for every fMRI scan. E.g., in the IR1 paradigm, there are 84 scans, so the RealWord text file will have 84 lines. Each line will have a 0, except for lines corresponding to the RealWord (or whatever) task, where the line will have a 1 instead. SPM will give you an opportunity to form contrasts between the various conditions.
(2) Explicit Regressors -- This is a little more complex. Here, you explicitly encode the contrasts of interest in the text file. E.g., in the IR1 paradigm, there are equal numbers of RealWord scans as there are FalseFont scans, so for the contrast "RealWord-FalseFont" place a 1 for RealWord scans and a -1 for FalseFont scans. Place a 0 for Fix scans. You'll probably want to also include an orthogonal regressor for the contrast "(RealWord and FalseFont) - Fix", where there is a -2 for Fix scans and a 1 for either RealWord and FalseFont scans.
If there had been a differing number of RealWord and FalseFont scans, it's a little trickier. Say there were N RealWord scans and M FalseFont scans. Then you could place a M for RealWord scans and a -N for FalseFont scans. You want the sum of all of the numbers in the text file to equal zero.