Presumably, you have good reason for using a graphic code for this rather than words or numbers, (e.g., to facilitate easy scanning through a lot of data for certain levels of probability). Assuming your users (correctly) conceptualize probability as a quantitative proportion, here’s a few of thoughts:
Mini-pie chart, being a circle with some portion filled in to represent the probability.
Quarter filled = very unlikely (but hardly impossible).
Half filled = as likely as it is unlikely.
Three-quarter filled = very likely (but hardly certain).
Mini bar graph with one rectangle filled to indicate the degree of probability. This is better than the pie chart when you’re showing actual probability values, not ordinal quanta like “very unlikely” – “about 50-50” – “very likely.”
Splatter frame, being a square with a scatter of dots, to suggest randomness more than a pie or bar.
A few dots = very unlikely
Half-filled with dots = as likely as it is unlikely.
Three-quarter filled with dots (maybe some dots overlapping other dots) = very likely.
Shade coding, sort of the same idea as a spatter frame, but more suitable for small icons, especially icons that code other information (e.g., by their size or shape). Avoid three shades of gray –they may be hard to tell apart. Instead:
White (maybe with one small black dot to suggest some uncertainty) = very unlikely
Gray = as likely as it is unlikely
Black (maybe with one small white dot) = very likely.
Crispness coding, where degree of blur implies low probability. This is good when you are trying to represent the uncertainty about another value (e.g., by the blurriness of a frame around the variable’s value, or by blurring the ends of a bar in a bar chart).
Color, which can be used redundantly with shade to maximize accessibility. This is good if there is an emotional valence associated with the probability values. Stick with simple population stereotypes when choosing colors (e.g., green, yellow, red).
For general principles on graphically coding variables see my post Putting the G into GUI.
Graphic codes are not necessarily self-documenting, so be prepared to provide some, perhaps as a legend or as redundant text for each value. This could be the actual probability value, if it is known, or ranked word phrases like you suggest. Both your graphic code and your word phrases may need some testing. The phrases you provide for example may have different meaning for different people and contexts. Is “quite likely” more likely than equally likely and unlikely, or just not particularly unlikely? Is “not very likely,” still likely, just not very likely? If so, is it more or less likely than “quite likely”?