Suppose we assert the proposition "bananas are green, brown or yellow."
Suppose we then discover a natural banana tree that has mauve banana's, but which produces toxic fruit. We may have difficulty deciding whether our proposition still stands. Is such a toxic banana still a banana? Did we mean that only edible bananas are green, brown or yellow? Has our proposition been falsified?
I think that precision in meaning is generally quite good when our proposition has a particular verificationist intent. If we were programming a robot to collect bananas for humans to eat, toxic bananas would not be bananas (edible fruits) for our intended purposes. In that case, our original proposition stands undefeated by the mauve fruits.
On the other hand, if we were categorizing plant species, we might consider toxic mauve bananas to be bananas in good standing* (by morphology, genetic similarity, potential for cross-breeding, etc.), thereby falsifying our original claim.
In these two cases, the intended mechanisms of verification make the clarification of meaning fairly straightforward.
However, things become much more shaky when our goal is extension of meaning beyond verification. If we don't explicitly specify how our proposition should be verified, the meaning becomes more uncertain. Mauve, toxic bananas either are or are not defeaters of the proposition depending on our verification protocols. The blanket statement that bananas are green, brown or yellow is highly uncertain without knowing precisely how we would verify the claim.
*One doesn't often use "bananas in good standing" in a sentence!