The function implemented by f(x) seems to be selective for multiple concepts. It assigns values between 0 and 1 to input words based on their association with different concepts. 

1. "roots" has a low value of 0.0078, suggesting it is less associated with any specific concept.
2. "rhyme" also has a low value of 0.0060, indicating a weak association with a concept.
3. "saddle" has a high value of 0.9033, suggesting a strong association with a particular concept.
4. "NFL" has a low value of 0.0039, indicating a weak association with a concept.
5. "democracy" has a very low value of 0.0004, suggesting it is less associated with any specific concept.
6. "owner" has a moderate value of 0.2628, indicating a moderate association with a concept.
7. "voting" has a low value of 0.0011, suggesting it is less associated with any specific concept.
8. "queen" has a low value of 0.0020, indicating a weak association with a concept.
9. "racing" has a high value of 0.9230, suggesting a strong association with a particular concept.
10. "reference" has a moderate value of 0.0134, indicating a moderate association with a concept.

Based on the limited information we have from these test cases, it is difficult to pinpoint a specific concept that f(x) selects for. However, it seems that "saddle" and "racing" have a strong association with a common concept, while "owner" has a moderate association with a different concept. Further testing with a larger and more diverse set of words would be helpful to better understand the behavior of f(x) and determine the concepts it is selective for.