The kind of complexity that you are presumably referring to is the number of inflected and derived forms that can be formed from a single root. There are many more forms of the verb, or the noun, in Latin than there are in Spanish, French or Italian, and the same is true of the relationship of Old Norse to Norwegian, Sanskrit to Hindi, Ancient Greek to Modern Greek. The ancestoral Proto-Indo-European language had a very rich system of grammar whereby many semantic properties (such as the fact of being male, singular, indirect object) are encoded in a single word. Languages change over time, which is by definition "evolution".
Language change is not unidirection in this respect. The proto-Bantu language had a limited amount of inflection and derivation with mere dozens of forms, more comparable to what you find in Romance languages, but many later individual language have vastly complicated especially the verb system to that you can often derive millions of forms from a single root and in the case of Shona it exceeds 100 trillion forms per root. This happened by compressing former phrases into single words. On the one hand, you need fewer words to express a given meaning; on the other hand, the rules for building words become more complex. The problem with attempts to discern a direction of "simplification" in language is that there are numerous cross-cutting kinds of "complexity", often at cross-purposes.