Many semantic taggers are trained on valency encoding resources such as FrameNet or PropBank. But what if the roles were reversed and semantic taggers were needed to mine valency patterns in the first place? While semantic taggers are not needed to detect complements like the subject and object in Estonian, they are indispensable for differentiating the semantic types of adverbials. In this paper, we test the possibilities of two strategies for annotating Estonian spatial adverbials for valency pattern mining. First, we use LLMs to automatically tag physical locations in a test set of 1000 nominal adverbials with added example sentences. GPT-4o was able to receive an F-score of 0.85, with most mistakes being attributed to words denoting organisations having both a physical and abstract locational meaning depending on the context. In the second experiment, we use existing semantic type annotation from a dictionary to automatically find instances where all of a particular verb’s adverbial dependents in a specific spatial case are locations. These could then be used to automatically annotate all of a verb’s dependents as locations by only knowing their case. We found that out of 21,267 verb-case pairs in the sample, 18,8% had a single dominant semantic type with an additional 10% needing minimal manual verification before use. Our findings indicated that both of the aforementioned methods can be used to annotate other semantic types in further work.