Understanding the meaning of words is crucial for many tasks that involve\nhuman-machine interaction. This has been tackled by research in Word Sense\nDisambiguation (WSD) in the Natural Language Processing (NLP) field. Recently,\nWSD and many other NLP tasks have taken advantage of embeddings-based\nrepresentation of words, sentences, and documents. However, when it comes to\nWSD, most embeddings models suffer from ambiguity as they do not capture the\ndifferent possible meanings of the words. Even when they do, the list of\npossible meanings for a word (sense inventory) has to be known in advance at\ntraining time to be included in the embeddings space. Unfortunately, there are\nsituations in which such a sense inventory is not known in advance (e.g., an\nontology selected at run-time), or it evolves with time and its status diverges\nfrom the one at training time. This hampers the use of embeddings models for\nWSD. Furthermore, traditional WSD techniques do not perform well in situations\nin which the available linguistic information is very scarce, such as the case\nof keyword-based queries. In this paper, we propose an approach to keyword\ndisambiguation which grounds on a semantic relatedness between words and senses\nprovided by an external inventory (ontology) that is not known at training\ntime. Building on previous works, we present a semantic relatedness measure\nthat uses word embeddings, and explore different disambiguation algorithms to\nalso exploit both word and sentence representations. Experimental results show\nthat this approach achieves results comparable with the state of the art when\napplied for WSD, without training for a particular domain.\n