About me – Valerio Basile – Perspectivist Semanticist
Much of modern Natural Language Processing, as well as other subfields of Artificial Intelligence, is based on some form of supervised learning. Since when the rule-based systems have been overcome by statistical models, we have seen Hidden Markov Models, Support Vector Machines, Convolutional and Recurrent Neural Networks, and more recently Transformer Networks each replacing the previous state of the art. In a way or another, all these models learn from data produced by humans, crowdsourced or otherwise.
Let us begin with a quick primer on how linguistic annotation is traditionally conducted. The basic components are the following:
With these premises, the act of annotating a set is an iterative process, where each annotator expresses their judgment about the target phenomenon on an instance at a time, in the modalities defined by the annotation scheme.
About me – Valerio Basile – Perspectivist Semanticist