Por João António Rodrigues.
Argument mining is the automatic identification and classification of arguments in natural language expressions.
An argument conveys some idea or proposed action, in order for its holder to persuade, be understood or solve a problem.
It is structured in a claim (aka conclusion) and premises (aka reasons or evidence; and also enthymemes when they are not explicit) that justify or refute the claim, as in the following example:
Proponents of immigration maintain that, [Premise: according to Article 13 of the Universal Declaration of Human Rights, everyone has the right to leave or enter a country, along with movement within it] ...
Some argue that [Claim: the freedom of movement both within and between countries is a basic human right], and that the restrictive immigration policies, typical of nation-state, violate this human right of freedom of movement.
To handle computationally an argument we need to represent its semantics.
Current natural language processing techniques have been successful at capturing the meaning of words and sentences with distributional semantics representations (aka embeddings) together with different neural network architectures. Such techniques rely heavily on the type and amount of source data.
Transfer learning techniques acquire semantic representations from different sources and tasks to leverage target tasks with less available data.
Given that argument mining data is scarce, I will present the quantitative results of novel experimentations with distributional semantics and transfer learning techniques to leverage and solve argument mining tasks.
More information: https://moodle.ciencias.ulisboa.pt/course/view.php?id=2964#section-8
Transmissão em direto via Zoom.
