Module A: Computational Modelling of Communicative Goals

Given a communicative goal – such as informing, narrating, summarizing, persuading, recommending, criticizing, complaining, etc.– as part of the input for the generation process, this project will propose a dynamic approach to automatically determine the plan and structure of the text to be generated.

The purpose of this module is to obtain multiple text planners – i.e. models for structuring and producing a wide variety of texts according to the communicative goal pursued–. To achieve this and obtain the aforementioned results, this module consists of two activities, which are next explained.

Objectives OBJ1, OBJ2 and OBJ3 will be fulfilled with this module.

Activity 1. Definition and characterization of the communicative goals

The objective of this activity is to analyze, define and characterize the communicative goal associated with a given text. First, an exhaustive analysis and review of the literature will be carried out to obtain a new taxonomy –based on the existing ones (Jakobson, 1960; Tedick, 2002; Hébert, 2011)- that considers the most relevant aspects and that can be addressed computationally. Subsequently, the specific characteristics and patterns of the language will be identified through its elements and expressions that are unequivocally associated with each of the defined communicative goals. To this end, NLP tools and resources will be used.

As a result of this task, a new taxonomy will be obtained to classify and characterize the communicative goals at a linguistic level, being also appropriate from a computational point of view. This knowledge will be used in activity A.2 for the automatic development of communicative language models and their subsequent materialization in text planners.

Milestone: taxonomy of communicative goals

Activity 2. Definition and obtaining of communicative language models. Text planners.

Once the communicative objectives have been defined and characterized, this activity consists of automatically learning the structure of various texts associated with a communicative goal. In other words, learning a communicative model of the language that will be materialized in a text planner. This corresponds to the macro-planning stage of the traditional NLG process, the purpose of which is to decide “what to say”. In the approach proposed in this project, the “what to say” will be at the level of structure and organization of the text, leaving the decision on the specific type of content to be included for the next module. The text planners will be obtained using corpus of documents associated to the same communicative objective.

This is a complex task, since, on the one hand, the same document may have a predominant communicative goal and its content may also contain other communicative goal. For example, in the case of hotel reviews, the predominant communicative goal may be to “complain” if the user did not like the hotel in which he or she stayed. However, a part of the review may contain text exposing their experience in the hotel. Thus, we would also find ourselves with the communicative sub-goal “inform” –”I stayed at this hotel in February 2017 because I had an event in this city […]”–. On the other hand, the expressive richness of a language and its constant evolution means that the way of writing changes with time, and a certain flexibility or variability is allowed in the type of structures present in the texts. This means that several texts on the same subject, sharing the same communicative goal, can be written in many ways, which constitutes an additional challenge for the project.

In order to carry out this activity successfully, machine learning models will be developed, including deep learning algorithms, in which specific characteristics relating to the characterization of communicative goals will be integrated –the intention, for example– to complement traditional linguistic characteristics –root or morphological category of the word, among others– in order to learn the unique and key aspects of each text that identify them with its communicative goal.

As a result of this task, the communicative language models obtained will be materialized in different text planners, and the resources and tools generated will be made available to the research community.

Milestone: obtaining text planners associated with communicative goals


  • Jakobson, R. (1960). Closing statements: Linguistics and Poetics, Style in language, T.A. Sebeok, New-York.
  • Tedick, D.J. (Ed.). (2002). Proficiency-oriented language instruction and assessment: A curriculum handbook for teachers. CARLA Working Paper Series. Minneapolis, MN: University of Minnesota, The Center for Advanced Research on Language Acquisition
  • Hébert, L. (2011). The Functions of Language. Signo.