The abstract describes work on learning probabilistic logic programs, a form of declarative representation that captures relational structure and supports neurosymbolic reasoning. These programs encode dependencies as weighted compositional rules, offering interpretability alongside expressive power.
The central problem addressed is how to induce such programs from data. The authors propose a method that uses functional gradient guided language models, but the abstract is truncated and does not specify the algorithm, experiments, or results.
Because the source is only a partial abstract, this summary is limited to the stated motivation and approach. No further claims can be made from the available text.