Traditional Named Entity Recognition (NER) metrics rely on syntactic span and type matching. While this may be appropriate if NER is framed as a sequence labeling task, the advent of generative models allows for alternative approaches, where model outputs are not extracted, but generated. In this perspective, generative NER calls for more semantically-aware scoring mechanisms. Generative models may produce valid aliases, paraphrases, or type variants that differ from the annotated gold, while also producing unsupported predictions. We propose a taxonomy of such cases and introduce an embedding-based parametric method to evaluate generative NER based on three metrics: soft precision, soft recall and unsupported rate. We validate this framework using standard and noisy datasets, across state-of-the-art systems such as GPT and GLiNER alongside older encoder-based models. We find that our metrics mostly agree with traditional ones on better-established, less noisy benchmarks, but differ in more recent, noisier ones, and can capture edge cases that arise from free-form generation.

Knowledge Graphs (KGs) are a practical representation of information that combines structural flexibility with expressiveness. With the recent advent of Large Language Models (LLMs), it has become possible to extract KGs from plain text. However, extracting facts from text is only one part of building a usable knowledge graph. A construction system must also reconcile entity mentions, control its relation vocabulary, integrate information across passages, and preserve a traceable connection between graph facts and their source. We present LMKG, a modular LLM-based pipeline that supports both open and ontology-constrained extraction. We also propose two construction strategies for long documents: fold, which accumulates a graph over successive chunks, and enrich, which combines whole-document and chunk-level extraction before graph refinement. We evaluate LMKG on three standard benchmarks: Text2KGBench, MINE-1, and HotpotQA. We find that our ontology conditioning substantially improves conformance to a given ontology in Text2KGBench. Our system consistently recovers facts, reaching the highest reported fact recovery of 90.80% in MINE-1. We observe degraded downstream question-answering performance on HotpotQA when supplying graphs rather than full passages.