Ever since we have been talking to machines, language has become the universal interface of Artificial Intelligence, and those who can build systems that understand and generate language have joined the European list of the most needed profiles. But language carries culture, variety and inequality with it: the craft of the AI Natural Language Processing Engineer is technical and, at the same time, inevitably humanistic.
Every era of computing has had its interface: the command line, the window, the touchscreen. The current one speaks. And behind every system that understands, interprets or generates human language lies the work of the AI Natural Language Processing Engineer, the professional of Natural Language Processing (NLP) who follows projects from the analysis of requirements to the development of models and on to operation.
A discipline with deep academic roots
NLP was not born with generative models. The current definition is offered by the course of the Hugging Face platform, a reference for the community: “NLP is a field of linguistics and machine learning focused on understanding everything related to human language” (Hugging Face); linguistics and machine learning together. Stanford University’s course devoted to the subject trains “the necessary skills to design, implement, and understand their own neural network models, using the Pytorch framework” (Stanford CS224N), and the textbook by Jurafsky and Martin remains the conceptual map of the discipline. The classic tasks – text classification, entity recognition, question answering, translation, summarisation – coexist today with the building of applications founded on Large Language Models (LLMs), between the adaptation of pre-trained models and Retrieval-Augmented Generation techniques.
The humanistic side of a technical craft
What distinguishes this profile from the other AI crafts is the nature of the material: language is not a datum like the others. It carries variety, registers, dialects, cultures, and the inequalities of those who speak it. From this follow three practical responsibilities. The first concerns data: the construction and annotation of corpora with documented criteria, and awareness of what each corpus includes and excludes. The second concerns evaluation: metrics relevant to the task, tests on different linguistic varieties and domains, the measurement of hallucinations in generative systems and of biases towards languages and groups of speakers; a system that works only for standard English is a faulty system even when the averages absolve it. The third concerns integration: transparency towards users who interact with conversational systems, an obligation that Regulation (EU) 2024/1689 (the AI Act) devotes precisely to systems intended to interact with people and to generated content, and the safeguarding of personal data in texts under Regulation (EU) 2016/679 (the GDPR), because free text is, of personal data, the most undisciplined vehicle.
Named, in full, in the European needs list
Market demand, for once, need not be deduced: it is written. The skills-needs analysis of the EU co-funded Artificial Intelligence Skills Alliance (ARISA) names the profile in full among the most needed: “The AI practitioners’ roles that are needed most are data scientists, data engineers and especially machine learning engineers including NLP engineers and computer vision engineers”. And the growth in demand for AI competences is a phenomenon that cuts across the European markets: Stanford University’s AI Index records, on data from the analytics provider Lightcast, shares of postings with AI skills that in Luxembourg, at 3.4 per cent, and Spain, at 3.3 per cent, exceed the United States itself, at 2.6.
The Italian benchmark
The UNI 11621-8:2026 standard recognises the AI Natural Language Processing Engineer as a standalone profile, with competences set out according to the methodology of the European e-Competence Framework, the UNI EN 16234-1 standard; the assessment and certification of professional competences take place at bodies accredited under the UNI CEI EN ISO/IEC 17024 standard, within the framework of Accredia Information Circular DC No. 21/2026.
Conclusions
The AI Natural Language Processing Engineer works on the most human material there is with the most recent tools there are: this is why the craft demands, together, engineering rigour and linguistic sensitivity. Now that language is the very interface of Artificial Intelligence, this double competence is not an academic luxury: it is what separates the systems that understand from the systems that seem to understand.
In the light of the above, one wonders whether the rush to conversational applications will preserve this double sensitivity, or whether haste will hand users brilliant interfaces that treat badly the languages, and the people, less represented in the data.
