Bringing NL Back with P: Defending Linguistic Methods in NLP for

Future AI Applications

Fabio Meroni

1,2 a

X23 – Science in Society, Treviglio, Italy

Università degli Studi di Bergamo, Bergamo, Italy

Keywords: Natural Language Processing, Computational Linguistics, Artificial Intelligence, Natural Language Formalisation,

Lexicon-Grammar, NooJ.

Abstract: In the development of Artificial Intelligence (AI) tools for Natural Language Processing (NLP) applications,

this position paper promotes the (re)introduction of rule-based, linguistically informed methodologies, with

particular attention to addressing the challenges posed by low-resource languages and research ethics when it

comes to the enhancement of machine intelligence by means of linguistic intelligence. NLP, as a rapidly

evolving subfield of AI, has seen a proliferation of contributions in recent years. However, the predominant

reliance on statistically driven approaches has reduced NLP to a pursuit of superficial aesthetic results,

neglecting the foundational linguistic structures that underpin natural language processing. Consequently, the

marginalization of linguists within the field has stalled progress toward a deeper understanding of Natural

Language Formalization (NLF). Without targeted intervention, these issues threaten to persist, undermining

the potential of NLP to achieve its full intellectual and practical promise. This paper argues for a renewed

integration of the science of natural language (NL) into its processing (P) within an interdisciplinary

framework that emphasizes collaboration between computational linguists and AI researchers, and presents a

methodological proposition of a possible way to include linguistic resources in a richly informed AI

application using NooJ.

1 INTRODUCTION

This position paper advocates for the inclusion of

rule-based linguistic tools in the development of new

Natural Language Processing (NLP) applications.

This approach is not forcedly intended to replace AI-

driven tools, but rather to complement them by

integrating an additional layer of functionality that

accounts for the description of actual linguistic

resources, for the sake of grammaticality and

scientific rigour. After offering an overview of the

state-of-the-art in theoretical and methodological

approaches to NLP, this contribution presents how

computational tools rooted in “linguistic methods”

(Silberztein, 2024) bring valuable insights and grant

a critical foundation for the development of linguistic

intelligence, a key component in achieving the

broader objective of “true” artificial intelligence. By

combining these methodologies, the following

paragraphs show how the industry can envisage

building more robust and versatile NLP applications

https://orcid.org/0009-0005-5853-6005

that leverage the strengths of linguistic theory when

applied to computational formalisation, in order to

achieve higher flexibility in a number of language

processing tasks including text generation, spell-

checking, automatic Part Of Speech (POS) tagging

and machine translation. This proposal insists on how

NooJ, as a software and a methodological staple

rooted in lexicon-grammar (Gross, 1994), can

provide an alternative perspective by emphasising

language processing in a representational way rather

than a purely statistical one, supporting the linguistic

approach to NLP first and foremost when

approaching the study of low-resource languages.

2 TROUBLED TIMES OF NLP

Stating that NLP primarily aims to formalise the

complexities of natural language may appear so self-

evident that one might choose to omit it altogether

when speaking about this domain of studies. One could

1062

Meroni, F.

Bringing NL Back with P: Defending Linguistic Methods in NLP for Future AI Applications.

DOI: 10.5220/0013259100003890

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 3, pages 1062-1068

ISBN: 978-989-758-737-5; ISSN: 2184-433X

argue that acknowledging this foundational goal is

crucial, as it highlights the inherent challenges

involved in research accounts made in this field: by

recognising the importance of Natural Language

Formalisation (NLF), researchers and practitioners

should underscore the interdisciplinary nature of NLP,

which integrates elements from linguistics and

computer science to create models that approximate

human understanding, communication, and reasoning.

However, recent accounts in the history of this

discipline have been talking about a sort of paradigm

shift: after a first phase of absolute reliance on

linguistic modelling, a second (ongoing) wave sees

AI taking the driving seat, marking a “renaissance” of

NLP (see at least Fanni et al., 2023 and Jiang et al.,

2023, p. 2). As a consequence, NLP methods that find

their very basis in the linguistically-accurate

description of phenomena occurring in natural

language instantiations and productions are now

widely judged to be outdated if not even obsolete.

According to the retelling of Anitha S. Pillai and

Roberto Tedesco (2024):

[...] especially after the advent of very

powerful conversational agents able to

simulate a human being and interact with

the user in a very convincing way, AI and

the historical field of Natural Language

Processing almost become synonymous

(think of the abilities of GPT-3-derived

models; for example, ChatGPT1). (Pillai &

Tedesco, 2024, p. vii).

The same two authors proceed with a very bold

statement when it comes to their diachronic overview

of research in this field:

Historically, NLP approaches took

inspiration from two very different research

fields: linguistics and computer science; in

particular, linguistics was adopted to

provide the theoretical basis on which to

develop algorithms trying to transfer the

insight of the theory to practical tasks.

Unfortunately, this process proved to be

quite difficult, as theories were typically too

abstract to be implemented as an effective

algorithm. On the other hand, computer

science provided plenty of approaches,

from AI and Formal Languages fields.

(Pillai & Tedesco, 2024, p. x).

The “difficulty” encountered in converting

linguistic theory into algorithmic form may not only

be a consequence of inherent theoretical complexity,

See for example how in 2024 the scientific journal

Mathematics released a special issue titled “Current Trends

in Natural Language Processing (NLP) and Human

but also of a limited willingness to work with linguists

to translate insights into implementable formats.

Thus, while there is undeniable value in linguistics as

a theoretical cornerstone, a certain reticence to

engage in cross-disciplinary adaptation has limited

the field's influence.

All in all, the argument presented by Pillai and

Tedesco (2023) reflects a subtly dismissive view of

the relationship between linguistics and NLP,

attributing limited success in early times to the

supposed impracticality of linguistic theory, yet

failing to account for the potential value of linguistics

as the very basis of scientifically relevant results in

the field if approached with rigour and adaptability.

Even though NLP continues to be defined as a

discipline that combines linguistics with AI and

computer science, we are now in the conditions to

believe NLP is more and more about the latter two

than the former, apparently foundational one

discipline that strives for scientificity and

explainability. In short, linguists have been excluded

from the discussion, while the remaining experts at

the table are celebrating their “advancements”

This paper hereby highlights the need for greater

scholarly agility in an interdisciplinary call to action,

with the objective of rendering linguistic theories

actionable within computational frameworks. In

today’s technological landscape, computational

linguistics (CL) must not simply provide

“inspiration”, but should actively integrate into AI,

offering a scaffold for interpretability, structure, and

meaningful context to NLP models. Embracing this

shift requires software engineers to re-engage with

the field together with their colleagues in language-

related studies, to refine theories in ways that make

them more computationally relevant without losing

the depth of insight that linguistics uniquely offers as

is very common nowadays.

This leads me to consider as such the current

status of NLP: more and more P, less and less NL.

3 THE IMPORTANCE OF

DEALING WITH NATURAL

LANGUAGE

Time for some recent history, this time in a more

positive light. Yogatama et al. (2019), long before the

popularisation of AI-powered chatbots (see the now

unsurpassably popular ChatGPT by OpenAI,

Language Technology (HLT)”. Unsurprisingly, no

contribution in it has any reference to linguistic theory.

Bringing NL Back with P: Defending Linguistic Methods in NLP for Future AI Applications

1063

launched in 2022), tempered the enthusiasm for the

apparent progress that NLP seemed to be making with

the advent of LLMs, emphasising the need to evaluate

model performance based on demonstrated linguistic

intelligence rather than solely on the plausibility of

their output. In fact, overly simplifying in a sense but

leaving nothing out of the frame, cutting-edge

examples of generative AI produce sentences that

resemble meaningful instantiations of natural

language, yet at their base they don’t follow any rules.

In a nutshell, the model is simply reassembling

patterns is seen before, based on statistical

likelihoods.

Emily Bender and Alexander Koller (2020) added

that, since human-analogous processing of natural

language is a big goal of research in artificial

intelligence, we should reconsider the futuristic

narrative that sees LLMs as their protagonists and

focus more on questions of machine intelligence

regarding language use. The same Emily Bender has

been at the centre of the scene since she first-authored

the illuminating paper which introduced the concept of

“stochastic parrot” (Bender et al., 2021) as a neologism

to designate the ignorance that AI shows with respect

to any semantic implication of the words they draw

from LLMs and subsequently employ; it is not a case

that Bender is a computational linguist, not a software

engineer, engaged in an ongoing project focused on the

production of linguistic formalisms through

unrestricted grammars

. Also Jiang et al. (2023, p. 43

and sq.) have insisted on the need for NLP AI-driven

tools to rely on some sort of rule-based knowledge,

including in this proposition of theirs the hunger for a

good deal of linguistic information.

4 THE STATUS OF NL AND NLP

In his publications, Max Silberztein opposes what he

calls “linguistic methods” to “statistical” or – with a

peculiar choice of terminology that could clash with

shared conventions

– “empirical methods”

(Silberztein, 2016; 2024), and always advocated for

the first. When defending the linguistic approach, he

outlined ten reasons why merely stochastic methods

should be disregarded (Silberztein, 2016, pp. 19-27),

of which the tenth contains the quintessential point of

See https://linguistics.washington.edu/research/projects-

and-grants/lingo-grammar-matrix.

In linguistic theory, “empirical linguistics” can be said to

be the study of language based on direct observation and

data collection from real-world language use (see at least

Sampson, 2002, and Schütze, 2016 [1996]). Many linguists,

especially field linguists, would say they take pride in being

my argumentation: one can’t derive from the

application of statistical methods, even when they

produce high-quality if not spectacular results in

terms of NLP tasks, any linguistic information of

scientific relevance, but mere mathematical

functions. His preface to Linguistic Resources for

Natural Language Processing (Silberztein, 2024) is a

brutally concise manifesto of such ideal.

This proves to be particularly relevant for the

study of low-resource languages. In the indigenous-

led position paper resulting from the initiative of

Lewis et al. (2020, pp. 35-36, p. 65, pp. 93-100),

language is a key element around which AI ethics,

representation, and inclusivity revolve. The authors

propose frameworks for Indigenous data governance

and culturally aware AI systems that prioritise

indigenous ownership and stewardship of digital

linguistic resources over the massive application of

statistical computing on transcribed corpora in

endangered languages carrying traditional

knowledge.

Linguists working in these contexts may face

some challenges when they try to introduce digital

methodologies in their language documentation tasks,

among which the most necessary is interlinearisation,

a process of aligning a text with its translation line by

line. So far, in the efforts intended to produce NLP

tools for linguists working with low-resource

languages, cutting-edge research in the field of

annotation and interlinearisation of texts has been

conducted using machine learning techniques applied

to training datasets. Zhao et al. (2020) proposed a

model based on Recurrent Neural Networks (RNN)

using source transcription and its translation for the

automatic gloss generation task, trying to overcome

the need for word alignment that is imperative when

using Conditional Random Fields (CRF), which had

been applied shortly before by Angelina McMillan-

Major (2020) for the same purpose. Diego Barriga

Martínez and colleagues (2021) have experimented

with both RNN and CRF, and described how factors

like systematicity and frequency of morphological

rules in the training corpus are directly proportional

to precision scoring. It is relevant to cite how Moeller

et al. (2020) focused on goals of morphological

accuracy, and showed a methodology based on

transformers that nevertheless includes, in an

empirical, as their empirical studies would involve

qualitative methods, such as discourse analysis or

ethnographic observations; in his terminological choice,

however, Silberztein only refers to the quantitative side of

empirical analysis, like statistical modelling of language-

use samples recollected in training corpora. On the other

hand, he calls “linguistic” all rule-based methods.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1064

intermediate passage, human intervention: after

partial inflectional paradigms were automatically

extracted from the interlinear glossed texts, the

linguist Andrew Brumleve was asked to “clean” the

resulting data and regroup words under common

features and inflectional paradigms, implementing

corrections where necessary (Moeller et al., 2020, p.

5256); such an operation of fine-tuning, reminiscent

of reinforcement learning from human feedback

(RLHF), could be eased with applications

implementing linguistic methods:

Giving explicitly the list of rules used to

disambiguate a training corpus, rather than

the corpus itself, would be more useful to

the scientific community, as these rules

could be examined, corrected, and

maintained, while at the same time being

applied to any text, and therefore be

checked, falsified, and refined at will. In

essence, this is what the linguistic approach

proposes. (Silberztein, 2024, p. 21).

This contrasts stochastic AI models, since “when

they produce inadequate results, they cannot be

corrected or tweaked: the model must be rebuilt from

scratch with new data” (Verdicchio, 2023, p. 12). The

assumption that statistical approaches are less time-

consuming than rule-based linguistic approaches thus

reveals to be false (Silberztein, 2016, p. 22); the

solution resides, again, in a better collaboration

network that calls for the inclusion of computational

linguists, who would be “more than happy” to help

computer engineers with their expertise (Silberztein,

2024). In a very feasible view, that nonetheless seems

rather utopian seen the state of the industry:

Formalising the lexical, morphological,

syntactic, and distributional properties of

the standard English vocabulary would

require the work of a dedicated team, much

smaller than the gigantic groups assigned to

the construction of tagged corpora for

statistical NLP applications (statistical

tagging or translation). A project like this

would be beneficial for the whole linguistic

community and would enable the

development of NLP software with

unequalled precision. (Silberztein, 2016, p.

27).

The Association internationale des utilisateurs de NooJ

[International Association of NooJ Users] is an

international scientific association and a non-profit

organisation devoted to the advancement of linguistic

research and CL. Its members consist of researchers,

educators, linguists, and developers who collaborate to

A hybrid approach would leverage both stochastic

patterns and validated linguistic rules to produce

more interpretable and reliable outputs. Such a

methodology would also minimise dependency on

large datasets, especially for low-resource languages

or language nuances not well-represented in the

available data, as well as for tasks that tackle out-of-

domain specialistic language. Huang et al. (2020)

already showed an example of combination of neural

machine translation (NMT) with a rule-based layer

made of dictionaries and formalised syntactic

information, with promising results that encourage

the integration of the latter to improve the

performance of the former: especially in out-of-

domain translations, their experiments showed that

this hybrid approach can help in achieving higher

precision. Also Wahde & Virgolin (2022) proposed a

hybrid model for a conversational agent that is more

transparent and interpretable thanks to a rule-based

knowledge base, along with a set of principles (the

five Is) that in their view governs transparency in AI:

Interpretability, Inherent capability to explain,

Independent data, Interactive learning, and

Inquisitiveness (Wahde & Virgolin, 2022, pp. 1858-

1859). It is from here that the approach presented in

this paper is introduced, with a renewed focus on low-

resource languages and the interest that linguists may

show in its regard.

5 INTRODUCING NOOJ FOR

HYBRID NLP

NooJ is here designated as a privileged choice for

integrating the linguistic resources here discussed in

future AI models. NooJ was introduced by the

aforementioned scholar Max Silberztein (2004) and

counts on a community in force of its being open

source

. NooJ offers a powerful environment to

develop resources that formalise linguistic

phenomena, and to use them to parse texts and

corpora, with NLF capacities that cover grammars of

every level of the Chomsky’s hierarchy, from Finite

State Automata (FSA) to Turing machines in the form

of Enhanced Recursive Transition Networks (ERTN),

that while being half a century old demonstrates to be

improve language processing techniques and support the

integration of NooJ in both academic and professional

settings. The association welcomes new members year-

round and gathers every year since 2006 for an international

conference dedicated to the various applications of the

software (https://nooj.univ-fcomte.fr/association.html).

Bringing NL Back with P: Defending Linguistic Methods in NLP for Future AI Applications

1065

still relevant with new applications in the field

(Chomsky & Schützenberger, 1963; Freidin, 2013).

Figure 1: _VG.nog, a syntactic grammar that comes with

the NooJ English module.

In ten years of history, contributions to CL using

NooJ have been innumerable. Low-resource

languages like Quechua (Duran, 2017; 2019) and

Kabyle (Aoughlis et al., 2013) now have access to

NLP thanks to NLF performed with the software.

Moreover, with the use of NooJ, to cite a recent

example, Walter Koza and María Mare (2023) were

able to formalise resources that can not only

recognise, but also generate texts with complex

constructions involving the juxtaposition of many

clitic pronouns in Spanish, while Mario Monteleone

(2020) had showed how NooJ linguistic resources

alone can be used to generate texts.

While it is true that NLF methodologies – in this

case Maurice Gross’ Lexicon-Grammar (Gross, 1994)

– can be used on their own for such tasks, this paper

proposes a methodology that integrates these rule-

based systems into a possible AI model that draws its

power from a deeper linguistic approach to leverage

both structured linguistic rules and machine learning.

6 INTEGRATING NLF INTO AI: A

PROPOSED METHODOLOGY

The advantage of this approach would lie in its dual

focus: rule-based linguistic methods provide

precision and control, while AI agents provide

adaptability and generalisation. This synergy creates

a more versatile and accurate language-

comprehensive system that can be applied across

various domains, from automated translation to

educational tools: this model would offer a level of

(linguistic) explainability lacking in pure AI systems,

making it more reliable and easier to fine-tune (as

lacunary performances would only require to touch

specific resources like dictionaries and grammars,

and not a whole model).

With a structured layer of linguistic rules, users

can potentially track and explain why the model made

certain linguistic choices while performing NLP

tasks. For instance, in the context of text generation

tasks, if a response is generated in a particular tense

or with a specific syntactic structure, the model could

reference the underlying linguistic rules it applied,

making its decision process interpretable.

The integration of this additional layer involves

NooJ’s linguistic resources called in action to

establish a solid linguistic foundation (see Figure 2).

Reference corpora are piped into NooJ to be

tokenised, annotated in electronic dictionaries and, in

the same way, the syntactic and morphological

structures present in such bodies of data are parsed by

finite-state grammars. In this scenario, the rule-based

framework serves as a guide that constrains and

informs the AI model, ensuring that its outputs adhere

closely to established language norms.

Figure 2: Flowchart for the proposed methodology.

To implement this methodology, a pipeline could

be developed where NooJ's resources first preprocess

the input textual data, structuring it in a way that

aligns with linguistic rules formalised thanks to

NooJ’s powerful linguistic engine. These structured

inputs are then fed into a machine learning model

trained to recognise and generate text that aligns with

this structure. The model could use techniques such

as transformers, which have a strong ability to capture

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1066

long-range dependencies and context in text; in any

case, this proposition doesn’t aim to propose any

particular instruction of such final step, and focuses

rather on the layer to be developed looking at

linguistically-informed ways of processing input

data: morphological, syntactic, and semantic parsing

are introduced as a pre-processing layer before on-

demand NLP tasks (such as text generation, spell-

checking and machine translation), with such layer

susceptible of the intervention of skilled linguists, the

humans in the loop (HITL).

By grounding these models in structured

linguistic data, developers would give them a robust

linguistic framework to work within, which helps

them not only to produce coherent and linguistically

accurate text, but also to make such output the result

of an operation that involves a “linguistic” rather than

merely “mathematical” intelligence.

7 CONCLUSIONS

This paper, oriented to a better future for NLP,

proposes a framework in which linguistic knowledge

participates as the main actor in the creation of agents

that can achieve actual goals in terms of linguistic

competence, aligning more closely with human

cognitive processes and thus contributing to a

significant qualitative step in the very definition of

Artificial Intelligence, while also aligning more closely

with the goals and needs of computational linguistics.

Modern NLP is indeed powered by machine

learning, but linguistic theories and methodologies

should continue to deeply inform its development.

NLP will be far more effective without linguistics

guiding how we as humans understand, structure, and

interpret language: here’s the reason why to address

an interdisciplinary effort that bridges a gap that is

now the elephant in the room for language scholars

interested in digital methodologies. AI researchers

and engineers should collaborate closely with

linguists, ensuring that models respect the

complexities of language structures and phenomena.

Such an approach would pave the way for AI systems

capable of genuinely engaging in natural, human-like

communication, enhancing their utility and

trustworthiness across a wide range of applications;

the first of which is the possibility to retrieve the

linguistic reasoning behind AI’s manipulation of

textual data, paving the way for a future in which low-

resource languages will see a degree of representation

that allows for the engagement in the development

and utilisation of digital resources.

NooJ complex linguistic resources could be

embedded directly into the AI pipeline through

command-line calls (noojapply.exe), acting as pre-

processing for NLP tasks. Moreover, if NooJ

resources are exposed through APIs, they can be

called to perform real-time checks as part of an AI

application, improving real-time analytical accuracy

of the model’s performance, so to finally bring back

NL together with P in a way that doesn’t only strive

for some semblable aesthetic imitation of the products

of human language, but that first and foremost cares

about the linguistic intelligence that can (and, in my

view, must) be made artificial for such purpose. With

such intelligence comes a renewed ethical framework

in approaching a vaster number of low-resource

languages, and with such ethical framework comes a

more sustainable future workflow.

ACKNOWLEDGEMENTS

Special thanks to Giuseppe Laquidara, who revised

an early version of this paper and helped me perfect it

with his invaluable advice.

REFERENCES

Aoughlis, F., Naït-Zerrad, K., Annouz, H., Kaci, F., &

Habet, M. S. (2014). A New Tamazight Module for

NooJ. In Koeva, S., Mesfar, S., & Silberztein, M. (eds.),

Formalising Natural Languages with NooJ 2013:

Selected papers from the NooJ 2013 International

Conference, pp. 13-21. Cambridge Scholars Publishing.

Barriga Martínez, D., & Mijangos, V., Gutierrez-Vasques, X.

(2021). Automatic Interlinear Glossing for Otomi

Language. In Mager, M., Oncevay, A., Rios, A., Meza

Ruiz, I. V., Palmer, A., Neubig, G., & Kann, K. (eds.),

Proceedings of the First Workshop on Natural Language

Processing for Indigenous Languages of the Americas,

pp. 34–43. Association for Computational Linguistics.

https://doi.org/10.18653/v1/2021.americasnlp-1.5.

Bender, E. M., & Koller, A. (2020). Climbing towards

NLU: On Meaning, Form, and Understanding in the

Age of Data. In Jurafsky, D., Chai, J., Schluter, N., &

Tetreault, J. (eds.), Proceedings of the 58th Annual

Meeting of the Association for Computational

Linguistics, pp. 5185–5198. Association for

Computational Linguistics. https://doi.org/10.18653/

v1/2020.acl-main.463.

Bender, E. M., Gebru, T.; McMillan-Major, A., &

Shmitchell, S. (2021). On the Dangers of Stochastic

Parrots: Can Language Models Be Too Big? 🦜. In

Proceedings of the 2021 ACM Conference on Fairness,

Accountability, and Transparency (FAccT '21), pp.

610–623. Association for Computing Machinery.

https://dl.acm.org/doi/10.1145/3442188.3445922.

Bringing NL Back with P: Defending Linguistic Methods in NLP for Future AI Applications

1067

Chomsky, N, & Schützenberger, M. P. (1963). The

algebraic theory of context free languages. In Braffort,

P., & Hirschberg, D. (eds.), Computer Programming

and Formal Systems, pp. 118–161. North Holland.

Duran M. (2017). Quechua Module for NooJ Multilingual

Linguistic Resources for MT. In Barone, L.,

Monteleone, M., Silberztein, M. (eds.), Automatic

Processing of Natural Language Electronic Texts with

NooJ: Selected Papers from the International

Conference NooJ 2016, pp. 48-63. Springer.

Duran, M. (2019). Can a robot help save an endangered

language? In Adda, G., Choukri, K., Kasinskaite-

Buddeberg, I., Mariani, J., Mazo, H., & Sakriani, S.

(eds.), Proceedings of the Language Technologies for

All, pp. 284-287. UNESCO.

Fanni, S.C., Febi, M., Aghakhanyan, G., Neri, E. (2023).

Natural Language Processing. In Klontzas, M.E.,

Fanni, S.C., Neri, E. (eds.), Introduction to Artificial

Intelligence. Imaging Informatics for Healthcare

Professionals (pp. 87-99). Springer, Cham.

https://doi.org/10.1007/978-3-031-25928-9_5.

Freidin, R. (2013). Noam Chomsky's contribution to

linguistics. In Allan, K. (ed.), The Oxford Handbook of

the History of Linguistics (pp. 438-467). Oxford

University Press. https://doi.org/10.1093/oxfordhb/

9780199585847.013.0020.

Gross, Maurice (1994). Constructing Lexicon-Grammars.

In Beryl T., Atkins, S. & Zampolli, A. (eds.),

Computational approaches to the lexicon, 213-263.

Oxford OUP.

Huang, J., Lee, K., & Kim, Y. (2020). Hybrid translation

with classification: Revisiting rule-based and neural

machine translation. Electronics, 9(2), 201.

https://doi.org/10.3390/electronics9020201.

Jiang, M., Lin, B. Y., Wang, S., Xu, Y., Yu, W., & Zhu, C.

(2023). Knowledge-augmented methods for natural

language processing. Springer.

Koza, W., & Mare, M. (2023). Computational modeling of

a generative grammar for clitic order in River Plate

Spanish. Revista Española de Lingüística

Aplicada/Spanish Journal of Applied Linguistics, 36(1),

270–306. https://doi.org/10.1075/resla.20053.koz.

Lewis, J. E., Abdilla, A., Arista, N., Baker, K.,

Benesiinaabandan, S., Brown, M., Cheung, M.,

Coleman, M., Cordes, A., Davison, J., Duncan, K.,

Garzon, S., Harrell, D. F., Jones, P. L.,

Kealiikanakaoleohaililani, K., Kelleher, M., Kite, S.,

Lagon, O., Leigh, J., Levesque, M., Mahelona, K.,

Moses, C., Nahuewai, I., Noe, K., Olson, D., Parker

Jones, 'Ō., Running Wolf, C., Running Wolf, M., Silva,

M., Fragnito, S., & Whaanga, H. (2020). Indigenous

Protocol and Artificial Intelligence Position Paper.

Project Report. Indigenous Protocol and Artificial

Intelligence Working Group and the Canadian Institute

for Advanced Research, Honolulu, HI.

McMillan-Major, A. (2020). Automating Gloss Generation

in Interlinear Glossed Text. Society for Computation in

Linguistics, 3(1), 338-349. https://doi.org/10.7275/

tsmk-sa32.

Moeller, S., Liu, L., Yang, C., Kann, K., & Hulden, M.

(2020). IGT2P: From Interlinear Glossed Texts to

Paradigms. In Proceedings of the 2020 Conference on

Empirical Methods in Natural Language Processing

(EMNLP), pp. 5251–5262. Association for

Computational Linguistics. https://doi.org/10.18653/

v1/2020.emnlp-main.424.

Monteleone, M. (2020). Automatic Text Generation: How

to Write the Plot of a Novel with NooJ. In Fehri, H.,

Mesfar, S. & Silberztein, M. (eds.), Formalizing

Natural Languages with NooJ 2019 and Its Natural

Language Processing Applications 13th International

Conference, NooJ 2019, Hammamet, Tunisia, June 7–

9, 2019 (pp. 135-146). Springer.

Pillai, A. S., & Tedesco, R. (2024). Machine learning and

deep learning in natural language processing. CRC

Press.

Sampson, G. (2002). Empirical linguistics. A&C Black.

Schütze, C. T. (2016 [1996]). The empirical base of

linguistics: Grammaticality judgments and linguistic

methodology. Language Science Press.

Silberztein, M. (2004). NooJ: A Cooperative, Object-

Oriented Architecture for NLP. In INTEX pour la

Linguistique et le traitement automatique des langues.

Cahiers de la MSH Ledoux, Presses Universitaires

de Franche-Comté. https://books.openedition.org/

pufc/30117.

Silberztein, M. (2016). Formalizing natural languages: The

NooJ approach. John Wiley & Sons.

Silberztein, M. (2024). Preface. In Silberztein, M. (ed.),

Linguistic Resources for Natural Language Processing

(pp. xviii-xvi). Springer Nature.

Verdicchio, M. (2023). Che cosa genera davvero l’IA

generativa? ParadoXa, 7(4), 29-43.

Wahde, M., & Virgolin, M. (2022). DAISY: An

Implementation of Five Core Principles for Transparent

and Accountable Conversational AI. International

Journal of Human–Computer Interaction, 39(9), 1856–

1873. https://doi.org/10.1080/10447318.2022.2081762.

Yogatama, D., de Masson d'Autume, C., Connor, J.,

Kocisky, T., Chrzanowski, M., Kong, L., Lazaridou, A.,

Ling, W., Yu, L., Dyer, C., & Blunsom, P. (2019).

Learning and evaluating general linguistic intelligence.

arXiv. https://arxiv.org/abs/1901.11373.

Zhao, X., Ozaki, S., Anastasopoulos, A., Neubig, G., &

Levin, L. (2020). Automatic Interlinear Glossing for

Under-Resourced Languages Leveraging Translations.

In Scott, D., Bel, N., & Zong, C. (eds.), Proceedings of

the 28th International Conference on Computational

Linguistics, pp. 5397–5408. International Committee

on Computational Linguistics. https://doi.org/10.

18653/v1/2020.coling-main.471.

Zianouka, Y., Hetsevich, Y., Latyshevich, D., & Dzenisiuk,

Z. (2022). Automatic Generation of Intonation Marks

and Prosodic Segmentation for Belarusian NooJ Module.

In Bigey, M., Richeton, A., Silberztein, M., Thomas, I.

(eds.), Formalizing Natural Languages: Applications to

Natural Language Processing and Digital Humanities.

15th International Conference (pp. 231-242), Besançon,

France, June 9–11, 2021. Springer.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1068