When producing morphologically complex words, speakers can draw on memorized examples of morphological structure that differ in specificity, ranging from word-specific information to broad generalizations. However, there is much that is not known about how memory retrieval operates in morphological production, particularly for languages with agglutinative structure, like Turkish. While it has long been suggested that item-specific memory does not play a significant role in such languages, there is empirical evidence for affix chunking in Turkish. This paper models morphological production as a sequential decision task that can be optimized with reinforcement learning. Applied to Turkish noun inflection, the model predicts a strategy of blending synthesis with multi-affix chunks gleaned from memorized words, but more in some parts of the paradigm than others, suggesting a fine-tuned relationship between synthesis and memory utilization.

Understanding emotion from speech requires sensitivity to both lexical and acoustic cues. However, it remains unclear whether large audio language models (LALMs) genuinely process acoustic information or rely primarily on lexical content. We present LISTEN (Lexical vs. Acoustic Speech Test for Emotion in Narratives), a controlled benchmark designed to disentangle lexical reliance from acoustic sensitivity in emotion understanding. Across evaluations of six state-of-the-art LALMs, we observe a consistent lexical dominance. Models predict "neutral" when lexical cues are neutral or absent, show limited gains under cue alignment, and fail to classify distinct emotions under cue conflict. In paralinguistic settings, performance approaches chance. These results indicate that current LALMs largely "transcribe" rather than "listen," relying heavily on lexical semantics while underutilizing acoustic cues. LISTEN offers a principled framework for assessing emotion understanding in multimodal models.

This paper offers a call to action. We urge our colleagues in the research community to play a greater role in the articulation of our findings to the public. To illustrate the stakes we present a case study on the initial stages of an LLM-based machine translation application's deployment in a real-world context: a text-2-911 system advertising capabilities in 55 languages for use in emergencies in which it may be difficult to call operators directly. We identify a number of common misconceptions about technologies such as these, concluding with a set of concrete recommendations and best practices for stakeholders at every stage of the development and deployment pipeline. While the advancement of scientific research often lies in solving the "hard" problems, we argue it is often the "easy" ones – problems for which the latest technology is often unnecessary – that are most overlooked.

Diffusion models produce high-fidelity speech but are inefficient for real-time use due to long denoising steps and challenges in modeling intonation and rhythm. To improve this, we propose Diffusion Loss-Guided Policy Optimization (DLPO), an RLHF framework for TTS diffusion models. DLPO integrates the original training loss into the reward function, preserving generative capabilities while reducing inefficiencies. Using naturalness scores as feedback, DLPO aligns reward optimization with the diffusion model's structure, improving speech quality. We evaluate DLPO on WaveGrad 2, a non-autoregressive diffusion-based TTS model. Results show significant improvements in objective metrics (UTMOS 3.65, NISQA 4.02) and subjective evaluations, with DLPO audio preferred 67% of the time. These findings demonstrate DLPO's potential for efficient, high-quality diffusion TTS in real-time, resource-limited settings.

Partly automated creation of interlinear glossed text (IGT) has the potential to assist in linguistic documentation. We argue that LLMs can make this process more accessible to linguists because of their capacity to follow natural-language instructions. We investigate the effectiveness of a retrieval-based LLM prompting approach to glossing, applied to the seven languages from the SIGMORPHON 2023 shared task. Our system beats the BERT-based shared task baseline for every language in the morpheme-level score category, and we show that a simple 3-best oracle has higher word-level scores than the challenge winner (a tuned sequence model) in five languages. In a case study on Tsez, we ask the LLM to automatically create and follow linguistic instructions, reducing errors on a confusing grammatical feature. Our results thus demonstrate the potential contributions which LLMs can make in interactive systems for glossing, both in making suggestions to human annotators and following directions.

Recent advancements in generative models have sparked a significant interest within the machine learning community. Particularly, diffusion models have demonstrated remarkable capabilities in synthesizing images and speech. Studies such as those by Lee et al. (2023), Black et al. (2023), Wang et al. (2023), and Fan et al. (2024) illustrate that Reinforcement Learning with Human Feedback (RLHF) can enhance diffusion models for image synthesis. However, due to architectural differences between these models and those employed in speech synthesis, it remains uncertain whether RLHF could similarly benefit speech synthesis models. In this paper, we explore the practical application of RLHF to diffusion-based text-to-speech synthesis, leveraging the mean opinion score (MOS) as predicted by UTokyo-SaruLab MOS prediction system (Saeki et al., 2022) as a proxy loss. We introduce diffusion model loss-guided RL policy optimization (DLPO) and compare it against other RLHF approaches, employing the NISQA speech quality and naturalness assessment model (Mittag et al., 2021) and human preference experiments for further evaluation. Our results show that RLHF can enhance diffusion-based text-to-speech synthesis models, and, moreover, DLPO can better improve diffusion models in generating natural and high quality speech audios.

This work investigates the in-context learning abilities of pretrained large language models (LLMs) when instructed to translate text from a low-resource language into a high-resource language as part of an automated machine translation pipeline. We conduct a set of experiments translating Southern Quechua to Spanish and examine the informativity of various types of context retrieved from a constrained database of digitized pedagogical materials (dictionaries and grammar lessons) and parallel corpora. Using both automatic and human evaluation of model output, we conduct ablation studies that manipulate (1) context type (morpheme translations, grammar descriptions, and corpus examples), (2) retrieval methods (automated vs. manual), and (3) model type. Our results suggest that even relatively small LLMs are capable of utilizing prompt context for zero-shot low-resource translation when provided a minimally sufficient amount of relevant linguistic information. However, the variable effects of context type, retrieval method, model type, and language-specific factors highlight the limitations of using even the best LLMs as translation systems for the majority of the world's 7,000+ languages and their speakers.

Introduction to the Special Issue.

This article situates ongoing debates about the governance policies of Archive of Our Own (AO3) in the context of the ideological tradition of technoliberalism, which links internet technologies with liberal ideals such as freedom of speech and individual choice. While often seen as a movement of the mid-2000s which swiftly gave way to straightforward corporatism, technoliberalism survives as a popular discourse in the AO3 debates, where popular rhetoric incorporates liberal defences of free speech and individual choice. Modern technoliberal arguments have developed in response to their critics, emphasizing individualistic content controls as an opportunity for self-expression and as protection from a historical tradition of censorship in fandom spaces.

Debates about the development of artificial superintelligence and its potential threats to humanity tend to assume that such a system would be historically unprecedented, and that its behavior must be predicted from first principles. I argue that this is not true: we can analyze multiagent intelligent systems (the best candidates for practical superintelligence) by comparing them to states, which also unite heterogeneous intelligences to achieve superhuman goals. States provide a model for several problems discussed in the literature on superintelligence, such as principal-agent problems and Instrumental Convergence. Philosophical arguments about governance, therefore, provide possible solutions to these problems, or point out problems in previously suggested solutions. In particular, the liberal concept of checks and balances, and Hannah Arendt’s concept of legitimacy, describe how state behavior is constrained by the preferences of constituents that could also apply to artificial systems. However, they also point out ways in which present-day computational developments could destabilize the international order by reducing the number of decision-makers involved in state actions. Thus, interstate competition not only serves as a model for the behavior of dangerous computational intelligences but also as the impetus for their development.

[This corrects the article DOI: 10.3389/fpsyg.2015.01793.].

This paper explores how Generative Adversarial Networks (GANs) learn representations of phonological phenomena. We analyze how GANs encode contrastive and non-contrastive nasality in French and English vowels by applying the ciwGAN architecture (Begus 2021a). Begus claims that ciwGAN encodes linguistically meaningful representations with categorical variables in its latent space and manipulating the latent variables shows an almost one to one corresponding control of the phonological features in ciwGAN's generated outputs. However, our results show an interactive effect of latent variables on the features in the generated outputs, which suggests the learned representations in neural networks are different from the phonological representations proposed by linguists. On the other hand, ciwGAN is able to distinguish contrastive and noncontrastive features in English and French by encoding them differently. Comparing the performance of GANs learning from different languages results in a better understanding of what language specific features contribute to developing language specific phonological representations. We also discuss the role of training data frequencies in phonological feature learning.

Maltese is often described as having a hybrid morphological system resulting from extensive contact between Semitic and Romance language varieties. Such a designation reflects an etymological divide as much as it does a larger tradition in the literature to consider concatenative and non-concatenative morphological patterns as distinct in the language architecture. Using a combination of computational modeling and information theoretic methods, we quantify the extent to which the phonology and etymology of a Maltese singular noun may predict the morphological process (affixal vs. templatic) as well as the specific plural allomorph (affix or template) relating a singular noun to its associated plural form(s) in the lexicon. The results indicate phonological pressures shape the organization of the Maltese lexicon with predictive power that extends beyond that of a word's etymology, in line with analogical theories of language change in contact.

We investigate how well words in the polysynthetic language Inuktitut can be translated by combining dictionary definitions, without use of a neural machine translation model trained on parallel text. Such a translation system would allow natural language technology to benefit from resources designed for community use in a language revitalization or education program, rather than requiring a separate parallel corpus. We show that the text-to-text generation capabilities of GPT-3 allow it to perform this task with BLEU scores of up to 18.5. We investigate prompting GPT-3 to provide multiple translations, which can help slightly, and providing it with grammar information, which is mostly ineffective. Finally, we test GPT-3's ability to derive morpheme definitions from whole-word translations, but find this process is prone to errors including hallucinations.

There are many challenges in morphological fieldwork annotation, it heavily relies on segmentation and feature labeling (which have both practical and theoretical drawbacks), it’s time-intensive, and the annotator needs to be linguistically trained and may still annotate things inconsistently. We propose a workflow that relies on unsupervised and active learning grounded in Word-and-Paradigm morphology (WP). Machine learning has the potential to greatly accelerate the annotation process and allow a human annotator to focus on problematic cases, while the WP approach makes for an annotation system that is word-based and relational, removing the need to make decisions about feature labeling and segmentation early in the process and allowing speakers of the language of interest to participate more actively, since linguistic training is not necessary. We present a proof-of-concept for the first step of the workflow, in a realistic fieldwork setting, annotators can process hundreds of forms per hour.

OSU’s inflection system is a transformer whose input is augmented with an analogical exemplar showing how to inflect a different word into the target cell. In addition, alignment-based heuristic features indicate how well the exemplar is likely to match the output. OSU’s scores substantially improve over the baseline transformer for instances where an exemplar is available, though not quite matching the challenge winner. In Part 2, the system shows a tendency to over-apply the majority pattern in English, but not Arabic.

We investigate how abstract processes like suffixation can be learned from morphological inflection task data using an analogical memory-based framework. In this framework, the inflection target form is specified by providing an example inflection of another word in the language. We show that this model is capable of near-baseline performance on the SigMorphon 2020 inflection challenge. Such a model can make predictions for unseen languages, allowing us to perform one-shot inflection on natural languages and investigate morphological transfer with synthetic probes. Accuracy for one-shot transfer can be unexpectedly high for some target languages (88% in Shona) and language families (53% across Romance). Probe experiments show that the model learns partially generalizable representations of prefixation, suffixation and reduplication, aiding its ability to transfer. We argue that the degree of generality of these process representations also helps to explain transfer results from previous research.

Existing work on making privacy policies accessible has explored new presentation forms such as color-coding based on the risk factors or summarization to assist users with conscious agreement. To facilitate a more personalized interaction with the policies, in this work, we propose an automated privacy policy question answering assistant that extracts a summary in response to the input user query. This is a challenging task because users articulate their privacy-related questions in a very different language than the legal language of the policy, making it difficult for the system to understand their inquiry. Moreover, existing annotated data in this domain are limited. We address these problems by paraphrasing to bring the style and language of the user's question closer to the language of privacy policies. Our content scoring module uses the existing in-domain data to find relevant information in the policy and incorporates it in a summary. Our pipeline is able to find an answer for 89% of the user queries in the privacyQA dataset.

In consequential domains such as recidivism prediction, facility inspection, and benefit assignment, it's important for individuals to know the decision-relevant information for the model's prediction. In addition, predictions should be fair both in terms of the outcome and the justification of the outcome. In other words, decision-relevant features should provide sufficient information for the predicted outcome and should be independent of the membership of individuals in protected groups such as race and gender. In this work, we focus on the problem of (un)fairness in the justification of the text-based neural models. We tie the explanatory power of the model to fairness in the outcome and propose a fairness-aware summarization mechanism to detect and counteract the bias in such models. Given a potentially biased natural language explanation for a decision, we use a multi-task neural model and an attribution mechanism based on integrated gradients to extract high-utility and low-bias justifications in form of a summary. The extracted summary is then used for training a model to make decisions for individuals. Results on several real world datasets suggest that our method drastically limits the demographic leakage in the input (fairness in justification) while moderately enhancing the fairness in the outcome. Our model is also effective in detecting and counteracting several types of data poisoning attacks that synthesize race-coded reasoning or irrelevant justifications.