This paper investigates prominence-aware automatic speec h recognition (ASR) by combining prominence detection and speech recognition for conversational Austrian German. Fi rst, prominence detectors were developed by fine-tuning wav2vec 2 models to classify word-level prominence. The detector was then used to automatically annotate prosodic prominence in a large corpus. Based on those annotations, we trained novel prominence-aware ASR systems that simultaneously transcr ibe words and their prominence levels. The integration of promi - nence information did not change performance compared to our baseline ASR system, while reaching a prominence detection accuracy of 85.53% for utterances where the recognized word sequence was correct. This paper shows that transforme r-based models can effectively encode prosodic information a nd represents a novel contribution to prosody-enhanced ASR, with potential applications for linguistic research and pr osody-informed dialogue systems.

Tokenization remains a fundamental yet underexplored bottleneck in natural language processing, with strategies largely static despite remarkable progress in model architectures. We present SupraTok, a novel tokenization architecture that reimagines subword segmentation through three innovations: cross-boundary pattern learning that discovers multi-word semantic units, entropy-driven data curation that optimizes training corpus quality, and multi-phase curriculum learning for stable convergence. Our approach extends Byte-Pair Encoding by learning "superword" tokens, coherent multi-word expressions that preserve semantic unity while maximizing compression efficiency. SupraTok achieves 31% improvement in English tokenization efficiency (5.91 versus 4.51 characters per token) compared to OpenAI's o200k tokenizer and 30% improvement over Google's Gemma 3 tokenizer (256k vocabulary), while maintaining competitive performance across 38 languages. When integrated with a GPT-2 scale model (124M parameters) trained on 10 billion tokens from the FineWeb-Edu dataset, SupraTok yields 8.4% improvement on HellaSWAG and 9.5% on MMLU benchmarks without architectural modifications. While these results are promising at this scale, further validation at larger model scales is needed. These findings suggest that efficient tokenization can complement architectural innovations as a path to improved language model performance.

We investigate the problem of segmenting unlabeled speech into word-like units and clustering these to create a lexicon. Prior work can be categorized into two frameworks. Bottom-up methods first determine boundaries and then cluster the fixed segmented words into a lexicon. In contrast, top-down methods incorporate information from the clustered words to inform boundary selection. However, it is unclear whether top-down information is necessary to improve segmentation. To explore this, we look at two similar approaches that differ in whether top-down clustering informs boundary selection. Our simple bottom-up strategy predicts word boundaries using the dissimilarity between adjacent self-supervised features, then clusters the resulting segments to construct a lexicon. Our top-down system is an updated version of the ES-KMeans dynamic programming method that iteratively uses K-means to update its boundaries. On the five-language ZeroSpeech benchmarks, both approaches achieve comparable state-of-the-art results, with the bottom-up system being nearly five times faster. Through detailed analyses, we show that the top-down influence of ES-KMeans can be beneficial (depending on factors like the candidate boundaries), but in many cases the simple bottom-up method performs just as well. For both methods, we show that the clustering step is a limiting factor. Therefore, we recommend that future work focus on improved clustering techniques and learning more discriminative word-like representations. Project code repository: https://github.com/s-malan/prom-seg-clus.

Byte-Pair Encoding (BPE) has become a widely adopted subword tokenization method in modern language models due to its simplicity and strong empirical performance across downstream tasks. However, applying BPE to unsegmented languages such as Chinese presents significant challenges, as its frequency-driven merge operation is agnostic to linguistic boundaries. To address this, we propose two entropy-informed pre-tokenization strategies that guide BPE segmentation using unsupervised information-theoretic cues. The first approach uses pointwise mutual information and left/right entropy to identify coherent character spans, while the second leverages predictive entropy derived from a pretrained GPT-2 model to detect boundary uncertainty. We evaluate both methods on a subset of the PKU dataset and demonstrate substantial improvements in segmentation precision, recall, and F1 score compared to standard BPE. Our results suggest that entropy-guided pre-tokenization not only enhances alignment with gold-standard linguistic units but also offers a promising direction for improving tokenization quality in low-resource and multilingual settings.

Language models provide a key framework for studying linguistic theories based on prediction, but phonological analysis using large language models (LLMs) is difficult; there are few phonological benchmarks beyond English and the standard input representation used in LLMs (subwords of graphemes) is not suitable for analyzing the representation of phonemes. In this work, we demonstrate how word segmentation can be used as a phonological probing task, allowing us to study the representations learned by phoneme-based language models trained on child-directed speech across 31 languages. Following computational models of word segmentation, we present unsupervised methods for extracting word boundaries from a trained model using the observation that prediction-error peaks at the start of words. We also use linear probes to identify that these models implicitly track word boundaries, even when they do not appear in training. This cross-lingual work corroborates statistical learning theories of acquisition and empirically motivates new methods for training subword tokenizers.

Contemporary Quranic Orthography (CQO) relies on a precise system of phonetic notation that can be traced back to the early stages of Islam, when the Quran was mainly oral in nature and the first written renderings of it served as memory aids for this oral tradition. The early systems of diacritical marks created on top of the Quranic Consonantal Text (QCT) motivated the creation and further development of a fine-grained system of phonetic notation that represented tajwid-the rules of recitation. We explored the systematicity of the rules of tajwid, as they are encountered in the Cairo Quran, using a fully and accurately encoded digital edition of the Quranic text. For this purpose, we developed a python module that can remove or add the orthographic layer of tajwid from a Quranic text in CQO. The interesting characteristic of these two sets of rules is that they address the complete Quranic text of the Cairo Quran, so they can be used as precise witnesses to study its phonetic and prosodic processes. From a computational point of view, the text of the Cairo Quran can be used as a linchpin to align and compare Quranic manuscripts, due to its richness and completeness. This will let us create a very powerful framework to work with the Arabic script, not just within an isolated text, but automatically exploring a specific textual phenomenon in other connected manuscripts. Having all the texts mapped among each other can serve as a powerful tool to study the nature of the notation systems of diacritics added to the consonantal skeleton.

An important step in understanding how children acquire languages is studying how infants learn word segmentation. It has been established in previous research that infants may use statistical regularities in speech to learn word segmentation. The research of Goldwater et al., demonstrated that incorporating context in models improves their ability to learn word segmentation. We implemented two of their models, a unigram and bigram model, to examine how context can improve statistical word segmentation. The results are consistent with our hypothesis that the bigram model outperforms the unigram model at predicting word segmentation. Extending the work of Goldwater et al., we also explored basic ways to model how young children might use previously learned words to segment new utterances.

Recent advancements in machine learning have significantly improved speech recognition, but recognizing speech from non-fluent or accented speakers remains a challenge. Previous efforts, relying on rule-based pronunciation patterns, have struggled to fully capture non-native errors. We propose two data-driven approaches using speech corpora to automatically detect mispronunciation patterns. By aligning non-native phones with their native counterparts using attention maps, we achieved a 5.7% improvement in speech recognition on native English datasets and a 12.8% improvement for non-native English speakers, particularly Korean speakers. Our method offers practical advancements for robust Automatic Speech Recognition (ASR) systems particularly for situations where prior linguistic knowledge is not applicable.

Inspired by early research on exploring naturally annotated data for Chinese Word Segmentation (CWS), and also by recent research on integration of speech and text processing, this work for the first time proposes to explicitly mine word boundaries from speech-text parallel data. We employ the Montreal Forced Aligner (MFA) toolkit to perform character-level alignment on speech-text data, giving pauses as candidate word boundaries. Based on detailed analysis of collected pauses, we propose an effective probability-based strategy for filtering unreliable word boundaries. To more effectively utilize word boundaries as extra training data, we also propose a robust complete-then-train (CTT) strategy. We conduct cross-domain CWS experiments on two target domains, i.e., ZX and AISHELL2. We have annotated about 1,000 sentences as the evaluation data of AISHELL2. Experiments demonstrate the effectiveness of our proposed approach.

Language models are typically trained on large corpora of text in their default orthographic form. However, this is not the only option; representing data as streams of phonemes can offer unique advantages, from deeper insights into phonological language acquisition to improved performance on sound-based tasks. The challenge lies in evaluating the impact of phoneme-based training, as most benchmarks are also orthographic. To address this, we develop a pipeline to convert text datasets into a continuous stream of phonemes. We apply this pipeline to the 100-million-word pre-training dataset from the BabyLM challenge, as well as to standard language and grammatical benchmarks, enabling us to pre-train and evaluate a model using phonemic input representations. Our results show that while phoneme-based training slightly reduces performance on traditional language understanding tasks, it offers valuable analytical and practical benefits.