Human language learning takes place under limited exposure and is shaped by structured regularities in the input, whereas standard Transformers apply a largely uniform sequence of computations at every token position. We ask whether dynamic latent segmentation can provide a useful inductive bias for language modeling under BabyLM-scale data constraints. BabyConceptLM causally encodes tokens, groups contextual states into learned contiguous latent segments, processes the shorter sequence with a deeper latent-level backbone, and returns to token resolution through a leakage-safe causal readout. We evaluate two profiles using recipe-matched token-only controls, with analyses of computational cost and segmentation behavior. Across English, multilingual, and Chinese comparisons, BabyConceptLM trails matched token-only controls by 0.60–1.47 points. Estimated padding-aware dense-forward FLOPs are reduced by 26% and 39% for the English and multilingual 5/8/3 models, respectively, but increase for the Chinese 2/4/2 configuration. Segmentation audits show weak alignment with word boundaries and predominantly length-driven behavior, with the larger model frequently reaching its eight-token segment cap. These results show that latent segmentation can yield a favorable compute–accuracy trade-off in some configurations, but does not by itself produce linguistically structured units or improved predictive performance. BabyConceptLM serves as a diagnostic test of hierarchical latent computation under restricted language exposure.

Large Language Model (LLMs) can be used to write or modify documents, presenting a challenge for understanding the intent behind their use. For example, benign uses may involve using LLM on a human-written document to improve its grammar or to translate it into another language. However, a document entirely produced by a LLM may be more likely to be used to spread misinformation than simple translation (, from use by malicious actors or simply by hallucinating). Prior works in Machine Generated Text (MGT) detection mostly focus on simply identifying whether a document was human or machine written, ignoring these fine-grained uses. In this paper, we introduce a HiErarchical, length-RObust machine-influenced text detector (HERO), which learns to separate text samples of varying lengths from four primary types: human-written, machine-generated, machine-polished, and machine-translated. HERO accomplishes this by combining predictions from length-specialist models that have been trained with Subcategory Guidance. Specifically, for categories that are easily confused (, different source languages), our Subcategory Guidance module encourages separation of the fine-grained categories, boosting performance. Extensive experiments across five LLMs and six domains demonstrate the benefits of our HERO, outperforming the state-of-the-art by 2.5-3 mAP on average.

Fine-grained address entity recognition (FGAER) from multi-turn spoken dialogues is particularly challenging. The major reason lies in that a full address is often formed through a conversation process. Different parts of an address are distributed through multiple turns of a dialogue with spoken noises. It is nontrivial to extract by turn and combine them. This challenge has not been well emphasized by main-stream entity extraction algorithms. To address this issue, we propose in this paper a logic-guided fine-grained address recognition method (Log-FGAER), where we formulate the address hierarchy relationship as the logic rule and softly apply it in a probabilistic manner to improve the accuracy of FGAER. In addition, we provide an ontology-based data augmentation methodology that employs ChatGPT to augment a spoken dialogue dataset with labeled address entities. Experiments are conducted using datasets generated by the proposed data augmentation technique and derived from real-world scenarios. The results of the experiment demonstrate the efficacy of our proposal.