WordSeg: Standardizing unsupervised word form segmentation from text.

A basic task in first language acquisition likely involves discovering the boundaries between words or morphemes in input where these basic units are not overtly segmented. A number of unsupervised learning algorithms have been proposed in the last 20 years for these purposes, some of which have been implemented computationally, but whose results remain difficult to compare across papers. We created a tool that is open source, enables reproducible results, and encourages cumulative science in this domain. WordSeg has a modular architecture: It combines a set of corpora description routines, multiple algorithms varying in complexity and cognitive assumptions (including several that were not publicly available, or insufficiently documented), and a rich evaluation package. In the paper, we illustrate the use of this package by analyzing a corpus of child-directed speech in various ways, which further allows us to make recommendations for experimental design of follow-up work. Supplementary materials allow readers to reproduce every result in this paper, and detailed online instructions further enable them to go beyond what we have done. Moreover, the system can be installed within container software that ensures a stable and reliable environment. Finally, by virtue of its modular architecture and transparency, WordSeg can work as an open-source platform, to which other researchers can add their own segmentation algorithms.

Variation in the input: a case study of manner class frequencies.

ABSTRACT What are the sources of variation in the input, and how much do they matter for language acquisition? This study examines frequency variation in manner-of-articulation classes in child and adult input. The null hypothesis is that segmental frequency distributions of language varieties are unigram (modelable by stationary, ergodic processes), and that languages are unitary (modelable as a single language variety). Experiment I showed that English segments are not unigram; they exhibit a 'bursty' distribution in which the local frequency varies more than expected by chance alone. Experiment II showed the English segments are approximately unitary: the natural background variation in segmental frequencies that arises within a single language variety is much larger than numerical differences across varieties. Variation in segmental frequencies seems to be driven by variation in discourse topic; topic-associated words cause bursts/lulls in local segmental frequencies. The article concludes with some methodological recommendations for comparing language samples.

A description of externally recorded womb sounds in human subjects during gestation.

OBJECTIVE: Reducing environmental noise benefits premature infants in neonatal intensive care units (NICU), but excessive reduction may lead to sensory deprivation, compromising development. Instead of minimal noise levels, environments that mimic intrauterine soundscapes may facilitate infant development by providing a sound environment reflecting fetal life. This soundscape may support autonomic and emotional development in preterm infants. We aimed to assess the efficacy and feasibility of external non-invasive recordings in pregnant women, endeavoring to capture intra-abdominal or womb sounds during pregnancy with electronic stethoscopes and build a womb sound library to assess sound trends with gestational development. We also compared these sounds to popular commercial womb sounds marketed to new parents. STUDY DESIGN: Intra-abdominal sounds from 50 mothers in their second and third trimester (13 to 40 weeks) of pregnancy were recorded for 6 minutes in a quiet clinic room with 4 electronic stethoscopes, placed in the right upper and lower quadrants, and left upper and lower quadrants of the abdomen. These recording were partitioned into 2-minute intervals in three different positions: standing, sitting and lying supine. Maternal and gestational age, Body Mass Index (BMI) and time since last meal were collected during recordings. Recordings were analyzed using long-term average spectral and waveform analysis, and compared to sounds from non-pregnant abdomens and commercially-marketed womb sounds selected for their availability, popularity, and claims they mimic the intrauterine environment. RESULTS: Maternal sounds shared certain common characteristics, but varied with gestational age. With fetal development, the maternal abdomen filtered high (500-5,000 Hz) and mid-frequency (100-500 Hz) energy bands, but no change appeared in contributions from low-frequency signals (10-100 Hz) with gestational age. Variation appeared between mothers, suggesting a resonant chamber role for intra-abdominal space. Compared to commercially-marketed sounds, womb signals were dominated by bowel sounds, were of lower frequency, and showed more variation in intensity. CONCLUSIONS: High-fidelity intra-abdominal or womb sounds during pregnancy can be recorded non-invasively. Recordings vary with gestational age, and show a predominance of low frequency noise and bowel sounds which are distinct from popular commercial products. Such recordings may be utilized to determine whether sounds influence preterm infant development in the NICU.

The role of the striatum in linguistic selection: Evidence from Huntington's disease and computational modeling.

Though accumulating evidence indicates that the striatum is recruited during language processing, the specific function of this subcortical structure in language remains to be elucidated. To answer this question, we used Huntington's disease as a model of striatal lesion. We investigated the morphological deficit of 30 early Huntington's disease patients with a novel linguistic task that can be modeled within an explicit theory of linguistic computation. Behavioral results reflected an impairment in HD patients on the linguistic task. Computational model-based analysis compared the behavioral data to simulated data from two distinct lesion models, a selection deficit model and a grammatical deficit model. This analysis revealed that the impairment derives from an increased randomness in the process of selecting between grammatical alternatives, rather than from a disruption of grammatical knowledge per se. Voxel-based morphometry permitted to correlate this impairment to dorsal striatal degeneration. We thus show that the striatum holds a role in the selection of linguistic alternatives, just as in the selection of motor and cognitive programs.

Learning diphone-based segmentation.

This paper reconsiders the diphone-based word segmentation model of Cairns, Shillcock, Chater, and Levy (1997) and Hockema (2006), previously thought to be unlearnable. A statistically principled learning model is developed using Bayes' theorem and reasonable assumptions about infants' implicit knowledge. The ability to recover phrase-medial word boundaries is tested using phonetic corpora derived from spontaneous interactions with children and adults. The (unsupervised and semi-supervised) learning models are shown to exhibit several crucial properties. First, only a small amount of language exposure is required to achieve the model's ceiling performance, equivalent to between 1 day and 1 month of caregiver input. Second, the models are robust to variation, both in the free parameter and the input representation. Finally, both the learning and baseline models exhibit undersegmentation, argued to have significant ramifications for speech processing as a whole.

Linking speech errors and phonological grammars: Insights from Harmonic Grammar networks.

Phonological grammars characterize distinctions between relatively well-formed (unmarked) and relatively ill-formed (marked) phonological structures. We review evidence that markedness influences speech error probabilities. Specifically, although errors result in both unmarked as well as marked structures, there is a markedness asymmetry: errors are more likely to produce unmarked outcomes. We show that stochastic disruption to the computational mechanisms realizing a Harmonic Grammar (HG) can account for the broad empirical patterns of speech errors. We demonstrate that our proposal can account for the general markedness asymmetry. We also develop methods for linking particular HG proposals to speech error distributions, and illustrate these methods using a simple HG and a set of initial consonant errors in English.