Artificial Language Learning Paradigm Research Articles

Semantic Cues Facilitate Structural Generalizations in Artificial Language Learning

ABSTRACT Natural languages contain systematic relationships between verb meaning and verb argument structure. Artificial language learning studies typically remove those relationships and instead pair verb meanings randomly with structures. Adult participants in such studies can detect statistical regularities associated with words in these languages and their use of novel words will adhere to those statistical regularities. However, word use in natural languages is associated with more than distributional statistics. Using an artificial language learning paradigm, we asked how a relationship between verb meaning and sentence structure affected learning and structure generalization. Twenty-four English-speaking adults watched videos described in an artificial language with two possible sentence structures. Half of the participants (statistics-only condition) learned a language with no relationship between verb meaning and sentence structure. The other half (semantics condition) learned a language in which verb meaning predicted which structures a verb occurred in. Although all learners were able to comprehend the learned structures with novel verbs, participants in the semantics conditions made grammaticality judgments and productions with novel verbs that were more consistent with the target language than participants in the statistics-only condition. The availability of semantic cues to verb subcategory supports artificial language learning.

Does Early Unit Size Impact the Formation of Linguistic Predictions? Grammatical Gender as a Case Study

AbstractMaking adults learn from larger linguistic units can facilitate learning article–noun agreement. Here we ask whether initial exposure to larger units improves learning by increasing the predictive associations between the article and noun. Using an artificial language learning paradigm, we taught 106 Hebrew‐speaking participants novel article–noun associations with either segmented input first or unsegmented input first, and tested their learning of the article–noun association and their ability to use articles to predict nouns. Our results showed that participants exposed to unsegmented input first were more likely to treat the article–noun unit as one word and were more accurate at learning the correct article–noun associations. However, participants in the unsegmented‐first condition did not show increased gaze to the target compared to those in the segmented‐first condition. We discuss how these findings inform our understanding of the challenges that adults face when learning a second language.

Experimental evidence: Directionality preference of tone absorption

Tone absorption is a very common tone alternation process in African and Asian tone languages, in which a part of the contour tone is absorbed byanother tone, and this process can take place in two directions. Typologically, right-dominant tone absorption is much more attested than the left-dominantcounterpart, although both are predicted possible. Such typological asymmetry hypothesizes the argument that there is a learnability difference between left- and right-dominant absorption. Specifically, given that the two patterns are equally complex, we hypothesized the difference is due to variations in phonetic groundedness. An artificial language learning paradigm was used for this study. We extracted possible tone absorption patterns from Tianjin Mandarin with reference to the Cantonese tonal inventory and designed an artificial language incorporating these absorption patterns. 56 native Cantonese participants took part in a word naming task, consisting of a pretest, a learning, and a test phase. They were taught a monosyllabic name of a color, and a monosyllabic name of a monster, and asked to combine them to form a disyllabic compound. Results showed that the right-dominant learners had significantly higher accuracy rates after learning. The analysis of reaction time (RT) revealed that right-dominant learners had significantly shorter RTs than the left-dominant group learners. Both findings showed that right-dominant tone absorption was better learned, suggesting that it is a universal directionality preference.

Low tone bias during perception of period doubling

Period-doubled voice consists of two alternating periods with multiple frequencies and is often perceived as rough with an indeterminate pitch. Past pitch-matching studies in period-doubled voice found that the perceived pitch was lower as the degree of amplitude and frequency modulation between the two alternating periods increased. The perceptual outcome also differed across f0s and modulation types: a lower f0 prompted earlier identification of a lower pitch, and the matched pitch dropped more quickly in frequency- than amplitude-modulated tokens (Sun & Xu, 2002; Bergan & Titze, 2001). However, it is unclear how listeners perceive period doubling when identifying linguistic tones. In an artificial language learning paradigm, this study used resynthesized stimuli with alternating amplitudes and/or frequencies of varying degrees, based on a production study of period-doubled voice (Huang, 2022). Listeners were native speakers of English and Mandarin. We confirm the positive relationship between the modulation degree and the proportion of low tones heard, and find that frequency modulation biased listeners to choose more low-tone options than amplitude modulation. However, a higher f0 (300 Hz) leads to a low-tone percept in more amplitude-modulated tokens than a lower f0 (200 Hz). Both English and Mandarin listeners behaved similarly, suggesting that pitch perception during period doubling is not language-specific. Furthermore, period doubling is predicted to signal low tones in languages, even when the f0 is high.

Speaking clearly improves speech segmentation by statistical learning under optimal listening conditions

This study investigated the effect of speaking style on speech segmentation by statistical learning under optimal and adverse listening conditions. Similar to the intelligibility and memory benefits found in previous studies, enhanced acoustic-phonetic cues of the listener-oriented clear speech could improve speech segmentation by statistical learning compared to conversational speech. Yet, it could not be precluded that hyper-articulated clear speech, reported to have less pervasive coarticulation, would result in worse segmentation than conversational speech. We tested these predictions using an artificial language learning paradigm. Listeners who acquired English before age six were played continuous repetitions of the ‘words’ of an artificial language, spoken either clearly or conversationally and presented either in quiet or in noise at a signal-to-noise ratio of +3 or 0 dB SPL. Next, they recognized the artificial words in a two-alternative forced-choice test. Results supported the prediction that clear speech facilitates segmentation by statistical learning more than conversational speech but only in the quiet listening condition. This suggests that listeners can use clear speech acoustic-phonetic enhancements to guide speech processing dependent on domain-general, signal-independent statistical computations. However, there was no clear speech benefit in noise at either signal-to-noise ratio. We discuss possible mechanisms that could explain these results.

The role of metacognition in recognition of the content of statistical learning.

Despite theoretical debate on the extent to which statistical learning is incidental or modulated by explicit instructions and conscious awareness of the content of statistical learning, no study has ever investigated the metacognition of statistical learning. We used an artificial language-learning paradigm and a segmentation task that required splitting a continuous stream of syllables into discrete recurrent constituents. During this task, statistical learning potentially produces knowledge of discrete constituents as well as about statistical regularities that are embodied in familiarization input. We measured metacognitive sensitivity and efficiency (using hierarchical Bayesian modelling to estimate metacognitive sensitivity and efficiency) to probe the role of conscious awareness in recognition of constituents extracted from the familiarization input and recognition of novel constituents embodying the same statistical regularities as these extracted constituents. Novel constituents are conceptualized to represent recognition of statistical structure rather than recognition of items retrieved from memory as whole constituents. We found that participants are equally sensitive to both types of learning products, yet subject them to varying degrees of conscious processing during the postfamiliarization recognition test. The data point to the contribution of conscious awareness to at least some types of statistical learning content.

Individual Differences in Learning Abilities Impact Structure Addition: Better Learners Create More Structured Languages.

Over the last decade, iterated learning studies have provided compelling evidence for the claim that linguistic structure can emerge from non-structured input, through the process of transmission. However, it is unclear whether individuals differ in their tendency to add structure, an issue with implications for understanding who are the agents of change. Here, we identify and test two contrasting predictions: The first sees learning as a pre-requisite for structure addition, and predicts a positive correlation between learning accuracy and structure addition, whereas the second maintains that it is those learners who struggle with learning and reproducing their input who add structure to it. This prediction is hard to test in standard iterated learning paradigms since each learner is exposed to a different input, and since structure and accuracy are computed using the same test items. Here, we test these contrasting predictions in two experiments using a one-generation artificial language learning paradigm designed to provide independent measures of learning accuracy and structure addition. Adults (N=48 in each study) were exposed to a semi-regular language (with probabilistic structure) and had to learn it: Learning was assessed using seen items, whereas structure addition was calculated over unseen items. In both studies, we find a strong positive correlation between individuals' ability to learn the language and their tendency to add structure to it: Better learners also produced more structured languages. These findings suggest a strong link between learning and generalization. We discuss the implications of these findings for iterated language models and theories of language change more generally.

The Role of Case Marking and Word Order in Cross‐Linguistic Structural Priming in Late L2 Acquisition

AbstractSeveral studies found cross‐linguistic structural priming with various language combinations. Here, we investigated the role of two important domains of language variation: case marking and word order, for transitive and ditransitive structures. We varied these features in an artificial language learning paradigm, using three different artificial language versions in a between‐subjects design. Priming was assessed between Dutch (no overt case marking, SVO word order) and a) an SVO order version, b) a case marking version, and c) an SOV order version. Similar within‐language and cross‐linguistic priming was found in all versions for transitives, indicating that cross‐linguistic structural priming was not hindered. In contrast, for ditransitives we found similar within‐language priming for all versions, but no cross‐linguistic priming. The finding that cross‐linguistic priming is possible between languages that vary in morphological marking or word order, is compatible with studies showing cross‐linguistic priming between natural languages that differ on these dimensions.

Learners’ harmonic preferences are modulated by lexical retrieval difficulty

Typological work has established the existence of language universals – features or combinations of features that (co-) occur in unrelated languages more frequently than expected by chance. The origins of language universals are a fundamental question in language sciences as these universals are considered a reflection of cognitive mechanisms underlying human language. In this study, we use a miniature artificial language learning paradigm to explore whether a well-known universal – a preference for harmonic word ordering between adpositional and verb phrases (i.e., placing the head in a consistent position with respect to the complements across the two phrase types) – originates in biases during language learning and whether this preference interacts with memory constraints (operationalized as lexical retrieval difficulty). We first trained participants on miniature languages containing adpositional phrases of a deterministic word order (either prepositional or postpositional inputs) and then briefly exposed them to simple transitive sentences (verb-subject-object and subject-object-verb order, equally frequent in the input). At test, we asked learners to describe simple transitive scenes. We found that in the hard lexical retrieval condition, learners exposed to a postpositional language showed a preference for harmonic ordering but learners of the prepositional language did not, which is only partially consistent to the typological distribution. In the easy lexical retrieval condition, learners of neither the postpositional nor prepositional language showed a preference for harmonic ordering, indicating that this preference is modulated by memory constraints.

Phonetically natural rules benefit from a learning bias: a re-examination of vowel harmony and disharmony

Substance-based phonological theories predict that a preference for phonetically natural rules (those which reflect constraints on speech production and perception) is encoded in synchronic grammars, and translates into learning biases. Some previous work has shown evidence for such biases, but methodological concerns with these studies mean that the question warrants further investigation. We revisit this issue by focusing on the learning of palatal vowel harmony (phonetically natural) compared to disharmony (phonetically unnatural). In addition, we investigate the role of memory consolidation during sleep on rule learning. We use an artificial language learning paradigm with two test phases separated by twelve hours. We observe a robust effect of phonetic naturalness: vowel harmony is learned better than vowel disharmony. For both rules, performance remains stable after twelve hours, regardless of the presence or absence of sleep.

Neural bases of learning and recognition of statistical regularities.

Statistical learning is a set of cognitive mechanisms allowing for extracting regularities from the environment and segmenting continuous sensory input into discrete units. The current study used functional magnetic resonance imaging (fMRI) (N=25) in conjunction with an artificial language learning paradigm to provide new insight into the neural mechanisms of statistical learning, considering both the online process of extracting statistical regularities and the subsequent offline recognition of learned patterns. Notably, prior fMRI studies on statistical learning have not contrasted neural activation during the learning and recognition experimental phases. Here, we found that learning is supported by the superior temporal gyrus and the anterior cingulate gyrus, while subsequent recognition relied on the left inferior frontal gyrus. Besides, prior studies only assessed the brain response during the recognition of trained words relative to novel nonwords. Hence, a further key goal of this study was to understand how the brain supports recognition of discrete constituents from the continuous input versus recognition of mere statistical structure that is used to build new constituents that are statistically congruent with the ones from the input. Behaviorally, recognition performance indicated that statistically congruent novel tokens were less likely to be endorsed as parts of the familiar environment than discrete constituents. fMRI data showed that the left intraparietal sulcus and angular gyrus support the recognition of old discrete constituents relative to novel statistically congruent items, likely reflecting an additional contribution from memory representations for trained items.

The Language-specific Use of Fundamental Frequency Rise in Segmentation of an Artificial Language: Evidence from Listeners of Taiwanese Southern Min.

Experience with native-language prosody encourages language-specific strategies for speech segmentation. Conflicting findings from previous research suggest that these strategies may not be abstracted away from the acoustic manifestation of prosodic features in the native speech. Using the artificial language learning paradigm, the current study explores this possibility in connection with listeners of a lexical tone language called Taiwanese Southern Min (TSM). In TSM, the only rising lexical tone occurs almost only on the final syllable of the language's tone sandhi domain and is phonetically associated with final lengthening. Based on these observations, Experiment I examined what constituted a sufficient finality cue for use by TSM listeners to support segmentation: (a) final fundamental frequency (F0) rise only; or (b) final F0 rise conjoined with final lengthening. The results showed that segmentation was inhibited by the former cue but facilitated by the latter. Experiment II showed that the facilitation cannot be attributed entirely to final lengthening, as a null effect was found when final lengthening was the sole prosodic cue to segmentation. It is thus assumed that acoustic details as fine-grained as the lengthening of the rising tone are involved in the modulation of the segmentation strategy whereby TSM listeners perceive F0 rise as signaling finality. The inhibitory effect of final F0 rise alone found in Experiment I motivated Experiment III, which revealed that initial F0 rise in the absence of lengthening cues improved TSM listeners' segmentation. It is speculated that such use of initial F0 rise might reflect a cross-linguistic segmentation solution.

Neutralization and homophony avoidance in phonological learning

Previous research has suggested that homophony avoidance plays a role in constraining language change; in particular, phonological contrasts are less likely to be neutralized if doing so would greatly increase the amount of homophony in the language. Most of the research on homophony avoidance has focused on the history of real languages, comparing attested and unattested (hypothetical) phonological changes. In this study, we take a novel approach by focusing on the language learner. Using an artificial language learning paradigm, we show that learners are less likely to acquire neutralizing phonological rules compared to non-neutralizing rules, but only if the neutralizing rules create homophony between lexical items encountered during learning. The results indicate that learners are biased against phonological patterns that create homophony, which could have an influence on language change. The results also suggest that lexical learning and phonological learning are highly integrated.

The importance of correctly characterising the English spelling system when devising and evaluating methods of reading instruction: Comment on Taylor, Davis, and Rastle (2017).

Taylor, Davis, and Rastle employed an artificial language learning paradigm to compare phonics and meaning-based approaches to reading instruction. Adults were taught consonant, vowel, and consonant (CVC) words composed of novel letters when the mappings between letters and sounds were completely systematic and the mappings between letters and meaning were completely arbitrary. At test, performance on naming tasks was better following training that emphasised the phonological rather than the semantic mappings, whereas performance on semantic tasks was similar in the two conditions. The authors concluded that these findings support phonics for early reading instruction in English. However, in our view, these conclusions are not justified given that the artificial language mischaracterised both the phonological and semantic mappings in English. Furthermore, the way participants studied the arbitrary letter-meaning correspondences bears little relation to meaning-based strategies used in schools. To compare phonics with meaning-based instruction it must be determined whether phonics is better than alternative forms of instruction that fully exploit the regularities within the semantic route. This is rarely assessed because of a widespread and mistaken assumption that underpins so much basic and applied research, namely, that the main function of spellings is to represent sounds.

Grammaticalized number, implicated presuppositions, and the plural

Plural morphology exhibits differing interpretations across languages. For example, in downward entailing contexts in English, the plural receives a one or more (or inclusive) interpretation, whereas in Korean-like languages the plural always receives a more than one (or exclusive) interpretation, regardless of context. Previous experimental work using an artificial language suggests that such differences may follow from structural properties of these languages (Liter, Heffner & Schmitt 2017), namely lack of grammaticalization of the plural/singular distinction. In this paper we adopt Sauerland, Anderssen & Yatsushiro’s (2005) implicated presupposition analysis of the plural (the English plural is semantically unmarked, whereas the Korean plural is semantically marked, carrying a presupposition that the cardinality of its referent is greater than one) in order to test two hypotheses about the interpretation of the plural. Using an artificial language learning paradigm identical to that in Liter, Heffner & Schmitt (2017) with non-grammaticalized number but with a much greater frequency of singular/plural NPs in the input, we test (i) whether semantic markedness of the plural should be linked to the non-grammaticalization of the number paradigm; or (ii) whether semantic markedness follows from insufficient statistical evidence for simplifying the lexical entry for the plural. Our results show that participants continue to assign an exclusive interpretation to plural morphology under the scope of negation, which is compatible with the hypothesis that non-grammaticalized number entails semantic markedness.

Contextual variability and exemplar strength in phonotactic learning.

Phonotactics-constraints on the position and combination of speech sounds within syllables-are subject to statistical differences that gradiently affect speaker and listener behavior (e.g., Vitevitch & Luce, 1999). What statistical properties drive the acquisition of such constraints? Because they are naturally highly correlated, previous work has been unable to dissociate the contribution of 2 properties: contextual variability (the number of unique phonological contexts in which a phonotactic pattern appears) and exemplar strength (the overall number of times the pattern appears). Using an artificial language learning paradigm, 3 experiments disentangled the effects of variability and strength, indexed by type and token frequency, respectively, on the learning of gradient phonotactics. When the 2 factors were decorrelated (Experiment 2), participants showed greater generalization of patterns advantaged for contextual variability, but not those advantaged for exemplar strength. When the 2 factors were anticorrelated (Experiment 3), participants preferred patterns advantaged in contextual variability, even though they were disadvantaged for exemplar strength. These results suggest that contextual variability is the key force driving phonotactic learning, as it allows learners to home in on the invariant features of the input. (PsycINFO Database Record

Zipf’s Law of Abbreviation and the Principle of Least Effort: Language users optimise a miniature lexicon for efficient communication

The linguist George Kingsley Zipf made a now classic observation about the relationship between a word’s length and its frequency; the more frequent a word is, the shorter it tends to be. He claimed that this “Law of Abbreviation” is a universal structural property of language. The Law of Abbreviation has since been documented in a wide range of human languages, and extended to animal communication systems and even computer programming languages. Zipf hypothesised that this universal design feature arises as a result of individuals optimising form-meaning mappings under competing pressures to communicate accurately but also efficiently—his famous Principle of Least Effort. In this study, we use a miniature artificial language learning paradigm to provide direct experimental evidence for this explanatory hypothesis. We show that language users optimise form-meaning mappings only when pressures for accuracy and efficiency both operate during a communicative task, supporting Zipf’s conjecture that the Principle of Least Effort can explain this universal feature of word length distributions.

Discovery of a Recursive Principle: An Artificial Grammar Investigation of Human Learning of a Counting Recursion Language.

Learning is typically understood as a process in which the behavior of an organism is progressively shaped until it closely approximates a target form. It is easy to comprehend how a motor skill or a vocabulary can be progressively learned—in each case, one can conceptualize a series of intermediate steps which lead to the formation of a proficient behavior. With grammar, it is more difficult to think in these terms. For example, center embedding recursive structures seem to involve a complex interplay between multiple symbolic rules which have to be in place simultaneously for the system to work at all, so it is not obvious how the mechanism could gradually come into being. Here, we offer empirical evidence from a new artificial language (or “artificial grammar”) learning paradigm, Locus Prediction, that, despite the conceptual conundrum, recursion acquisition occurs gradually, at least for a simple formal language. In particular, we focus on a variant of the simplest recursive language, anbn, and find evidence that (i) participants trained on two levels of structure (essentially ab and aabb) generalize to the next higher level (aaabbb) more readily than participants trained on one level of structure (ab) combined with a filler sentence; nevertheless, they do not generalize immediately; (ii) participants trained up to three levels (ab, aabb, aaabbb) generalize more readily to four levels than participants trained on two levels generalize to three; (iii) when we present the levels in succession, starting with the lower levels and including more and more of the higher levels, participants show evidence of transitioning between the levels gradually, exhibiting intermediate patterns of behavior on which they were not trained; (iv) the intermediate patterns of behavior are associated with perturbations of an attractor in the sense of dynamical systems theory. We argue that all of these behaviors indicate a theory of mental representation in which recursive systems lie on a continuum of grammar systems which are organized so that grammars that produce similar behaviors are near one another, and that people learning a recursive system are navigating progressively through the space of these grammars.

Simultaneous segmentation and generalisation of non-adjacent dependencies from continuous speech

Language learning requires mastering multiple tasks, including segmenting speech to identify words, and learning the syntactic role of these words within sentences. A key question in language acquisition research is the extent to which these tasks are sequential or successive, and consequently whether they may be driven by distinct or similar computations. We explored a classic artificial language learning paradigm, where the language structure is defined in terms of non-adjacent dependencies. We show that participants are able to use the same statistical information at the same time to segment continuous speech to both identify words and to generalise over the structure, when the generalisations were over novel speech that the participants had not previously experienced. We suggest that, in the absence of evidence to the contrary, the most economical explanation for the effects is that speech segmentation and grammatical generalisation are dependent on similar statistical processing mechanisms.

THE USE OF PROSODIC CUES IN LEARNING NEW WORDS IN AN UNFAMILIAR LANGUAGE

The artificial language learning paradigm was used to investigate to what extent the use of prosodic features is universally applicable or specifically language driven in learning an unfamiliar language, and how nonnative prosodic patterns can be learned. Listeners of unrelated languages—Dutch (n= 100) and Korean (n= 100)—participated. The words to be learned varied with prosodic cues: no prosody, fundamental frequency (F0) rise in initial and final position, final lengthening, and final lengthening plus F0 rise. Both listener groups performed well above chance level with the final lengthening cue, confirming its crosslinguistic use. As for final F0 rise, however, Dutch listeners did not use it until the second exposure session, whereas Korean listeners used it at initial exposure. Neither group used initial F0 rise. On the basis of these results, F0 and durational cues appear to be universal in the sense that they are used across languages for their universally applicable auditory-perceptual saliency, but how they are used is language specific and constrains the use of available prosodic cues in processing a nonnative language. A discussion on how these findings bear on theories of second language (L2) speech perception and learning is provided.