Indirect evidence and the poverty of the stimulus: the case of anaphoric one

It is widely held that children's linguistic input underdetermines the correct grammar, and that language learning must therefore be guided by innate linguistic constraints. Here, we show that a Bayesian model can learn a standard poverty-of-stimulus example, anaphoric one , from realistic input by relying on indirect evidence, without a linguistic constraint assumed to be necessary. Our demonstration does, however, assume other linguistic knowledge; thus, we reduce the problem of learning anaphoric one to that of learning this other knowledge. We discuss whether this other knowledge may itself be acquired without linguistic constraints.     

Automatic syntactic computation and use of semantic information during sentence comprehension

Summary The paper represents a contribution to the issue of the differential effects of syntactic, semantic and pragmatic information in the process of sentence comprehension. The results of four experiments are reported, in which sentences containing syntactic, semantic and pragmatic violations were presented with different tasks to see how these violations affected language comprehension, how easily they were detected and whether they affected sentence comprehension even when they were not explicitly recognized. The main results can be summarized as follows. First, the most relevant cues used to comprehend a sentence are not so much syntactic, but semantic or pragmatic. Second, with difficult or anomalous linguistic input, syntactic cues become more important. Third, even when the reader is not aware of syntactic violations, these seem to affect his processing of the linguistic input. This suggests the hypothesis of an automatic computation of syntactic information, the output of which is, however, used only when evidence coming from other sources, semantic or pragmatic, is insufficient or not clear enough.     

What infants know about syntax but couldn't have learned: experimental evidence for syntactic structure at 18 months

Generative linguistic theory stands on the hypothesis that grammar cannot be acquired solely on the basis of an analysis of the input, but depends, in addition, on innate structure within the learner to guide the process of acquisition. This hypothesis derives from a logical argument, however, and its consequences have never been examined experimentally with infant learners. Challenges to this hypothesis, claiming that an analysis of the input is indeed sufficient to explain grammatical acquisition, have recently gained attention. We demonstrate with novel experimentation the insufficiency of this countervailing view. Focusing on the syntactic structures required to determine the antecedent for the pronoun one, we demonstrate that the input to children does not contain sufficient information to support unaided learning. Nonetheless, we show that 18-month-old infants do have command of the syntax of one. Because this syntactic knowledge could not have been gleaned exclusively from the input, infants' mastery of this aspect of syntax constitutes evidence for the contribution of innate structure within the learner in acquiring a grammar.     

Intraverbal acquisition of semantic concepts by preschoolers

Intraverbal learning is a process through which semantic knowledge is acquired from purely linguistic information. The concern of this study is to investigate the role of intraverbal learning in the preschool-aged child's acquisition of semantic concepts. Three- and 4-year-old children were either shown pictures illustrating novel concepts or they were told verbal definitions of novel concepts, the definitions containing information about key physical and functional properties. An additional manipulation consisted of including or omitting information about the superordinate category membership of the concept. Concept acquisition was assessed via a battery of comprehension tasks; these tasks assessed both referential and formal knowledge of the concepts. Results indicate a correspondence between the type of information received and the type of knowledge acquired, with intraverbal information resulting in broader knowledge than pictorial information. Purely pictorial information was associated with referential knowledge only, while intraverbal information enhanced both referential and formal knowledge.     

Constructing Semantic Models From Words,Images, and Emojis

A number of recent models of semantics combine linguistic information, derived from text corpora, and visual information, derived from image collections, demonstrating that the resulting multimodal models are better than either of their unimodal counterparts, in accounting for behavioral data. Empirical work on semantic processing has shown that emotion also plays an important role especially in abstract concepts; however, models integrating emotion along with linguistic and visual information are lacking. Here, we first improve on visual and affective representations, derived from state-of-the-art existing models, by choosing models that best fit available human semantic data and extending the number of concepts they cover. Crucially then, we assess whether adding affective representations (obtained from a neural network model designed to predict emojis from co-occurring text) improves the model's ability to fit semantic similarity/relatedness judgments from a purely linguistic and linguistic–visual model. We find that, given specific weights assigned to the models, adding both visual and affective representations improves performance, with visual representations providing an improvement especially for more concrete words, and affective representations improving especially the fit for more abstract words.     

Automatically deriving readers’ knowledge structures from texts

Latent semantic analysis (LSA) serves as both a theory and a method for representing the meaning of words based on a statistical analysis of their contextual usage (Foltz, 1996; Landauer & Dumais, 1997). In experiments in the domains of psychology and history, we compared the representation of readers’ knowledge structures of information learned from texts with the representation generated by LSA. Results indicated that LSA’s representation is similar to readers’ representations. In addition, the degree to which the reader’s representation is similar to LSA’s representation is indicative of the amount of knowledge the reader has acquired and of the reader’s reading ability. This approach has implications both as a model of learning from text and as a practical tool for performing knowledge assessment.     

Visual problem solving in the absence of lexical semantics: Evidence from dementia

A case of dissociation between visual and praxic skills and linguistic, referential, and semantic skills is described in a case of dementia. Testing revealed that her visual processing was intact and that she was capable of a wide range of visually based tasks. These included copying geometric figures and block designs; matching items on the basis of number, color, or relative size, matching upper and lower case letters, sequencing numerals, orienting pictures, coins, and letters appropriately, and solving problems based on the identification of geometric patterns or on the identification of logical sequences based on pattern alternation, number, and size increases. In contrast, she was completely incapable of matching pictures on the basis of their semantic or referential meaning; similarly she had lost all comprehension of language. It was argued that this case demonstrates the capacities of a visual system cut off from all symbolic and semantic processes. When examined with other case studies this case provides information about the nature of and interaction between modality-specific perceptual processing and semantic processing, and information about the various cognitive factors involved in visual gnosis. The CT scan indicated frontal atrophy and marked anterior temporal atrophy, a pattern that is consistent with Pick's disease. The case was discussed in terms of the correlation between the pattern of observed strengths and deficits and the neurological pattern of atrophy.     

The development of children's understanding of "because" and "so"

Children's (5 years, 8 months to 8 years, 4 months) comprehension of "because" and "so" was assessed on both enactment and sentence completion tasks. The use of a within-subject design permitted performance on each task to be interpreted in terms of performance on the other. The pattern of results provided evidence against a componential model for the acquisition of causal connectives. This model holds that the two meaning components (cause, order) are acquired separately, with the order component being acquired several years later than the causal component. The results also were in accord with predictions derived from a contextual model of children's developing understanding of terms expressing logical relationships. This model posits that the understanding of relational terms is initially context dependent, such that children can understand relational terms when they express familiar relationships, but experience difficulty in forming representations of novel relationships solely on the basis of linguistic input. The lexical components of because and so are understood during the context-dependent stage, and the transition from context-dependent to context-independent understanding does not reflect further lexical development, but rather increasing facility in dealing with decontextualized linguistic input.     

Semantic knowledge as a determinant of developmental differences in recall

Two experiments examined the possible role of children's semantic knowledge and their ability to encode it in a cued-recall test. Performance of children aged 7, 10, and 13 was observed in encoding specificity tasks, which used homographs as the to-be-remembered words (TBRs).In Experiment I, children of all ages performed better when the output cue words prompted similar meanings to the input cues, than when output and input meanings were incompatible. In Experiment II, all children were required to recall identical items, but younger children, owing to lack of knowledge, were unable to encode pairs comprising input and TBR words incompatibly with output cues. In such circumstances, younger subjects outperformed older subjects. When input and output cues were incompatible for all ages, recall increased with age, possibly due to older childrens' using more effective retrieval strategies.The findings indicate that semantic encoding influenced recall performance at each age. This suggests that age differences in memory are due to differences in the amount of semantic knowledge a child possesses for encoding and cuing and not to capacity differences.     

The role of corpus size and syntax in deriving lexico-semantic representations for a wide range of concepts

One of the most significant recent advances in the study of semantic processing is the advent of models based on text and other corpora. In this study, we address what impact both the quantitative and qualitative properties of corpora have on mental representations derived from them. More precisely, we evaluate models with different linguistic and mental constraints on their ability to predict semantic relatedness between items from a vast range of domains and categories. We find that a model based on syntactic dependency relations captures significantly less of the variability for all kinds of words, regardless of the semantic relation between them or their abstractness. The largest difference was found for concrete nouns, which are commonly used to assess semantic processing. For both models we find that limited amounts of data suffice in order to obtain reliable predictions. Together, these findings suggest new constraints for the construction of mental models from corpora, both in terms of the corpus size and in terms of the linguistic properties that contribute to mental representations.     

听觉条件下词汇获得年龄效应产生机制的初探

通过三个实验考察听觉词汇的获得年龄效应。实验1采用基于语音的词汇判断任务,在错误率上存在词汇获得年龄效应,即早期习得词汇的正确率高于晚期习得词汇;实验2、3分别采用基于字形和语义的词汇判断任务,在反应时上存在词汇获得年龄效应,即早期习得词汇的反应快于晚期习得词汇。结果支持任意映射假说的观点：⒈词汇获得年龄效应产生于音、形、义加工的多个方面;⒉映射程度高低是决定词汇获得年龄效应大小的重要因素。     

Single-word predictions of upcoming language during comprehension: Evidence from the cumulative semantic interference task

Comprehenders predict upcoming speech and text on the basis of linguistic input. How many predictions do comprehenders make for an upcoming word? If a listener strongly expects to hear the word “sock”, is the word “shirt” partially expected as well, is it actively inhibited, or is it ignored? The present research addressed these questions by measuring the “downstream” effects of prediction on the processing of subsequently presented stimuli using the cumulative semantic interference paradigm. In three experiments, subjects named pictures (sock) that were presented either in isolation or after strongly constraining sentence frames (“After doing his laundry, Mark always seemed to be missing one …”). Naming sock slowed the subsequent naming of the picture shirt – the standard cumulative semantic interference effect. However, although picture naming was much faster after sentence frames, the interference effect was not modulated by the context (bare vs. sentence) in which either picture was presented. According to the only model of cumulative semantic interference that can account for such a pattern of data, this indicates that comprehenders pre-activated and maintained the pre-activation of best sentence completions (sock) but did not maintain the pre-activation of less likely completions (shirt). Thus, comprehenders predicted only the most probable completion for each sentence.     

Semantic priming effects for stimuli presented to the right and left visual fields

Studies that have addressed the possibility of hemispheric differences in semantic priming effects have yielded contradictory results. This paper reports the findings of two experiments intended to shed greater light on the issue of hemispheric differences in semantic priming. Experiment 1 used a hemiretinal paradigm and examined manual response latency and response accuracy to four types of word pairs; categorically related, syntactically related, unrelated, and pairs containing a nonword member. Experiment 2 examined the effects of unrecognized, disambiguating flank words on verbal responses to a centrally presented homograph. Experiment 1 yielded no significant visual field differences in magnitude of priming effects when response latency served as the dependent measure, although categorical relatedness facilitated response accuracy for left but not right visual field stimuli. In experiment 2, the disambiguating words had a significant effect on meaning interpretation of the homographs that was independent of visual field of presentation. Taken together, the results of these studies are interpreted as indicating that semantic aspects of linguistic input are automatically processed and can influence the content and latency of subsequent responses, whether presented to the left or right visual field.     

Effects of context-induced processing operations on semantic information acquired from discourse

Effects of context on processes involved in understanding discourse were investigated. An experimental context designed to directly affect discourse processing by inducing subjects to generate inferences involving text content was compared to a context in which subjects simply listened to and recalled the content of a text. The extent of inferential discourse processing was estimated from derived semantic information in subjects' text recalls. The experimental context produced effects on inferred and other derived information in subjects' text recalls which were predicted from a constructive model of discourse processing. Also consistent with this model, the amount of derived information increased with repeated exposures to the text. In a third condition, subjects repeatedly listened to the text in the experimental context, but only recalled it once on the last trial. The context produced very substantial effects on proportions of derived information under this condition. Context effects were still present one week later.     

A Probabilistic Model of Lexical and Syntactic Access and Disambiguation

The problems of access—retrieving linguistic structure from some mental grammar —and disambiguation—choosing among these structures to correctly parse ambiguous linguistic input—are fundamental to language understanding. The literature abounds with psychological results on lexical access, the access of idioms, syntactic rule access, parsing preferences, syntactic disambiguation, and the processing of garden-path sentences. Unfortunately, it has been difficult to combine models which account for these results to build a general, uniform model of access and disambiguation at the lexical, idiomatic, and syntactic levels. For example, psycholinguistic theories of lexical access and idiom access and parsing theories of syntactic rule access have almost no commonality in methodology or coverage of psycholinguistic data. This article presents a single probabilistic algorithm which models both the access and disambiguation of linguistic knowledge. The algorithm is based on a parallel parser which ranks constructions for access, and interpretations for disambiguation, by their conditional probability. Low-ranked constructions and interpretations are pruned through beam-search; this pruning accounts, among other things, for the garden-path effect. I show that this motivated probabilistic treatment accounts for a wide variety of psycholinguistic results, arguing for a more uniform representation of linguistic knowledge and for the use of probabilistically-enriched grammars and interpreters as models of human knowledge of and processing of language.     

Linguistic and psychometric factors in propositional reasoning

An experiment is reported which demonstrates the effects of linguistic and psychometric factors on adults' performance on a propositional reasoning task. The three linguistic factors were the semantic content in which the logical rule was embedded, the linguistic form of the logical rule, and the polarity of the major premise of the logical arguments. The two psychometric factors were the mode of response and the order of presentation of the different types of content. The results showed that the linguistic factors had a pronounced effect on adults' propositional reasoning abilities, whereas the influence of the psychometric factors was negligible. These findings are discussed in relation to operational and interpretational factors in reasoning.     

Representing the meanings of object and action words: the featural and unitary semantic space hypothesis

This paper presents the Featural and Unitary Semantic Space (FUSS) hypothesis of the meanings of object and action words. The hypothesis, implemented in a statistical model, is based on the following assumptions: First, it is assumed that the meanings of words are grounded in conceptual featural representations, some of which are organized according to modality. Second, it is assumed that conceptual featural representations are bound into lexico-semantic representations that provide an interface between conceptual knowledge and other linguistic information (syntax and phonology). Finally, the FUSS model employs the same principles and tools for objects and actions, modeling both domains in a single semantic space. We assess the plausibility of the model by showing that it can capture generalizations presented in the literature, in particular those related to category-related deficits, and show that it can predict semantic effects in behavioral experiments for object and action words better than other models such as Latent Semantic Analysis (Landauer & Dumais, 1997) and similarity metrics derived from Wordnet (Miller & Fellbaum, 1991).     

Applying automated deduction to natural language understanding

Very few natural language understanding applications employ methods from automated deduction. This is mainly because (i) a high level of interdisciplinary knowledge is required, (ii) there is a huge gap between formal semantic theory and practical implementation, and (iii) statistical rather than symbolic approaches dominate the current trends in natural language processing. Moreover, abduction rather than deduction is generally viewed as a promising way to apply reasoning in natural language understanding. We describe three applications where we show how first-order theorem proving and finite model construction can efficiently be employed in language understanding.The first is a text understanding system building semantic representations of texts, developed in the late 1990s. Theorem provers are here used to signal inconsistent interpretations and to check whether new contributions to the discourse are informative or not. This application shows that it is feasible to use general-purpose theorem provers for first-order logic, and that it pays off to use a battery of different inference engines as in practice they complement each other in terms of performance.The second application is a spoken-dialogue interface to a mobile robot and an automated home. We use the first-order theorem prover spass for checking inconsistencies and newness of information, but the inference tasks are complemented with the finite model builder mace used in parallel to the prover. The model builder is used to check for satisfiability of the input; in addition, the produced finite and minimal models are used to determine the actions that the robot or automated house has to execute. When the semantic representation of the dialogue as well as the number of objects in the context are kept fairly small, response times are acceptable to human users.The third demonstration of successful use of first-order inference engines comes from the task of recognising entailment between two (short) texts. We run a robust parser producing semantic representations for both texts, and use the theorem prover vampire to check whether one text entails the other. For many examples it is hard to compute the appropriate background knowledge in order to produce a proof, and the model builders mace and paradox are used to estimate the likelihood of an entailment.     

The acquisition of morphological knowledge investigated through artificial language learning

Affix knowledge plays an important role in visual word recognition, but little is known about how it is acquired. The authors present a new method of investigating the acquisition of affixes in which participants are trained on novel affixes presented in novel word contexts (e.g., sleepnept). Experiment 1 investigated the role of semantic information on affix acquisition by comparing a form-learning condition with a condition in which participants also received definitions for each novel word. Experiment 2 investigated the role of long-term consolidation on affix acquisition by comparing knowledge of learned affixes two days and nearly two months after training. Results demonstrated that episodic knowledge of affixes can be acquired shortly after a single training session using either form or semantic learning, but suggested that the development of lexicalized representations of affixes requires the provision of semantic information during learning as well as a substantial period of offline consolidation.