Graph‐Theoretic Properties of Networks Based on Word Association Norms: Implications for Models of Lexical Semantic Memory

We compared the ability of three different contextual models of lexical semantic memory (BEAGLE, Latent Semantic Analysis, and the Topic model) and of a simple associative model (POC) to predict the properties of semantic networks derived from word association norms. None of the semantic models were able to accurately predict all of the network properties. All three contextual models over‐predicted clustering in the norms, whereas the associative model under‐predicted clustering. Only a hybrid model that assumed that some of the responses were based on a contextual model and others on an associative network (POC) successfully predicted all of the network properties and predicted a word's top five associates as well as or better than the better of the two constituent models. The results suggest that participants switch between a contextual representation and an associative network when generating free associations. We discuss the role that each of these representations may play in lexical semantic memory. Concordant with recent multicomponent theories of semantic memory, the associative network may encode coordinate relations between concepts (e.g., the relation between pea and bean, or between sparrow and robin), and contextual representations may be used to process information about more abstract concepts.     

Spicy Adjectives and Nominal Donkeys: Capturing Semantic Deviance Using Compositionality in Distributional Spaces

Sophisticated senator and legislative onion. Whether or not you have ever heard of these things, we all have some intuition that one of them makes much less sense than the other. In this paper, we introduce a large dataset of human judgments about novel adjective‐noun phrases. We use these data to test an approach to semantic deviance based on phrase representations derived with compositional distributional semantic methods, that is, methods that derive word meanings from contextual information, and approximate phrase meanings by combining word meanings. We present several simple measures extracted from distributional representations of words and phrases, and we show that they have a significant impact on predicting the acceptability of novel adjective‐noun phrases even when a number of alternative measures classically employed in studies of compound processing and bigram plausibility are taken into account. Our results show that the extent to which an attributive adjective alters the distributional representation of the noun is the most significant factor in modeling the distinction between acceptable and deviant phrases. Our study extends current applications of compositional distributional semantic methods to linguistically and cognitively interesting problems, and it offers a new, quantitatively precise approach to the challenge of predicting when humans will find novel linguistic expressions acceptable and when they will not.     

Probing Lexical Ambiguity: Word Vectors Encode Number and Relatedness of Senses

Lexical ambiguity—the phenomenon of a single word having multiple, distinguishable senses—is pervasive in language. Both the degree of ambiguity of a word (roughly, its number of senses) and the relatedness of those senses have been found to have widespread effects on language acquisition and processing. Recently, distributional approaches to semantics, in which a word's meaning is determined by its contexts, have led to successful research quantifying the degree of ambiguity, but these measures have not distinguished between the ambiguity of words with multiple related senses versus multiple unrelated meanings. In this work, we present the first assessment of whether distributional meaning representations can capture the ambiguity structure of a word, including both the number and relatedness of senses. On a very large sample of English words, we find that some, but not all, distributional semantic representations that we test exhibit detectable differences between sets of monosemes (unambiguous words; N = 964), polysemes (with multiple related senses; N = 4,096), and homonyms (with multiple unrelated senses; N = 355). Our findings begin to answer open questions from earlier work regarding whether distributional semantic representations of words, which successfully capture various semantic relationships, also reflect fine-grained aspects of meaning structure that influence human behavior. Our findings emphasize the importance of measuring whether proposed lexical representations capture such distinctions: In addition to standard benchmarks that test the similarity structure of distributional semantic models, we need to also consider whether they have cognitively plausible ambiguity structure.     

Semantic classification of pictures and words

We provide new behavioural norms for semantic classification of pictures and words. The picture stimuli are 288 black and white line drawings from the International Picture Naming Project ([Székely, A., Jacobsen, T., D'Amico, S., Devescovi, A., Andonova, E., Herron, D., et al. (2004). A new on-line resource for psycholinguistic studies. Journal of Memory & Language, 51, 247–250]). We presented these pictures for classification in a living/nonliving decision, and in a separate version of the task presented the corresponding word labels for classification. We analyzed behavioural responses to a subset of the stimuli in order to explore questions about semantic processing. We found multiple semantic richness effects for both picture and word classification. Further, while lexical-level factors were related to semantic classification of words, they were not related to semantic classification of pictures. We argue that these results are consistent with privileged semantic access for pictures, and point to ways in which these data could be used to address other questions about picture processing and semantic memory.     

Semantic boost on episodic associations: an empirically-based computational model

Words become associated following repeated co-occurrence episodes. This process might be further determined by the semantic characteristics of the words. The present study focused on how semantic and episodic factors interact in incidental formation of word associations. First, we found that human participants associate semantically related words more easily than unrelated words; this advantage increased linearly with repeated co-occurrence. Second, we developed a computational model, SEMANT, suggesting a possible mechanism for this semantic-episodic interaction. In SEMANT, episodic associations are implemented through lateral connections between nodes in a pre-existent self-organized map of word semantics. These connections are strengthened at each instance of concomitant activation, proportionally with the amount of the overlapping activity waves of activated nodes. In computer simulations SEMANT replicated the dynamics of associative learning in humans and led to testable predictions concerning normal associative learning as well as impaired learning in a diffuse semantic system like that characteristic of schizophrenia.     

Visual and Affective Multimodal Models of Word Meaning in Language and Mind

One of the main limitations of natural language-based approaches to meaning is that they do not incorporate multimodal representations the way humans do. In this study, we evaluate how well different kinds of models account for people's representations of both concrete and abstract concepts. The models we compare include unimodal distributional linguistic models as well as multimodal models which combine linguistic with perceptual or affective information. There are two types of linguistic models: those based on text corpora and those derived from word association data. We present two new studies and a reanalysis of a series of previous studies. The studies demonstrate that both visual and affective multimodal models better capture behavior that reflects human representations than unimodal linguistic models. The size of the multimodal advantage depends on the nature of semantic representations involved, and it is especially pronounced for basic-level concepts that belong to the same superordinate category. Additional visual and affective features improve the accuracy of linguistic models based on text corpora more than those based on word associations; this suggests systematic qualitative differences between what information is encoded in natural language versus what information is reflected in word associations. Altogether, our work presents new evidence that multimodal information is important for capturing both abstract and concrete words and that fully representing word meaning requires more than purely linguistic information. Implications for both embodied and distributional views of semantic representation are discussed.     

Semantic meaning and associative responding

Abstract.— The hypothesis tested was that the search for a response word in a single word, free association situation proceeds in the direction of decreasing semantic similarity between the stimulus and response words. Semantic features were assigned to the 50 first Kent-Rosanoff words and each of the three most frequent response words to each stimulus word, obtained in a single word, free association situation (N =176). An index for the number of semantic features shared by the stimulus and response words was found to decrease with increasing response latency, thus confirming the hypothesis. The findings further supports the assumption that semantic meanings of stimulus and response words are determinants of free. associative responding.     

Discriminating Semantic from Episodic Relatedness in Young and Older Adults

ABSTRACT

The ability of young (aged 18–30) and older (aged 60–80) adults to discriminate pre-experimental (semantic) from experimental (episodic) associations was examined. Participants studied a list containing semantically related and unrelated word pairs and then made either associative recognition (Experiments 1a and b) or semantic relatedness (Experiment 2) judgments at various response deadlines. For associative recognition judgments, both young and older adults benefited from semantic relatedness, leading to more hits for related than unrelated pairs, and at the long response deadline, older adults' performance on those pairs matched that of young participants. Also, both young and older adults demonstrated superior discrimination for unrelated lures whose members had originally been studied in related pairs – evidence for recall-to-reject processing in both age groups. In making semantic relatedness judgments, both young and older adults showed an episodic priming effect. When older adults can rely on long-standing associations, their performance resembles that of young adults – both in associative recognition and in episodic priming.     

The Role of Negative Information in Distributional Semantic Learning

Distributional models of semantics learn word meanings from contextual co‐occurrence patterns across a large sample of natural language. Early models, such as LSA and HAL (Landauer & Dumais, 1997; Lund & Burgess, 1996), counted co‐occurrence events; later models, such as BEAGLE (Jones & Mewhort, 2007), replaced counting co‐occurrences with vector accumulation. All of these models learned from positive information only: Words that occur together within a context become related to each other. A recent class of distributional models, referred to as neural embedding models, are based on a prediction process embedded in the functioning of a neural network: Such models predict words that should surround a target word in a given context (e.g., word2vec; Mikolov, Sutskever, Chen, Corrado, & Dean, 2013). An error signal derived from the prediction is used to update each word's representation via backpropagation. However, another key difference in predictive models is their use of negative information in addition to positive information to develop a semantic representation. The models use negative examples to predict words that should not surround a word in a given context. As before, an error signal derived from the prediction prompts an update of the word's representation, a procedure referred to as negative sampling. Standard uses of word2vec recommend a greater or equal ratio of negative to positive sampling. The use of negative information in developing a representation of semantic information is often thought to be intimately associated with word2vec's prediction process. We assess the role of negative information in developing a semantic representation and show that its power does not reflect the use of a prediction mechanism. Finally, we show how negative information can be efficiently integrated into classic count‐based semantic models using parameter‐free analytical transformations.     

Spreading Activation in an Attractor Network With Latching Dynamics: Automatic Semantic Priming Revisited

Localist models of spreading activation (SA) and models assuming distributed representations offer very different takes on semantic priming, a widely investigated paradigm in word recognition and semantic memory research. In this study, we implemented SA in an attractor neural network model with distributed representations and created a unified framework for the two approaches. Our models assume a synaptic depression mechanism leading to autonomous transitions between encoded memory patterns (latching dynamics), which account for the major characteristics of automatic semantic priming in humans. Using computer simulations, we demonstrated how findings that challenged attractor‐based networks in the past, such as mediated and asymmetric priming, are a natural consequence of our present model’s dynamics. Puzzling results regarding backward priming were also given a straightforward explanation. In addition, the current model addresses some of the differences between semantic and associative relatedness and explains how these differences interact with stimulus onset asynchrony in priming experiments.     

Does<Emphasis Type="SmallCaps">LGHT</Emphasis> prime<Emphasis Type="SmallCaps">DARK</Emphasis>? Masked associative priming with addition neighbors

We present two masked priming lexical decision experiments in which we examined whether a nonword prime word would activate associative/semantic information from its corresponding addition neighbor (e.g.,lght-DARK via the addition neighborlight), producing associative/semantic priming. The rationale was the following: If a nonword prime with a missing letter produced a semantic/associative priming effect, this would clearly indicate that this nonword was activating the lexical/semantic representations of its base word, thereby reinforcing the models of visual-word recognition in which the orthographic representations produced bylght (orligt) andlight are quite similar (e.g., SOLAR, SERIOL, open-bigram, and overlap models). The results showed that the magnitude of the masked associative priming effect with subset primes was remarkably similar to that of the priming effect with the corresponding word prime. Furthermore, the magnitude of the associative priming effect was similar when the deleted letter was a vowel and when the deleted letter was a consonant.     

Semantic transparency in free stems: The effect of Orthography-Semantics Consistency on word recognition

A largely overlooked side effect in most studies of morphological priming is a consistent main effect of semantic transparency across priming conditions. That is, participants are faster at recognizing stems from transparent sets (e.g., farm) in comparison to stems from opaque sets (e.g., fruit), regardless of the preceding primes. This suggests that semantic transparency may also be consistently associated with some property of the stem word. We propose that this property might be traced back to the consistency, throughout the lexicon, between the orthographic form of a word and its meaning, here named Orthography-Semantics Consistency (OSC), and that an imbalance in OSC scores might explain the “stem transparency” effect. We exploited distributional semantic models to quantitatively characterize OSC, and tested its effect on visual word identification relying on large-scale data taken from the British Lexicon Project (BLP). Results indicated that (a) the “stem transparency” effect is solid and reliable, insofar as it holds in BLP lexical decision times (Experiment 1); (b) an imbalance in terms of OSC can account for it (Experiment 2); and (c) more generally, OSC explains variance in a large item sample from the BLP, proving to be an effective predictor in visual word access (Experiment 3).     

Semantic distance norms computed from an electronic dictionary (WordNet)

WordNet, an electronic dictionary (or lexical database), is a valuable resource for computational and cognitive scientists. Recent work on the computing of semantic distances among nodes (synsets) in WordNet has made it possible to build a large database of semantic distances for use in selecting word pairs for psychological research. The database now contains nearly 50,000 pairs of words that have values for semantic distance, associative strength, and similarity based on co-occurrence. Semantic distance was found to correlate weakly with these other measures but to correlate more strongly with another measure of semantic relatedness, featural similarity. Hierarchical clustering analysis suggested that the knowledge structure underlying semantic distance is similar in gross form to that underlying featural similarity. In experiments in which semantic similarity ratings were used, human participants were able to discriminate semantic distance. Thus, semantic distance as derived from WordNet appears distinct from other measures of word pair relatedness and is psychologically functional. This database may be downloaded fromwww.psychonomic.org/archive/.     

Feature overlap slows lexical selection: Evidence from the picture–word interference paradigm

How does the presence of a categorically related word influence picture naming latencies? In order to test competitive and noncompetitive accounts of lexical selection in spoken word production, we employed the picture–word interference (PWI) paradigm to investigate how conceptual feature overlap influences naming latencies when distractors are category coordinates of the target picture. Mahon et al. (2007. Lexical selection is not by competition: A reinterpretation of semantic interference and facilitation effects in the picture-word interference paradigm. Journal of Experimental Psychology. Learning, Memory, and Cognition, 33(3), 503–535. doi:10.1037/0278-7393.33.3.503) reported that semantically close distractors (e.g., zebra) facilitated target picture naming latencies (e.g., HORSE) compared to far distractors (e.g., whale). We failed to replicate a facilitation effect for within-category close versus far target–distractor pairings using near-identical materials based on feature production norms, instead obtaining reliably larger interference effects (Experiments 1 and 2). The interference effect did not show a monotonic increase across multiple levels of within-category semantic distance, although there was evidence of a linear trend when unrelated distractors were included in analyses (Experiment 2). Our results show that semantic interference in PWI is greater for semantically close than for far category coordinate relations, reflecting the extent of conceptual feature overlap between target and distractor. These findings are consistent with the assumptions of prominent competitive lexical selection models of speech production.     

Distributional analysis of semantic interference in picture naming

In picture–word interference experiments, participants name pictures (e.g., of a cat) while trying to ignore distractor words. Mean response time (RT) is typically longer with semantically related distractor words (e.g., dog) than with unrelated words (e.g., shoe), called semantic interference. Previous research has examined the RT distributional characteristics of distractor effects by performing ex-Gaussian analyses, which reveal whether effects are present in the normal part of the distribution (the μ parameter), its long right tail (the τ parameter), or both. One previous study linked the semantic interference effect selectively to the distribution tail. In the present study, we replicated the semantic interference effect in the mean picture naming RTs. Distributional analysis of the RTs and those of a previous study revealed that semantic interference was present in both μ and τ. These results provide evidence that the effect is not selectively linked to the τ parameter, and they warn against any simple one-to-one mapping between semantic interference and distributional parameters.     

Semantic relations differentially impact associative recognition memory: Electrophysiological evidence

Though associative recognition memory is thought to rely primarily on recollection, recent research indicates that familiarity might also make a substantial contribution when to-be-learned items are integrated into a coherent structure by means of an existing semantic relation. It remains unclear how different types of semantic relations, such as categorical (e.g., dancer–singer) and thematic (e.g., dancer–stage) relations might affect associative recognition, however. Using event-related potentials (ERPs), we addressed this question by manipulating the type of semantic link between paired words in an associative recognition memory experiment. An early midfrontal old/new effect, typically linked to familiarity, was observed across the relation types. In contrast, a robust left parietal old/new effect was found in the categorical condition only, suggesting a clear contribution of recollection to associative recognition for this kind of pairs. One interpretation of this pattern is that familiarity was sufficiently diagnostic for associative recognition of thematic relations, which could result from the integrative nature of the thematic relatedness compared to the similarity-based nature of categorical pairs. The present study suggests that the extent to which recollection and familiarity are involved in associative recognition is at least in part determined by the properties of semantic relations between the paired associates.     

Semantic distance norms computed from an electronic dictionary (WordNet).

WordNet, an electronic dictionary (or lexical database), is a valuable resource for computational and cognitive scientists. Recent work on the computing of semantic distances among nodes (synsets) in WordNet has made it possible to build a large database of semantic distances for use in selecting word pairs for psychological research. The database now contains nearly 50,000 pairs of words that have values for semantic distance, associative strength, and similarity based on co-occurrence. Semantic distance was found to correlate weakly with these other measures but to correlate more strongly with another measure of semantic relatedness, featural similarity. Hierarchical clustering analysis suggested that the knowledge structure underlying semantic distance is similar in gross form to that underlying featural similarity. In experiments in which semantic similarity ratings were used, human participants were able to discriminate semantic distance. Thus, semantic distance as derived from WordNet appears distinct from other measures of word pair relatedness and is psychologically functional. This database may be downloaded from www.psychonomic.org/archive/.     

Semantic feature production norms for a large set of objects and events

Semantic features produced by speakers of a language when given a word corresponding to a concept have provided insight into numerous behavioral phenomena concerning semantic representation in language-impaired and -unimpaired speakers. A number of theories concerning the organization of semantic memory have used features as their starting point. Here, we provide a set of feature norms collected from approximately 280 participants for a total of 456 words (169 nouns referring to objects, 71 nouns referring to events, and 216 verbs referring to events). Whereas a number of feature norms for object concepts already exist, we provide the first set of norms for event concepts. We have used these norms (for both objects and events) in research addressing questions concerning the similarities and differences between the semantic representation of objects and events and in research concerning the interface between semantics and syntax, given that events can be expressed in language as nouns or verbs. Some of this research is summarized here. These norms may be downloaded from www.psychonomic.org/archive.     

Semantic feature production norms for a large set of living and nonliving things

Semantic features have provided insight into numerous behavioral phenomena concerning concepts, categorization, and semantic memory in adults, children, and neuropsychological populations. Numerous theories and models in these areas are based on representations and computations involving semantic features. Consequently, empirically derived semantic feature production norms have played, and continue to play, a highly useful role in these domains. This article describes a set of feature norms collected from approximately 725 participants for 541 living (dog) and nonliving (chair) basic-level concepts, the largest such set of norms developed to date. This article describes the norms and numerous statistics associated with them. Our aim is to make these norms available to facilitate other research, while obviating the need to repeat the labor-intensive methods involved in collecting and analyzing such norms. The full set of norms may be downloaded from www.psychonomic.org/archive.