The English Sublexical Toolkit: Methods for indexing sound-spelling consistency.

This work introduces the English Sublexical Toolkit, a suite of tools that utilizes an experience-dependent learning framework of sublexical knowledge to extract regularities from the English lexicon. The Toolkit quantifies the empirical regularity of sublexical units in both the reading and spelling directions (i.e., grapheme-to-phoneme and phoneme-to-grapheme) and at multiple grain sizes (i.e., phoneme/grapheme and onset/rime unit size). It can extract multiple experience-dependent regularity indices for words or pseudowords, including both frequency indices (e.g., grapheme frequency) and conditional probability indices (e.g., grapheme-to-phoneme probability). These tools provide (1) superior estimates of the regularities that better reflect the complexity of the sublexical system relative to previously published indices and (2) completely novel indices of sublexical units such as phonographeme frequency (i.e., combined units of individual phonemes and graphemes that are independent of processing direction). We demonstrate that measures from the toolkit explain significant amounts of variance in empirical data (naming of real words and lexical decision), and either outperform or are comparable to the best available consistency measures. The flexibility of the toolkit is further demonstrated by its ability to readily index the probability of different pseudowords pronunciations, and we report that the measures account for the majority of variance in these empirically observed probabilities. Overall, this work provides a framework and resources that can be flexibly used to identify optimal corpus-based consistency measures that help explain reading/spelling behaviors for real and pseudowords.

HeLP: The Hebrew Lexicon project.

Lexicon projects (LPs) are large-scale data resources in different languages that present behavioral results from visual word recognition tasks. Analyses using LP data in multiple languages provide evidence regarding cross-linguistic differences as well as similarities in visual word recognition. Here we present the first LP in a Semitic language-the Hebrew Lexicon Project (HeLP). HeLP assembled lexical decision (LD) responses to 10,000 Hebrew words and nonwords, and naming responses to a subset of 5000 Hebrew words. We used the large-scale HeLP data to estimate the impact of general predictors (lexicality, frequency, word length, orthographic neighborhood density), and Hebrew-specific predictors (Semitic structure, presence of clitics, phonological entropy) of visual word recognition performance. Our results revealed the typical effects of lexicality and frequency obtained in many languages, but more complex impact of word length and neighborhood density. Considering Hebrew-specific characteristics, HeLP data revealed better recognition of words with a Semitic structure than words that do not conform to it, and a drop in performance for words comprising clitics. These effects varied, however, across LD and naming tasks. Lastly, a significant inhibitory effect of phonological ambiguity was found in both naming and LD. The implications of these findings for understanding reading in a Semitic language are discussed.

NSP-SCD: A corpus construction protocol for child-directed print in understudied languages.

Child-directed print corpora enable systematic psycholinguistic investigations, but this research infrastructure is not available in many understudied languages. Moreover, researchers of understudied languages are dependent on manual tagging because precise automatized parsers are not yet available. One plausible way forward is to limit the intensive work to a small-sized corpus. However, with little systematic enquiry about approaches to corpus construction, it is unclear how robust a small corpus can be made. The current study examines the potential of a non-sequential sampling protocol for small corpus development (NSP-SCD) through a cross-corpora and within-corpus analysis. A corpus comprising 17,584 words was developed by applying the protocol to a larger corpus of 150,595 words from children's books for 3-to-10-year-olds. While the larger corpus will by definition have more instances of unique words and unique orthographic units, still, the selectively sampled small corpus approximated the larger corpus for lexical and orthographic diversity and was equivalent for orthographic representation and word length. Psycholinguistic complexity increased by book level and varied by parts of speech. Finally, in a robustness check of lexical diversity, the non-sequentially sampled small corpus was more efficient compared to a same-sized corpus constructed by simply using all sentences from a few books (402 books vs. seven books). If a small corpus must be used then non-sequential sampling from books stratified by book level makes the corpus statistics better approximate what is found in larger corpora. Overall, the protocol shows promise as a tool to advance the science of child language acquisition in understudied languages.

Affordance norms for 2825 concrete nouns.

Objects are commonly described based on their relations to other objects (e.g., associations, semantic similarity, etc.) or their physical features (e.g., birds have wings, feathers, etc.). However, objects can also be described in terms of their actionable properties (i.e., affordances), which reflect interactive relations between actors and objects. While several normed datasets have been developed to categorize various aspects of meaning (e.g., semantic features, cue-target associations, etc.), to date, norms for affordances have not been generated. We address this limitation by developing a set of affordance norms for 2825 concrete nouns. Using an open-response format, we computed affordance strength (AFS; i.e., the probability of an item eliciting a particular action response), affordance proportion (AFP; i.e., the proportion of participants who provided a specific action response), and affordance set size (AFSS; i.e., the total number of unique action responses) for each item. Because our stimuli overlapped with Pexman et al.'s, Behavior Research Methods, 51, 453-466, (2019) body-object interaction norms (BOI), we tested whether AFS, AFP, and AFSS were related to BOI, as objects with more perceived action properties may be viewed as being more interactive. Additionally, we tested the relationship between AFS and AFP and two separate measures of relatedness: cosine similarity (Buchanan et al., Behavior Research Methods, 51, 1849-1863, 2019a, Behavior Research Methods, 51, 1878-1888, 2019b) and forward associative strength (Nelson et al., Behavior Research Methods, Instruments, & Computers, 36(3), 402-407, 2004). All analyses, however, revealed weak relationships between affordance measures and existing semantic norms, suggesting that affordance properties reflect a separate construct.

LASTU: A psycholinguistic search tool for Finnish lexical stimuli.

LASTU is a tool for searching for Finnish language stimulus words for psycholinguistic studies. The tool allows the user to query a number of properties, including forms, lemmas, frequencies, and morphological features. It also includes two new measures for quantifying lemma and form ambiguity. The tool is written in Python and is available for Windows and macOS platforms. It is available at https://osf.io/j8v6b/ . Included with the tool is a database based on a massive corpus of dependency-parsed Finnish language data crawled from the Internet (over 5 billion tokens). While LASTU has been developed for researchers working on the Finnish language, the openly available implementation can also be applied to other languages.

Semantic feature norms: a cross-method and cross-language comparison.

The ability to assign meaning to perceptual stimuli forms the basis of human behavior and the ability to use language. The meanings of things have primarily been probed using behavioral production norms and corpus-derived statistical methods. However, it is not known to what extent the collection method and the language being probed influence the resulting semantic feature vectors. In this study, we compare behavioral with corpus-based norms, across Finnish and English, using an all-to-all approach. To complete the set of norms required for this study, we present a new set of Finnish behavioral production norms, containing both abstract and concrete concepts. We found that all the norms provide largely similar information about the relationships of concrete objects and allow item-level mapping across norms sets. This validates the use of the corpus-derived norms which are easier to obtain than behavioral norms, which are labor-intensive to collect, for studies that do not depend on subtle differences in meaning between close semantic neighbors.

Can large language models help augment English psycholinguistic datasets?

Research on language and cognition relies extensively on psycholinguistic datasets or "norms". These datasets contain judgments of lexical properties like concreteness and age of acquisition, and can be used to norm experimental stimuli, discover empirical relationships in the lexicon, and stress-test computational models. However, collecting human judgments at scale is both time-consuming and expensive. This issue of scale is compounded for multi-dimensional norms and those incorporating context. The current work asks whether large language models (LLMs) can be leveraged to augment the creation of large, psycholinguistic datasets in English. I use GPT-4 to collect multiple kinds of semantic judgments (e.g., word similarity, contextualized sensorimotor associations, iconicity) for English words and compare these judgments against the human "gold standard". For each dataset, I find that GPT-4's judgments are positively correlated with human judgments, in some cases rivaling or even exceeding the average inter-annotator agreement displayed by humans. I then identify several ways in which LLM-generated norms differ from human-generated norms systematically. I also perform several "substitution analyses", which demonstrate that replacing human-generated norms with LLM-generated norms in a statistical model does not change the sign of parameter estimates (though in select cases, there are significant changes to their magnitude). I conclude by discussing the considerations and limitations associated with LLM-generated norms in general, including concerns of data contamination, the choice of LLM, external validity, construct validity, and data quality. Additionally, all of GPT-4's judgments (over 30,000 in total) are made available online for further analysis.

Standardizing norms for 1286 colored pictures in Cantonese.

This study established psycholinguistic norms in Cantonese for a set of 1286 colored pictures sourced from several picture databases, including 750 colored line drawings from MultiPic (Duñabeitia et al., 2018) and 536 photographs selected for McRae et al. (2005) concepts. The pictures underwent rigorous normalization processes. We provided picture characteristics including name and concept agreement, familiarity, visual complexity, and frequency of modal responses. Through correlational analyses, we observed strong interrelationships among these variables. We also compared the current Cantonese norming to other languages and demonstrated similarity and variations among different languages. Additionally, we embraced the multilingual diversity within the current sample, and found that higher Cantonese proficiency but lower non-native language proficiency were associated with better spoken picture naming. Last but not least, we validated the predictive power of normed variables calculated from typed responses to spoken picture naming, and the consistency between typed and spoken responses. The present norming provides a timely and valuable alternative for researchers in the field of psycholinguistics, especially those studying Cantonese production and lexical retrieval. All raw data, analysis scripts, and final norming results are available online as psycholinguistic norms for Cantonese in the following link at the Open Science Framework: https://osf.io/dz9j6/?view_only=a452d8a56c92430b9dedf21ac26b1bc1 .

Concreteness ratings for 36,000 Estonian words.

We present a collection of concreteness ratings for 35,979 words in Estonian. The data were collected via a web application from 2278 native Estonian speakers. Human ratings of concreteness have not been collected for Estonian beforehand. We compare our results to Aedmaa et al. (2018), who assigned concreteness ratings to 240,000 Estonian words by means of machine learning. We show that while these two datasets show reasonable correlation (R = 0.71), there are considerable differences in the distribution of the ratings, which we discuss in this paper. Furthermore, the results also raise questions about the importance of the type of scale used for collecting ratings. While most other datasets have been compiled based on questionnaires entailing five- or seven-point Likert scales, we used a continuous 0-10 scale. Comparing our rating distribution to those of other studies, we found that it is most similar to the distribution in Lahl et al. (Behavior Research Methods, 41(1), 13-19, 2009), who also used a 0-10 scale. Concreteness ratings for Estonian words are available at OSF .

Capturing individual differences in sentence processing: How reliable is the self-paced reading task?

Advances in research on language processing have originally come from group-level comparisons, but there is now a growing interest in individual differences. To investigate individual differences, tasks that have shown robust group-level differences are often used with the implicit assumption that they will also be reliable when used as an individual differences measure. Here, we examined whether one of the primary tasks used in psycholinguistic research on language processing, the self-paced reading task, can reliably measure individual differences in relative clause processing. We replicated the well-established effects of relative clauses at the group level, with object relative clauses being more difficult to process than subject relative clauses. However, when using difference scores, the reliability of the size of the relative clause effect was close to zero because the self-paced reading times for the different relative clause types were highly correlated within individuals. Nonetheless, we found that the self-paced reading task can be used to reliably capture individual differences in overall reading speed as well as key sentence regions when the two types of relative clause sentences are considered separately. Our results indicate that both the reliability and validity of different sentence regions need to be assessed to determine whether and when self-paced reading can be used to examine individual differences in language processing.

TACO: A Turkish database for abstract concepts.

The organization of abstract concepts reflects different dimensions, grounded in the brain regions coding for the corresponding experience. Normative measures of linguistic stimuli offer noteworthy insights into the organization of conceptual knowledge, but studies differ in the dimensions and classes of concepts considered. Additionally, most of the available information has been collected in English, without considering possible linguistic and cultural differences. Here, we aimed to create a comprehensive Turkish database for abstract concepts (TACO), including rarely investigated classes such as political concepts. We included 503 words-78 concrete (fruits, animals, tools) and 425 abstract (emotions, social, mental states, theoretical, quantity, space, political)-rated by 134 Turkish speakers for familiarity, imageability, age of acquisition, valence, arousal, quantity, space, theoretical, social, mental state, and political dimensions. We calculated dominance and exclusivity, indicating the dimension receiving the highest mean score for each word, and the position of the word along the unidimensional-multidimensional continuum, respectively. A principal component analysis (PCA) was conducted on the semantic dimensions. The results showed that mental state was the dominant dimension for most concepts. Moderate to low levels of exclusivity indicated that the concepts were multidimensional. PCA revealed three components: Component 1 captured the juxtaposition between social/mental state and magnitude polarities, Component 2 highlighted affective components, and Component 3 grouped together political and theoretical dimensions. The introduction of political concepts provided insights into the multidimensional nature of this unexplored class, closely intertwined with the theoretical dimension. TACO constitutes the first comprehensive Turkish database covering several abstract dimensions, paving the way for cross-linguistic and cross-cultural studies of semantic representations.

Decoding the essence of two-character Chinese words: Unveiling valence, arousal, concreteness, familiarity, and imageability through word norming.

Investigation of affective and semantic dimensions of words is essential for studying word processing. In this study, we expanded Tse et al.'s (Behav Res Methods 49:1503-1519, 2017; Behav Res Methods 55:4382-4402, 2023) Chinese Lexicon Project by norming five word dimensions (valence, arousal, familiarity, concreteness, and imageability) for over 25,000 two-character Chinese words presented in traditional script. Through regression models that controlled for other variables, we examined the relationships among these dimensions. We included ambiguity, quantified by the standard deviation of the ratings of a given lexical variable across different raters, as separate variables (e.g., valence ambiguity) to explore their connections with other variables. The intensity-ambiguity relationships (i.e., between normed variables and their ambiguities, like valence with valence ambiguity) were also examined. In these analyses with a large pool of words and controlling for other lexical variables, we replicated the asymmetric U-shaped valence-arousal relationship, which was moderated by valence and arousal ambiguities. We also observed a curvilinear relationship between valence and familiarity and between valence and concreteness. Replicating Brainerd et al.'s (J Exp Psychol Gen 150:1476-1499, 2021; J Mem Lang 121:104286, 2021) quadratic intensity-ambiguity relationships, we found that the ambiguity of valence, arousal, concreteness, and imageability decreases as the value of these variables is extremely low or extremely high, although this was not generalized to familiarity. While concreteness and imageability were strongly correlated, they displayed different relationships with arousal, valence, familiarity, and valence ambiguity, suggesting their distinct conceptual nature. These findings further our understanding of the affective and semantic dimensions of two-character Chinese words. The normed values of all these variables can be accessed via https://osf.io/hwkv7 .

How ready is speech-to-text for psychological language research? Evaluating the validity of AI-generated English transcripts for analyzing free-spoken responses in younger and older adults.

For the longest time, the gold standard in preparing spoken language corpora for text analysis in psychology was using human transcription. However, such standard comes at extensive cost, and creates barriers to quantitative spoken language analysis that recent advances in speech-to-text technology could address. The current study quantifies the accuracy of AI-generated transcripts compared to human-corrected transcripts across younger (n = 100) and older (n = 92) adults and two spoken language tasks. Further, it evaluates the validity of Linguistic Inquiry and Word Count (LIWC)-features extracted from these two kinds of transcripts, as well as transcripts specifically prepared for LIWC analyses via tagging. We find that overall, AI-generated transcripts are highly accurate with a word error rate of 2.50% to 3.36%, albeit being slightly less accurate for younger compared to older adults. LIWC features extracted from either transcripts are highly correlated, while the tagging procedure significantly alters filler word categories. Based on these results, automatic speech-to-text appears to be ready for psychological language research when using spoken language tasks in relatively quiet environments, unless filler words are of interest to researchers.

The Chinese lexicon of deaf readers: A database of character decisions and a comparison between deaf and hearing readers.

We present a psycholinguistic study investigating lexical effects on simplified Chinese character recognition by deaf readers. Prior research suggests that deaf readers exhibit efficient orthographic processing and decreased reliance on speech-based phonology in word recognition compared to hearing readers. In this large-scale character decision study (25 participants, each evaluating 2500 real characters and 2500 pseudo-characters), we analyzed various factors influencing character recognition accuracy and speed in deaf readers. Deaf participants demonstrated greater accuracy and faster recognition when characters were more frequent, were acquired earlier, had more strokes, displayed higher orthographic complexity, were more imageable in reference, or were less concrete in reference. Comparison with a previous study of hearing readers revealed that the facilitative effect of frequency on character decision accuracy was stronger for deaf readers than hearing readers. The effect of orthographic-phonological regularity differed significantly for the two groups, indicating that deaf readers rely more on orthographic structure and less on phonological information during character recognition. Notably, increased stroke counts (i.e., higher orthographic complexity) hindered hearing readers but facilitated recognition processes in deaf readers, suggesting that deaf readers excel at recognizing characters based on orthographic structure. The database generated from this large-scale character decision study offers a valuable resource for further research and practical applications in deaf education and literacy.

Body-object interaction ratings for 3600 French nouns.

Body-object interaction (BOI) measures the ease with which the human body can interact with the concept represented by a word. This research focuses on two main objectives: first, to establish French norms for the psycholinguistic variable BOI, and second, to investigate the contribution of BOI to language processing in French. We collected BOI ratings for 3600 French nouns from participants through an online platform. The inter- and intrastudy reliability of these new ratings indicate that the ratings are robust. We then aimed to determine the role of BOI in word recognition. A hierarchical regression analysis was conducted using lexical decision reaction times (RTs) as the dependent variable. BOI was found to be a significant predictor of lexical decision latencies, beyond the contribution of word length, frequency, orthographic distinctiveness, and imageability. Contrary to previous findings in English, higher BOI values were associated with longer RTs in French, indicating an inhibitory effect of BOI on French word processing. Methodological differences may account for this divergent result. Taken together, the results of this study show the independent contribution of BOI to word recognition in French. This supports the notion that sensorimotor information is a crucial component of language processing. By providing a reliable and sizable BOI database for French nouns, we offer a valuable resource for psycholinguistic and language processing research. This research underscores the complex relationship between language, cognition, and sensorimotor experiences, advancing our comprehension of language processing mechanisms.

Spectral fingerprints or spectral tilt? Evidence for distinct oscillatory signatures of memory formation.

Decreases in low-frequency power (2-30 Hz) alongside high-frequency power increases (>40 Hz) have been demonstrated to predict successful memory formation. Parsimoniously, this change in the frequency spectrum can be explained by one factor, a change in the tilt of the power spectrum (from steep to flat) indicating engaged brain regions. A competing view is that the change in the power spectrum contains several distinct brain oscillatory fingerprints, each serving different computations. Here, we contrast these two theories in a parallel magnetoencephalography (MEG)-intracranial electroencephalography (iEEG) study in which healthy participants and epilepsy patients, respectively, studied either familiar verbal material or unfamiliar faces. We investigated whether modulations in specific frequency bands can be dissociated in time and space and by experimental manipulation. Both MEG and iEEG data show that decreases in alpha/beta power specifically predicted the encoding of words but not faces, whereas increases in gamma power and decreases in theta power predicted memory formation irrespective of material. Critically, these different oscillatory signatures of memory encoding were evident in different brain regions. Moreover, high-frequency gamma power increases occurred significantly earlier compared to low-frequency theta power decreases. These results show that simple "spectral tilt" cannot explain common oscillatory changes and demonstrate that brain oscillations in different frequency bands serve different functions for memory encoding.

The Cortical Organization of Syntax.

Syntax, the structure of sentences, enables humans to express an infinite range of meanings through finite means. The neurobiology of syntax has been intensely studied but with little consensus. Two main candidate regions have been identified: the posterior inferior frontal gyrus (pIFG) and the posterior middle temporal gyrus (pMTG). Integrating research in linguistics, psycholinguistics, and neuroscience, we propose a neuroanatomical framework for syntax that attributes distinct syntactic computations to these regions in a unified model. The key theoretical advances are adopting a modern lexicalized view of syntax in which the lexicon and syntactic rules are intertwined, and recognizing a computational asymmetry in the role of syntax during comprehension and production. Our model postulates a hierarchical lexical-syntactic function to the pMTG, which interconnects previously identified speech perception and conceptual-semantic systems in the temporal and inferior parietal lobes, crucial for both sentence production and comprehension. These relational hierarchies are transformed via the pIFG into morpho-syntactic sequences, primarily tied to production. We show how this architecture provides a better account of the full range of data and is consistent with recent proposals regarding the organization of phonological processes in the brain.

Presuppositions of determiners are immediately used to disambiguate utterance meaning: A mouse-tracking study on the German language.

The present study investigated how listeners understand and process the definite and the indefinite determiner. While the definite determiner clearly conveys a uniqueness presupposition, the status of the anti-uniqueness inference associated with the indefinite determiner is less clear. In a forced choice production task, we observed that participants make use of the information about number usually associated with the two determiners to convey a message. In a subsequent mouse-tracking task, participants had to select one of two potential referents presented on screen according to an auditorily presented stimulus sentence. The data revealed that participants use the information about uniqueness or anti-uniqueness encoded in determiners to disambiguate sentence meaning as early as possible, but only when they are exclusively faced with felicitous uses of determiners.

Novel stress phonotactics are learnable by English speakers: Novel tone phonotactics are not.

Speech errors are sensitive to newly learned phonotactic constraints. For example, if speakers produce strings of syllables in which /f/ is an onset if the vowel is /æ/, but a coda if the vowel is /I/, their slips will respect that constraint after a period of sleep. Constraints in which the contextual factor is nonlinguistic, however, do not appear to be learnable by this method-for example, /f/ is an onset if the speech rate is fast, but /f/ is a coda if the speech rate is slow. The present study demonstrated that adult English speakers can learn (after a sleep period) constraints based on stress (e.g., /f/ is an onset if the syllable is stressed, but /f/ is a coda if the syllable is unstressed), but cannot learn analogous constraints based on tone (e.g., /f/ is an onset if the tone is rising, but /f/ is a coda if the tone is falling). The results are consistent with the fact that, in English, stress is a relevant lexical phonological property (e.g., "INsight" and "inCITE" are different words), but tone is not (e.g., "yes!" and "yes?" are the same word, despite their different pragmatic functions). The results provide useful constraints on how consolidation effects in learning may interact with early learning experiences.

The language of character strengths: Predicting morally valued traits on social media.

OBJECTIVE: Social media is increasingly being used to study psychological constructs. This study is the first to use Twitter language to investigate the 24 Values in Action Inventory of Character Strengths, which have been shown to predict important life domains such as well-being. METHOD: We use both a top-down closed-vocabulary (Linguistic Inquiry and Word Count) and a data-driven open-vocabulary (Differential Language Analysis) approach to analyze 3,937,768 tweets from 4,423 participants (64.3% female), who answered a 240-item survey on character strengths. RESULTS: We present the language profiles of (a) a global positivity factor accounting for 36% of the variances in the strengths, and (b) each of the 24 individual strengths, for which we find largely face-valid language associations. Machine learning models trained on language data to predict character strengths reach out-of-sample prediction accuracies comparable to previous work on personality (rmedian = 0.28, ranging from 0.13 to 0.51). CONCLUSIONS: The findings suggest that Twitter can be used to characterize and predict character strengths. This technique could be used to measure the character strengths of large populations unobtrusively and cost-effectively.