multisensory perception

Perceptual expertise is widely assumed to refine multisensory perception, yet the empirical literature appears inconsistent. Trained musicians show narrower audiovisual temporal integration windows and quicker segregation, whereas experienced wine tasters seem more influenced by a wine's color in their aroma and flavor judgments than novices are. We argue that the inconsistency arises from a conflation of two formally distinct quantities: the prior probability that concurrent signals arise from a common source, and the coupling between the values of signals that share one. We present a generative model in which both are free parameters, along with channel precision and a decision criterion, that contains causal-inference and coupling-prior models as special cases. The model shows that the correlation between the values of two signals across an environment reflects only the product of the two priors, so that a correlation-tracking learner acquires the coupling when the signals always share a source, whereas separating the two priors requires representing causal structure; that binary common-source reports confound the common-cause prior with the decision criterion; that the coupling trades off against the measurement noise of the influencing channel unless its precision is measured independently; and that the two priors leave distinct signatures in how cross-modal influence varies with disparity. When the source is fixed, as for the color and odor of a wine, tighter coupling alone produces the conjunction of greater displacement by color and better discrimination between wines reported for experienced tasters, which no change in odor precision can produce. Simulations of a test battery combining continuous estimates, common-source reports and discrimination thresholds show that the parameters can be recovered, that a within-subject dissociation can be detected with conventional sample sizes under illustrative effect sizes, and that recovery fails in certain cases. The account yields a rule classifying environments by their base rate of common causation and their coupling strength, and the same ambiguity attends claims about altered multisensory processing in autism and aging. We also show that the most cited demonstration of expert susceptibility to color contains no expertise manipulation, and suggest that apparent expert biases may reflectwind calibration to real-world statistics.

Multisensory perception is a cornerstone paradigm for understanding how the brain constructs coherent representations of the world from noisy, fragmented sensory inputs. For decades, researchers have used the magnitude of crossmodal illusions, the width of the temporal binding window, and related behavioral indices as direct proxies for 'integration strength,' and have leveraged these measures to compare multisensory function across developmental, clinical, and aging populations. Here we argue that this descriptive practice is frequently compromised: behavioral readouts of multisensory integration are composite measures jointly shaped by unisensory precision, amodal priors, and the binding process itself, and cannot be interpreted in isolation. Drawing on simulations within a Bayesian Causal Inference framework, we show how identical behavioral patterns can arise from very different underlying causes, leading to systematic misattribution of group differences to 'deficits' or 'enhancements' in integration. We review complementary computational frameworks, including coupling priors, drift diffusion, multisensory correlation detection, and statistical facilitation models, and outline their respective explanatory limits. Finally, we provide a model-based inference pipeline, from experimental design and unisensory baselines to parameter estimation and interpretation, that aims to help disentangles sensory fidelity, prior expectations, and integrative tendency. Adopting this normative approach can facilitate cumulative progress in basic multisensory research and for its translation to neuropsychiatric assessment, lifespan research, and artificial perceptual systems.

ادراک چندحسی یکی از پارادایم‌های اساسی برای درک این است که مغز چگونه بازنمایی‌های منسجم از جهان را از ورودی‌های حسی پر سر و صدا و تکه‌تکه می‌سازد. برای دهه‌ها، پژوهشگران از بزرگی خطاهای توهمات بین‌حسی، پهنای پنجره اتصال زمانی و شاخص‌های رفتاری مرتبط به عنوان نمایه‌های مستقیم برای «قدرت ادغام» استفاده کرده‌اند و این سنجه‌ها را برای مقایسه عملکرد چندحسی در جمعیت‌های رشدی، بالینی و سالمندی به کار برده‌اند. در این مقاله استدلال می‌شود که این رویه توصیفی اغلب با مشکل مواجه است: خروجی‌های رفتاری ادغام چندحسی، سنجه‌هایی مرکب هستند که به طور مشترک تحت تأثیر دقت تک‌حسی، پیش‌انگاره‌های آمودال و خود فرایند اتصال قرار دارند و نمی‌توان آن‌ها را به طور مجزا تفسیر کرد. با استفاده از شبیه‌سازی‌ها در چارچوب استنتاج علّی بیزی (Bayesian Causal Inference)، نشان می‌دهیم که الگوهای رفتاری یکسان می‌توانند از علل زیربنایی بسیار متفاوتی ناشی شوند و این امر منجر به نسبت‌دهی نادرست و نظام‌مند تفاوت‌های گروهی به «نقص» یا «افزایش» در ادغام می‌شود. چارچوب‌های محاسباتی مکمل، از جمله پیش‌انگاره‌های اتصال، مدل‌های انتشار رانش (drift diffusion)، شناسایی همبستگی چندحسی و مدل‌های تسهیل آماری را مرور کرده و حدود توضیحی هر یک را تبیین می‌کنیم. در نهایت، یک خط لوله استنتاج مبتنی بر مدل، از طراحی آزمایش و خطوط پایه تک‌حسی تا برآورد پارامتر و تفسیر نتایج ارائه می‌دهیم که هدف آن کمک به تفکیک وفاداری حسی، انتظارات قبلی و گرایش به ادغام است. اتخاذ این رویکرد هنجاری می‌تواند پیشرفت انباشتی در پژوهش‌های بنیادی چندحسی و ترجمه آن به ارزیابی‌های عصب‌روان‌پزشکی، پژوهش‌های طول عمر و سامانه‌های ادراکی مصنوعی را تسهیل کند.