When we learn a second language (L2), our brain does not start with a blank slate. Instead, it views the new language through a "filter" constructed by our native tongue (L1). This filter—composed of ingrained phonological rules, rhythmic patterns, and physical motor habits—systematically distorts how we perceive and produce foreign sounds.
Research across decades, from historical lexicography to real-time MRI studies, suggests that "foreign accents" are not random errors. Rather, they are the predictable result of the brain attempting to map novel acoustic data onto a pre-existing, optimized mental map.
The Perceptual Trap: Hearing What You Know
The most fundamental barrier to mastering a new language is not the tongue, but the ear. This phenomenon, often called perceptual assimilation, occurs when the brain ignores acoustic differences that are irrelevant in one's native language.
As Hui Feng and colleagues demonstrate, Chinese learners of English often struggle with vowel contrasts like "bet" vs. "bat" because their native phonetic system treats these distinct sounds as mere variations of a single category. This creates a "perceptual trap" where the listener is functionally deaf to the very cues they need to distinguish.
This distortion is so powerful that it can lead to "hallucinations." Adriana Guevara-Rukoz and Sharon Peperkamp found that Japanese listeners, whose native phonotactics forbid most consonant clusters, actually "hear" illusory vowels between consonants in foreign words. This isn't a conscious choice; it is an automatic perceptual repair to make the input fit the L1 "legal" structure.
Prosodic Transfer and Stress Deafness
Beyond individual sounds (segments), the "music" of a language—its rhythm, stress, and intonation (suprasegmentals)—is also subject to heavy interference.
Learners often experience stress deafness, a term highlighted by Jingli Lu and others to describe the difficulty speakers of syllable-timed languages (like Cantonese) have in perceiving emphasis in stress-timed languages (like English). While a native English speaker uses a "cocktail" of pitch, duration, and intensity to signal stress, a tonal-language speaker might rely almost exclusively on pitch, remaining insensitive to the timing cues that native speakers take for granted.
Helen Meng and Emily Nava argue that this prosodic transfer—applying the staccato rhythm of an L1 to the flowing, reduced-vowel rhythm of an L2—is often what most hinders a speaker's perceived fluency and intelligibility.
The Physicality of Speech: Articulatory Settings
Even when a learner understands the rules, their body may resist. Speech is a highly coordinated motor task. Shrikanth Narayanan and Louis Goldstein have used real-time MRI (rtMRI) to show that proficient L2 speakers often substitute complex target gestures with "hidden" motor patterns from their native language. For example, a learner might produce an English "r" sound that sounds correct but uses a throat constriction height borrowed from their L1.
Furthermore, Andrés Benítez and colleagues found that speakers actually adopt different articulatory settings—the default resting posture of the tongue and jaw—depending on which language they are about to speak. A persistent foreign accent may stem from a failure to adopt the "physiological default" of the target language, forcing the learner to fight against their own mouth's resting state.
Orthographic Interference: The "False Friend" of Script
The way we read also distorts how we speak. Lei Li and Hong You point to orthographic interference, where learners apply the spelling-to-sound rules of their native language to the written form of the L2.
This is particularly evident in loanwords. Yunmi Park shows that Japanese learners of Korean often misspell or mispronounce words like "Internet" because they are viewing the word through the phonetic transcription habits of Japanese. The familiar script acts as a "false friend," triggering native phonological rules that override the actual sounds of the target language.
Modeling the Noisy Channel
Understanding these systematic distortions has profound implications for technology. Preethi Jyothi and Mark A. Hasegawa-Johnson have modeled the non-native listener as a noisy channel. By understanding the specific "noise" (the L1 filter) introduced by a transcriber, researchers can actually recover the original signal with high accuracy.
Similarly, in Computer-Aided Pronunciation Training (CAPT), researchers like Nancy F. Chen and Lan Wang are moving away from generic "correctness" scores. Instead, they use large-scale corpora, such as the iCALL corpus, to build "error dictionaries" that predict mistakes based on a learner's specific L1. By recognizing that a Spanish-speaking child's errors are different from a Mandarin-speaking adult's, these systems can provide the targeted, physical feedback necessary to help learners finally bypass their native language filter.