The Transcript Is an Argument, Not a Mirror

If you're like most defenders, you read the police interrogation transcript the way you'd read a deposition: as a faithful record of what was said. Peer-reviewed research says that assumption is wrong. The transcript is a selective, constructed document that reflects the officer's case theory as much as the suspect's words.

The strongest empirical evidence comes from Martha Komter at the Netherlands Institute for the Study of Crime and Law Enforcement, who analyzed a corpus of 34 audio-recorded police interrogations matched against the written reports officers produced. Her conclusion: police transcripts are "co-productions" of the interrogation, oriented toward building a case rather than neutrally capturing events. As she puts it, in court suspects are held accountable for "what the police wrote down as their statement," which disregards the fact that the transcript itself is a joint product shaped by the interrogator (Frontiers in Communication).

The most striking documented example: in one theft case, half an hour of interactive persuasion between officer and suspect disappeared between two adjacent written lines of the police report. Nothing in the document signals the omission. The reader sees a seamless statement; the recording tells a different story.

Komter's analysis of how that compression works should change how you read every transcript in your discovery pile:

• The report's "monologue style" transforms question-and-answer interaction into a seemingly volunteered narrative by the suspect. Coached, pressured, or repeated answers read as spontaneous admissions.
• Institutions write for future readership, not accuracy. Citing Garfinkel, Komter notes these documents exist to display "correct procedures," not to represent events objectively.
• Officers themselves view contemporaneous transcription as "a necessary evil" that detracts from what they consider the core business: interrogating the suspect.
• The suspect often doesn't know what the officer is typing, or how long the typing will last, so the written record incorporates words the suspect never reviewed.

The practical takeaway for your 2026 practice: treat the transcript as an argument to be cross-examined, and the recording as the primary source. That means comparing the written account against the audio line by line, flagging monologue-style passages that may compress question-answer sequences, and noting what's absent (breaks, repeated questions, the persuasion that produced the "voluntary" statement). The same skepticism you bring to a police interrogation transcript should extend to the other recorded evidence in the case, which is why grounded review of 911 call transcript analysis and jail call transcription belongs in the same workflow as interrogation review.

Komter is candid that her own transcripts are selective too, oriented to her research questions. Selectivity is unavoidable in transcription. The difference is whether you know what was left out. When you work from a transcript with word-level timestamps and citations back to the exact moment of the recording, the omissions become visible instead of buried.

What 2026 Case Law Changed: Figueroa and the Foreign-Language Trap

On August 7, 2026, the Massachusetts Appeals Court handed down a decision that changes how defense attorneys should think about every foreign-language police interrogation transcript in their files. In Commonwealth v. Figueroa, No. 25-P-1027, the court affirmed a conviction for assault and battery by means of a dangerous weapon and held that a trial judge has discretion, not a duty, to require an English transcript of a recorded interview conducted in another language.

Here's what that means in practice. Under Commonwealth v. Portillo, 462 Mass. 324 (2012), the Commonwealth must provide an English transcript before using a defendant's foreign-language statements in its case-in-chief. But Figueroa holds that Portillo does not extend to impeachment: once a defendant testifies, the interrogating detective may describe prior statements even when no English transcript of the recording exists. As the Serpa Law analysis puts it, the protections around a foreign-language interrogation are strongest before the defendant testifies, and taking the stand may make admissible the very statements the judge kept out of the case-in-chief.

That creates a doctrinal tension worth sitting with. Martha Komter's peer-reviewed research, built on 34 matched recordings and police reports, documents that police transcripts are "co-productions" oriented toward building a case. In one documented theft case, half an hour of interactive persuasion vanished between two adjacent written lines. If a defense attorney's own translation of the recording is the only check on the officer's account, Figueroa allows impeachment on an unverified version of what the defendant actually said.

The practical lesson is record-building, and Boston defense attorney Patrick J. Murphy argues the point directly: objections to untranslated interrogations have to be built at the motion hearing, not raised for the first time on appeal. Figueroa turned on what the defendant failed to do. Before trial, you want to:

• Demand the recording and a certified translation early, in writing, so the request is in the record.
• Compare the officer's account against the recording line by line, with timestamps, before the hearing.
• Put on evidence of any translation inaccuracy, and seek exclusion or a continuance where accuracy is in question.
• Argue inability to cross-examine the statements through the officer, the ground Figueroa found missing.

The court itself noted that prudence favors translating the interview and providing a copy to defense counsel, so a judge may order the Commonwealth to produce one. That discretion is not a rule of automatic exclusion, which means the strength of your motion record is the whole game. A grounded review workspace like Evidence Copilot helps here: word-level timestamps and citation-backed excerpts let you show the judge exactly where the officer's account diverges from the recording, the same approach used for 911 call transcript analysis and jail call transcription in the same case file.

The same verification discipline applies to any police interrogation transcript you contest, in any language.

Mapping Miranda Against Word-Level Timestamps

A police interrogation transcript looks like a clean record, but the suppression motion lives in the seconds between the warnings. Miranda bars questioning the moment a suspect indicates "in any manner and at any stage" a desire for counsel, so your workflow has to find that moment before the State does.

Start by marking the warnings. The detective reads the card; you timestamp the start, the end, and whether each warning was intelligible. Then flag waiver ambiguity: was the suspect asked yes-or-no, did the detective paraphrase the waiver, did the suspect ask a clarifying question that was answered with more questioning? Per Miranda v. Arizona, any waiver must be "voluntarily, knowingly and intelligently," and ambiguity cuts against the State. Finally, catch every invocation cue. "Should I get a lawyer?" "I think I need one." "My mom told me not to talk." "Can I call my brother?" Each is a flag, and the transcript needs to show you saw it.

Timestamped, speaker-labeled transcripts make this citeable. You can point to the exact second the detective stopped reading warnings, the second the suspect said "I guess," and the second the next question began. That precision is the difference between a colorable motion and a hearing where the judge asks which line you're relying on.

Two transcript types matter here. A working transcript (AI-generated, human-verified) is what you build the motion from: fast, searchable, marked AI-generated and pending human review. A certified transcript (human, court-usable) is what you attach when the transcript itself is the exhibit. The 2026 tooling landscape treats this as a validity requirement, with verbatim formatting and speaker diarization as admissibility baselines rather than nice-to-haves.

The practical move: run the interrogation through your transcript tool the day it lands in discovery. Mark warnings, waiver, and every invocation cue against word-level timestamps. Then run the same sweep on the 911 call and the jail calls, because the invocation question rarely stays inside one recording. The same workflow that catches a mid-interrogation "lawyer" also catches a pre-arrest "lawyer" on the 911 line, and a post-booking "lawyer" on a jail call, and suppression turns on which recording the State picks to enter first.

If you want a workspace that builds a chronological custodial-interrogation timeline and drafts a suppression motion outline with both sides quoted and cited, Evidence Copilot runs that playbook. Every AI output stays marked for human review; the legal conclusion is yours.

A Line-by-Line Review Checklist You Can Run This Week

Print the transcript, pull up the recording, and give yourself one hour. That's all a disciplined line-by-line review of a police interrogation transcript takes to run, and it catches the errors that matter: omissions, substitutions, and misattributed speech that a quick skim never surfaces.

Start with the omission pass. Peer-reviewed research on police transcripts found half an hour of interactive persuasion omitted from between two adjacent written lines in one documented case, drawn from a corpus of 34 matched interrogation recordings and reports (Komter, 2022). Play the recording at the timestamps flanking every gap and ask what happened in the space the paper version skipped.

Then run the substitution check. Flag any passage where the transcript's monologue style converts a question-answer sequence into a "seemingly volunteered narrative" by the suspect, and note precise details (a location, a time) that shifted into more prosecutor-friendly phrasing.

• Verify verbatim formatting survived: filler words, false starts, pauses, and notations like [inaudible] and [crosstalk]. A cleaned-up transcript misrepresents the original recording.
• QC speaker attribution on overlapping speech. Diarization errors on similar voices are a validity problem, not a cosmetic one.
• Map every Miranda warning, waiver, and invocation cue to a word-level timestamp, since an invocation "in any manner and at any stage" stops questioning (Miranda).
• Escalate hard audio (overlapping speakers, poor quality, heavy accents) to human transcription when the transcript needs to hold up in court.

Know which product you're holding. An AI-generated transcript is a working transcript that requires human verification; a certified human transcript is the court-usable version. Use the working copy to find the moments, then verify the ones you'll cite.

Finally, fold the interrogation file into the rest of the case. The same grounded-citation workflow applies to 911 call transcript analysis and jail call transcription, so inconsistencies between the interrogation, the emergency call, and the jail recordings surface in one place. In Evidence Copilot, every answer cites the exact second of video it came from, and every AI output is marked pending human review. The transcript is the argument; the recording is the proof. Check one against the other.