Reversible sentence pairs

On screen

1. Open On screen. Type a client code if you want the session filed under one: initials or a number, never a name.
2. Pick the sentences, the number of trials, and how the sentence is given: written with the voice on request, written and spoken, spoken only, or written only.
3. Press Start the session and turn the screen to the client. The client reads or hears the sentence and taps a picture. Say it plays the sentence (Say it again when the page has already spoken it), and every press is counted.
4. Your controls stay small at the top: Skip (no response) and End session. Each answer is saved the moment it is given, so a closed tab or a flat battery loses nothing: the home screen offers to resume.

At the table, with printed cards

1. Print the picture cards once from Print cards. Each card carries a small gray code, like 2B, so you can find it fast.
2. Open At the table, pick the plan, and say where you sit: across from the client or beside them.
3. Each trial shows the sentence to say and the two cards to lay out, drawn the way you see the table from your seat. Tap the card the client picks. No response, Asked to repeat and Undo last are one tap each.
4. No device at the table? Print the data sheet for the plan, circle the answers on paper, and enter them afterwards: At the table, then Use a code with the code printed on the sheet, gives the same trials in the same order.

Plan codes

Every session follows a plan with a code like M20-4821. The letter is the sentence set (M mixed, A active only, P passive only, B an active block, then a passive block), the number is the trial count, the four digits fix the order, and an N or X at the end means pairs 6 to 10 or all ten pairs (no letter: pairs 1 to 5, so codes from v1.0.0 still rebuild the same trials). The same code always gives the same trials in the same order, on any device. A finished session retires its code, so the next session gets a new order unless you type a code on purpose.

How a plan is balanced

• Mixed, 20 trials: every picture once with its active sentence and once with its passive sentence. 40: twice each. 10: every picture once; in each pair one picture gets its active sentence and the other its passive, so each type gets five.
• Active only or passive only: 10 trials is every picture once, 20 is twice, 40 is four times.
• Active, then passive: the same counts, split into an active block followed by a passive block.
• Pairs: all the counts above are for one set of five pairs. With all ten pairs, 40 mixed trials is every picture once with each sentence type, 20 is every picture once, and 10 is one picture from each pair (five actives, five passives); 20 active-only trials is every picture once.
• The target sits in the first position (left, or top when the pictures are stacked) in half the trials of each sentence type, give or take one when the count is odd. When an item comes up more than once, it appears in each position equally often.
• Order rules: never the same pair twice in a row; never the target in the same position more than three times in a row; in mixed plans, never more than four sentences of the same type in a row.

• Active and passive are counted apart. The gap between the two counts is the first thing to look at.
• The guessing line. With two pictures, guessing lands near half right. Under each count, the record says which totals would be unusual for guessing alone: less than a 1-in-20 chance on a one-sided binomial test. For 10 trials that is 9 or more, or 1 or fewer; for 20 trials, 15 or more, or 5 or fewer; for 40, 26 or more, or 14 or fewer. Below five trials, knowing and guessing can't be told apart.
• Well under half is information too. Picking the reversed picture again and again isn't what guessing looks like. One common reason is taking the first-named character as the doer, which works for every active sentence and fails for every passive one.
• Doer on the left, doer on the right. In every A picture the doer is on the left; in every B picture the doer is on the right. People who read left to right tend to expect the doer on the left (spatial agency research, listed in Sources). A big gap between those two rows can mean the layout, not the grammar, is steering the choice.
• Position and picks. The target is balanced between the two positions, so a client answering from the sentence picks each side about equally often. Many more picks on one side can point to a side preference or a visual field cut, which is worth ruling out before reading anything else.
• Pair 4. Doctors usually help patients, so 4A can come out right from world knowledge alone. 4B only comes out right from the grammar.
• Repeats. On screen, how often Say it or Say it again was pressed. At the table, how often the client asked to hear the sentence again.
• Time to answer (on screen only) runs from the moment the pictures appear to the tap, so with a spoken sentence it includes the time the voice takes. The record shows the median for each sentence type. Compare it within one client, on one device, with the same settings.
• What the record is not. It is a practice log. There are no norms, no cut-offs and no severity ratings, and nothing on this page is a test or a diagnosis. Reading it is your call.

Busy pictures pull the eye away from the sentence. The AI-made set this kit started from put houses, trees, a garden gnome, dust clouds and speech bubbles around five simple sentences. The look to aim for became the plain pictures of the old Dick & Jane readers: two characters, one action, nothing else.

Carefully controlled images evoked significantly less disproportionate visual attention than clipart.

stimulus-driven factors unrelated to linguistic comprehension may influence performance

Same sentence, before and after

The cat chased the dog.

Gone: the houses, trees, flowers, fence, bench, gnome, tipped pot, dust clouds and the Yikes! bubble. What's left is the two animals and the chase. In the new pair the cat keeps the left spot in both pictures, and only the direction of the chase changes.

Cut

• Scenery: houses, trees, a rose trellis, hydrangeas, a fence, benches, a garden gnome, a tipped flowerpot, falling leaves, dust clouds and an AI watermark.
• Extra people and things: parents, kids and a stroller behind the hug, and a backpack on the grass.
• Words: the Yay! and Yikes! bubbles, a thought bubble with a second little chase inside it, and the German sound word KRACH!. A sentence task shouldn't carry extra reading, and Yay! over the chaser hints at the answer.
• The cat-and-dog garden, reused as the setting for the car crash.

Kept

• One shared scene within each pair. That was the right instinct for a minimal pair.
• Friendly, recognizable characters, in color.

Fixed

• Old pair 1: the cat and dog switch sides between the two pictures and run the same way in both. Now they keep their spots and the chase changes direction.
• Old pair 2: in the second picture the girl only rests a hand on the boy's chest, which doesn't read as a hug. Now the hugger's arms go all the way around, leaning in with eyes closed, and the one being hugged stands with arms down, surprised.
• Old pair 3: the first crash is nose to nose with both vehicles dented, so either one could be the hitter, and the vehicles switch sides. Now the moving vehicle has speed lines, the other is parked, and a star marks where they touch.
• The old set stopped at pair 3. Pairs 4 and 5 now have pictures too.

The rules every pair follows

• Same two characters in the same spots in both pictures. Only the direction of the action flips, so each pair has one picture where the action runs left to right and one where it runs right to left.
• Everything else matches between A and B: size, colors, shadows, ground.
• The doer shows the action and the receiver shows the effect.
• No words and no background: a white page, a soft ground patch, flat color and one ink outline.
• The two characters differ in lightness as well as color, so the pairs still work on a black-and-white photocopy.

• The original sheet's sentences describe the A pictures. The B sentences are new, so both pictures in a pair can be the target, and each pair gives four trials: A active, A passive, B active, B passive.
• Pair 4 leans on world knowledge. Doctors usually help patients, so The doctor helped the patient can be right for the wrong reason. 4B turns it around, and only the grammar gets it right.
• Pair 5 says was bitten instead of was bit.
• With two pictures, a guess is right half the time, so one answer says little. Practice mode keeps separate counts for active and passive sentences because the gap between those two numbers is the thing to watch.
• Adult clients may find cartoon animals young. The drawings are made from code, so a line-art or more grown-up version is quick to make.

The kit is one HTML file that makes no network requests: no libraries, no fonts fetched from a server, no tracking. It opens offline, and it never runs everything at once.

• Idle until asked. At startup the page draws the home screen and stops. There is no animation and nothing repeating in the background, at any point.
• One copy of each drawing. The ten pictures are stored once, as SVG, and every card, thumbnail and practice screen draws them by reference (<use>). That keeps the file small and every screen quick to build.
• Screens build when opened. The gallery, the records list and the chart are made when you open them, not at startup.
• Print pages build one per frame. Opening Print cards lays the sheets out one page per screen refresh, so the page stays responsive while it works and the preview shows up with the first page. The Print button then only opens the print window.
• Speech only when a sentence is said. The voice speaks one sentence at a time and stops when you leave the screen.
• Records math on demand. A record's counts and a client's chart are worked out only when that record or client is opened.

• Records stay in this browser, on this device, in the page's own local storage. Nothing is uploaded and there is no account.
• Use a client code or initials, never a full name.
• Clearing the browser's site data erases the records. Download backup (under Records) saves them to a file; Restore backup puts them back, on this device or another one. Restoring adds sessions and never replaces ones already here.
• Each browser keeps its own records, and a private window keeps none after it closes. If the browser refuses to keep records at all, the page says so, and the CSV download still works.
• The CSV has one row per trial, ready for a spreadsheet.

Quester Gauntlet activated. Internal operational seed: turn the ten-picture page into a clinic-ready kit: plan a balanced session, run it on screen or with printed cards, keep the record on the device, and print cut-ready cards, sentence strips, an answer key and a data sheet, with nothing heavy running until the clinician asks for it.

Sun Tzu's lens for this round: the victorious strategist seeks battle only after the victory has been won. The battles worth winning in advance are the ones that happen at the table: a card that won't cut straight, a trial order that leaks the answer through position, a record that disappears, a page that freezes an old laptop. The gauntlet ran on five fronts, each in the order the brief set: pre-mortem, council, mirror, future scenarios, devil's advocate, wheat and chaff.

I. The plan

Order the ground before the troops arrive.

Plan. Replace coin-flip trials with a plan: a fixed list of trials, balanced by sentence type, picture and position, and rebuilt exactly from a short code.

Pre-mortem. The first practice mode picked the pair, the picture, the sentence type and the side at random on every tap. Twenty taps could easily mean thirteen passives and seven actives, or the target on the left six times running. A clinician comparing active with passive would be comparing unequal piles, and a side preference could pass for grammar.

Council.

• Aphasia clinician: the same number of actives and passives, every picture as the target, and a way to run the identical list again next week.
• Research methodologist: counterbalance the target's position within each sentence type, cap runs on one side at three, and never let the same pair come up twice in a row, or memory of the last trial does the work.
• Treatment researcher: mixed and blocked presentation are different tasks, so offer both.

Mirror. Randomize the trials turned inside out is make the order unpredictable and the counts impossible to get wrong. Seeded randomness does both: the order looks random, and the counts are fixed by construction.

Future scenarios. A data sheet printed on Monday and entered on Thursday rebuilds from its code, trial for trial. A covering clinician runs the same list from the same code. New pairs added later slot into the same plan builder, and the code format doesn't change.

Devil's advocate. Isn't a replayable list learnable? Over many sessions, possibly. That is why a finished session retires its code and the next one gets a new order; the code is there for when repeating is the point.

Wheat and chaff. Wheat: balance, position counterbalancing, run limits, a short code. Chaff: adaptive difficulty, streaks, levels, and anything that changes the list mid-session.

Orders. Plans M, A, P and B at 10, 20 or 40 trials; balanced positions; three order rules; the whole plan rebuilt from its code.

II. Two fronts: on screen and at the table

Fight where the client is, not where the software is.

Plan. Run the same plan two ways: the client taps pictures on a screen, or the clinician lays out printed cards and taps what the client picked.

Pre-mortem. The client double-taps and two answers are counted. A phone held upright stacks the pictures and left stops meaning left. The clinician sits across the table, reads client's left off the screen, lays the card on their own left, and every position is logged backwards. The browser has no voice. The tablet sleeps halfway through and the session is gone.

Council.

• Aphasia clinician: at the table I sit across from the client. Show me the cards the way I see them, and let me log each answer with one tap.
• Stroke rehabilitation clinician: a visual field cut often comes with the stroke that caused the aphasia. Offer one picture above the other, so both stay near the middle of the view.
• Accessibility designer: large targets, a large sentence, no motion, and the clinician's controls small and out of the client's way.

Mirror. Record what the client picked turned inside out is never make the clinician translate a side. At the table the logging buttons are the cards themselves, drawn as the clinician sees them, and the page keeps the left-and-right books.

Future scenarios. Telepractice: the on-screen mode runs in a shared browser window and the client points, or says top or bottom. An interruption: every answer is saved as it happens, and the home screen offers to resume. A client who can't tap reliably: the client points and the clinician taps; the on-screen mode doesn't care whose finger it is.

Devil's advocate. Two modes, twice the things to break? Only if they were two programs. Both read the same plan, write the same record and feed the same summary; only the input differs.

Wheat and chaff. Wheat: one tap per answer, resume after an interruption, skip for no response, counted repeats, a seat setting. Chaff: sound effects, animation, points, a countdown.

Orders. On screen: presentation, voice speed and voice, marks, layout, and a trial counter the client only sees if you want them to. At the table: seat, undo, no response and repeats. Both save after every answer.

III. The record

Count what was won, and say what the count can't tell you.

Plan. Every finished session becomes a record on the device: a summary, every trial and a note, with a CSV export, a backup file, and a chart per client across sessions.

Pre-mortem. Browser data gets cleared and months of sessions vanish. A client's full name goes onto a shared tablet. Seven of ten gets read as above chance. A private window silently keeps nothing, and nobody notices until the session is over.

Council.

• Aphasia clinician: active and passive side by side, by pair, with the trial list, printable for the chart.
• Statistician: with two pictures a guess is right half the time. Say which totals would be unusual for guessing alone, and let small counts stay unreadable.
• Privacy officer: codes not names, nothing leaves the device, and the backup is the clinician's own file.

Mirror. Show how well the client did turned inside out is show what else could have produced this count: guessing, a side preference, a doer-on-the-left habit, world knowledge on pair 4. The summary splits the count along each of those lines.

Future scenarios. Six weeks of sessions: the chart shows active and passive by date against the guessing line, with the session list as its table. A new laptop: download the backup and restore it there. A clinic spreadsheet: one CSV row per trial.

Devil's advocate. Won't a percentage read like a test score? It can. So counts come first and percentages second, the guessing line sits under every count, there are no cut-offs and no pass or fail, and the notes say plainly that this is not a test.

Wheat and chaff. Wheat: counts by sentence type, pair, doer side and position; repeats; median time; CSV; backup and restore. Chaff: accounts, cloud sync, norms, severity scales.

Orders. Records on the device, the summary, CSV, backup and restore, a per-client chart with its table twin, and a warning whenever the browser won't keep a record.

IV. The print shop

A card that cuts crooked loses the session before it starts.

Plan. One Print cards button that lays out cut-ready sheets on plain paper: picture cards at three sizes, sentence strips, an answer key, and a data sheet for the current plan.

Pre-mortem. The browser's margins push the bottom row onto a blank extra page. Fit to page shrinks the cards off their cut lines. Codes on the sentence strips let the client match codes instead of reading. An A4 printer crops a Letter layout. Thirty snips a sheet.

Council.

• Print designer: cut lines that run edge to edge, so a paper trimmer takes each one in a single pass; fixed page sizes with a safety margin; no sentences on picture cards unless asked for.
• Aphasia clinician: one pair per page is the size I use at the table; the whole set on one sheet is for a take-home pack.
• Photocopier: the two characters in every pair differ in lightness as well as color, so a black-and-white copy still tells them apart.

Mirror. Print the pictures turned inside out is print the session. The data sheet carries the same plan as the screen, so paper and screen agree trial for trial.

Future scenarios. Home practice: a family gets the one-sheet set and the strips. Laminated cards used for months: the corner codes survive every sort. A clinic on A4: one switch resizes every sheet.

Devil's advocate. Why not ship a PDF? A PDF can't follow a new plan, can't switch between Letter and A4, and is one more file to lose. The page builds its print sheets from the same drawings it shows on screen.

Wheat and chaff. Wheat: three card sizes, strips, answer key, data sheet, paper switch, page-by-page preview. Chaff: bleed and crop marks for a print shop, decorative borders, card backs, two-sided printing.

Orders. A print screen with options and a preview; fixed-size sheets; dashed, edge-to-edge cut lines; the answer key and the data sheet.

V. The supply line

An army that marches everything at once arrives exhausted.

Plan. Build only what is on screen, and never run everything at once, so the kit stays quick on an old laptop, tablet or phone.

Pre-mortem. Building every print page, twenty thumbnails, the chart and the gallery at startup freezes an old device for seconds. An animation loop keeps the fan running in a forgotten tab. The file grows past what an email will carry.

Council.

• Front-end engineer: store each drawing once and reuse it by reference; build screens lazily; lay out print pages one per frame.
• Clinic IT: no network calls, no installs, no plug-ins: a file that opens offline on a locked-down machine.
• Aphasia clinician with an old tablet: it has to open fast and stay fast.

Mirror. Make it fast turned inside out is make it idle. The page does nothing at all until someone asks it to.

Future scenarios. Fifty pairs instead of five: still one copy of each drawing, still one print page per frame. Years of records: browsers cap local storage at a few megabytes, so the page shows how much it holds and warns the moment a save fails.

Devil's advocate. Is one page per frame overkill for ten pictures? Today either way takes a fraction of a second. It is cheap insurance for old devices, and for the day the set grows.

Wheat and chaff. Wheat: lazy screens, shared drawings, paced print layout, nothing running in the background. Chaff: frameworks, service workers, fonts fetched from the web, analytics.

Orders. As built; see How the page stays light.

Synthesis

The victory before the battle: a plan that can't leak the answer, two ways to run it that write the same record, paper that cuts straight and matches the screen, and a page that stays idle until it's asked. What's left for the session is what should be left: the client, two pictures, and the grammar.

Round one made the pictures. Its print plan, a separate PDF and ZIP, is replaced by the Print cards screen, which lays out the same one-pair-per-page cards and the answer key from the drawings inside this file.

Internal operational seed: replace the busy AI pictures with ten primer-plain minimal-pair pictures (two actors, fixed spots, one action, no background) for all five pairs, then deliver them in forms a clinician can use at the table.

Sun Tzu's lens for the whole campaign: a sentence-picture task is a battlefield where only one thing should move, and that thing is who does what to whom.

I. Know the enemy

Count the clutter before you fight it.

Plan. Inventory everything in the six old pictures and sort each piece by one test: does it help show who did what?

Council.

• Aphasia clinician: the wrong picture in a pair should differ from the right one in one way only, who did what. Anything else is a second test hiding inside the first.
• Eye-tracking researcher: people with aphasia spend more time on whichever picture stands out physically, and more so when comprehension is more impaired (Heuer, Ivanova & Hallowell, 2017). A garden gnome is a magnet.
• Psycholinguist: speech bubbles put extra words into a word task, and Yay! over the chaser leaks the answer.

Mirror. Flip what makes a picture helpful? into what could make someone pick wrong for a reason that isn't grammar? Every answer to the second question went on the cut list.

Wheat and chaff. Wheat: one shared scene per pair, and characters with personality. Chaff: about a dozen kinds of props, four bits of text, and the garden that got reused for a car crash.

Devil's advocate. Aren't rich scenes more engaging and more like real life? In a storybook, yes. Here the scene's only job is to hold still while the grammar moves. The faces can carry the charm.

Pre-mortem. Cleaning up the backgrounds alone would still have failed: the actors switch sides between pictures, the second hug is a hand on a chest, and the crash is nose to nose with both vehicles dented.

Orders. Strip each picture to two actors and one action, fix the unclear actions, and draw pairs 4 and 5.

II. Choose the ground

Pick ground where only the meaning can move.

Plan. Write rules that make picture A and picture B identical except for who does what.

Council.

• Psycholinguist: people who read left to right tend to picture the doer on the left, and passive sentences shift that expectation (spatial agency research, 2020). If every doer stands on the left, position answers the question instead of grammar.
• Treatment researcher: people with aphasia can fall back on position-based strategies, and after sentence treatment they used them less (Mack & Thompson, 2017). The pictures shouldn't reward that shortcut.
• Aphasia clinician: every picture should get to be the target, not just the foil, so both directions of the action get tested.

Mirror. Keep the actors still turned inside out is let the action move. The cat is always on the left and the dog always on the right; in 1A the chase runs left to right, and in 1B it runs right to left. Every pair has one of each.

Wheat and chaff. Wheat: the five original sentence pairs. Chaff: the idea that each pair has one answer picture. New sentences for the B pictures mean each pair gives four trials.

Devil's advocate. Won't a client notice that the direction flips? Probably. Noticing the flip still doesn't say who's doing it. They have to map the sentence onto the roles either way.

Pre-mortem. Pair 4 leaks. Doctors help patients in real life, so The doctor helped the patient can be answered from world knowledge. The fix is 4B, the patient helping the doctor up, where only the grammar gets you there.

Orders. The five rules listed under What changed.

III. Arm the troops

Give the doer the action and the receiver the reaction.

Plan. Draw like an old first-grade reader: white page, big clear shapes, one action, friendly faces. New characters in that spirit, not copies of the book's.

Council.

• Primer illustrator: every action needs two signals, one on the doer and one on the receiver, so the direction reads at a glance.
• Print and accessibility designer: colors should differ in lightness too, so a photocopy still works, and nothing should be so thin it disappears on paper.
• Aphasia clinician: faces should be readable but calm, because a frightening face becomes the thing everyone looks at.

Mirror. Simple turned inside out is easy to misread, so every action got a pair of cues:

• Chase: speed lines behind the chaser; the chased one looks back, sweating.
• Hug: the hugger's arms go all the way around, eyes closed; the other stands with arms down, surprised.
• Hit: the moving vehicle has speed lines; the other is parked, and a star marks the hit.
• Help: the helper stands and pulls; the helped one sits on the floor.
• Bite: the biter's jaws hold the tail tip; the bitten one jumps, wide-eyed.

Wheat and chaff. Wheat: flat color, a soft ground patch so nobody floats, one shadow each. Chaff: backgrounds, words, texture and props.

Devil's advocate. Cartoon animals can feel childish to an adult client. That's a real risk. The style is plain rather than babyish, and because the pictures are drawn in code, a line-art or more grown-up version is quick to make. That call belongs to the clinician.

Pre-mortem. Two catches in review. The biting animal's squeezed-shut eyes read as pain, which muddied who was hurt, so it now has open, focused eyes and only the bitten one reacts. The car's speed lines ran off the edge of the frame, so that scene moved inward.

Orders. Ten pictures, reviewed in color and in grayscale.

IV. The campaign

A plan that has already failed on paper won't fail at the table.

Plan. Deliver what gets used: printable cards, image files, and a way to practice.

Council.

• Print designer: US Letter, one pair per page, vector art that stays sharp at any size, small gray labels, and no sentences on the picture pages.
• Aphasia clinician: the answer key on its own page, with an active and a passive sentence for every picture.
• Eye-tracking researcher: in practice mode both pictures are the same size, and they trade places at random.

Mirror. Deliver pictures turned inside out is deliver a way to use them. Practice mode reads the sentence aloud, shuffles the pictures and keeps separate counts for active and passive.

Wheat and chaff. Wheat: the PDF, the PNG and SVG files, and practice mode. Chaff: sound effects, animation and points that would turn practice into a game.

Devil's advocate. Does anyone need a practice page when the cards exist? Not necessarily. The cards work on their own. The page is for a phone or tablet.

Pre-mortem. AI image generators drift between two versions of the same scene. Code-drawn pairs can't, so A and B match to the pixel. Voices differ from device to device, so the words can stay on screen.

Orders. Ship this page, the PDF and the ZIP.

Synthesis

Sun Tzu's best victory is the one won before the fight starts. Here that means arranging the pictures so the only road to the right answer runs through the grammar: fixed spots, a flipped action, a doer who acts and a receiver who reacts, and nothing else on the page.

• Heuer, Ivanova & Hallowell (2017). More than the verbal stimulus matters: visual attention in language assessment for people with aphasia using multiple-choice image displays. Journal of Speech, Language, and Hearing Research 60(5).Eye-tracked 41 people with aphasia on 40 picture sets. The picture that differed physically drew more looking time, and more so with more severe comprehension problems.
• Hallowell & Heuer (2005). Eye tracking evaluation of clipart versus carefully designed images in multiple-choice image displays for comprehension assessment. Clinical Aphasiology Conference.Carefully built picture sets spread attention more evenly than clipart sets showing the same content.
• Heuer & Hallowell (2006). Influences of image characteristics on visual attention in a multiple-choice image array with and without presentation of a verbal stimulus. Clinical Aphasiology Conference.Makes the case that the physical features of pictures can get in the way of a fair comprehension test in aphasia.
• Meltzer, McArdle, Schafer & Braun (2010). Neural aspects of sentence comprehension: syntactic complexity, reversibility, and reanalysis. Cerebral Cortex 20(8).Describes the standard clinical task: hear a sentence, then pick the matching picture from a set that includes a role-reversed foil.
• Mack & Thompson (2017). Recovery of online sentence processing in aphasia: eye movement changes resulting from treatment of underlying forms. Journal of Speech, Language, and Hearing Research 60(5).Before treatment, passives were understood less reliably than actives. After treatment, participants leaned less on a position-based response strategy.
• Martin, Kohen, Kalinyak-Fliszar & Guerrero (2013). Comprehension of sentences with reversible semantic roles is sensitive to phonological STM capacity. Clinical Aphasiology Conference.Sentences with reversible roles are hard for many people with agrammatic aphasia, especially passives.
• What is agentic about the Spatial Agency Bias? How pragmatic relevance contributes to the spatial representations of actions (2020). University of Padua research archive.People tend to choose pictures with the doer on the left. Across five studies, passive sentences pushed the doer toward the right.
• Zenz (2015). The effects of color in recognition of images in multiple-choice displays by people with and without aphasia. Master's thesis, University of Wisconsin–Milwaukee.Looks at how color affects picture recognition in these displays. The set keeps flat color and was also checked in grayscale.