If you don't like the orientation of the coordinates, this is where matrix rotation comes into play.  If you think of a 2D solution as dots on a flat piece of paper, if you spin the paper around, you can see them from different angles, but the distances between the dots don't change.  We can do the same in three dimensions, like turning a cube all around in your hand.  The only important thing is that we don't change the distances between points, since figuring out distances is the purpose of an MDS analysis.  

In order to do rotation, we can spin the coordinates around the x-axis, y-axis, or z-axis.  Here is a great visual example of rotating around each of these axes from ​jsfiddle.net/greggman/Lqg93y1v/. 

We tell the script to do this by inputting how much of a rotation we want in radians.  Below is a handy guide for converting angles to radians from this site.

You can put π into R as "pi", so pi/2 is a 90º rotation, pi is a 180º rotation, and (3 * pi)/2 is a 270º degree rotation.  Let's say we want to rotate our solution 90º around the z-axis.  Below is what this would look like.  As you can see in the first plot, the points have moved counterclockwise a fourth of the way around.  If you look at the updated dimension scores in the new dataframe MDS_points_rotated, you'll notice that because we're rotating around the z-axis, none of the D3 scores have changed.  In the second plot below, this means that all of the points are at the same height as in the original plot.

If you want to rotate around more than one axis, change the input of the relevant rotation chunk to the output of the previous rotation (MDS_points to MDS_rotated_points).  Here is what it would look like if we also rotated the already rotated solution 90º around the y-axis.

Use the rotations to help you understand how listeners were grouping the stimuli.  In this case, the original dimension scores are clearer, since for Dimension 1 at least, we can also see a divide with points with short V1 on the left and those with long V1 on the right.  Perhaps we would want to rotate -45º along the z-axis to make this distinction sharper.  You can also think about what makes most sense visually for your audience, such as putting /i/ in the top left corner in a study looking at vowels. 

​Once you're happy with your rotation, be sure to save the image(s) with the ggsave chunk at the bottom of the script.  There is also a chunk for saving the new dimension scores, which you can use later if you want to correlate acoustic/phonological properties with the dimension scores to find out what cues listeners were using to group the stimuli.

This script will save text files of the dimension scores for 1, 2, 3, 4, and 5 dimensional solutions.  The dimension scores are the points in space of each stimulus. In the image below, we see the 3 dimensional solution of our Finnish length data, combined across contexts.  You can think of D1, D2, and D3 as (x, y, z) coordinates.  For example, in row 7, we see that the average CVCCVV token for speaker N (averaged across pata, tiki, and kupu contexts) is placed by the MDS solution at x = 0.00527, y = -0.59382, z = -0.41867.  

When Dimension 1 x Dimension 2 (i.e. x and y coordinates) and Dimension 1 x Dimension 3 (i.e. x and z coordinates) are plotted , we see where CVCCVV_N is placed relative to other stimuli.  You can think of the second graph (Dim1 x Dim3) as viewing the first graph from above. 

In these plots, you can somewhat see that the stimuli are grouping together mainly by vowel length rather than consonant length (e.g. Dim 1 has mostly short V1 on the left and long V1 on the right), but it's still difficult to interpret.  For this reason, we recommend rotating the solution (i.e. moving all of the points in a certain way by a specific amount) to provide a better visualization.  This doesn't change the position of the points relative to each other, which is the important part in an MDS solution.  My colleague Ryan Lidster did this in Excel, but since that was rather clunky, I'm working with him on creating an R script to aid in rotating MDS solutions and plotting them.  Stay tuned!  [Edit 4/13/26: We finally have a script for visualization that includes rotation!  See the next blog entry.]

Added 4/13/26:
Finally, this new version of the script outputs matrices of Euclidean distances between stimuli.  These are useful if you want to compare MDS distances for contrasts to their accuracy in a discrimination task, as we did in our 2023 paper.  I'll have more information about this in a later blog entry.

As you can see, the header cVccV_upu at the top left should be the header for the first column of numbers.  If you want to make a table with these results, I recommend opening this file in Excel and moving all of the headers one column to the right.  (Note that you do NOT need to do this to use the R script for multi-dimensional scaling described in the following blog post.)  The file will now look like this:

This file shows us that, for example, [kuppu] tokens and [kuuppu] tokens were grouped together 19.7% of the time (column B, row 6).  Since this is the similarities for all speakers combined, the numbers along the diagonal show how often the sound files of the same stimulus spoken by different speakers were grouped together (not very often; English speakers are bad at length).

​The R code can handle up to 4 different contexts.  If you have more than 4 contexts, have more than 2 versions (i.e. Version A and Version B for counterbalanced order of presentation), or find any errors with the code, let me know at daidoned AT uncw.edu and I can modify/fix the script.

Make sure you label your columns this way so that the R script will work correctly.
Subject: The participant's ID 
Version: The version of the task that the participant did

• For example, in our Finnish length task we had two versions for the different orders of slides that we counterbalanced.  In version A, the "ata" context (pata, paata, patta, etc.) was on slide 1, the "iki" context (tiki, tiiki, tikki, etc.) was on slide 2, and the "upu" context (kupu, kuupu, kuppu, etc.) was on slide 3.  In version B, the "upu" context was first, the "iki" context was second, and the "ata" context was third.  Be sure to use the letters "A" and "B" for your coding instead of numbers.  If you only have one version, label them all version A.  You'll still need this column for the R code.  If you have more than 2 versions, the R code won't work.  Send me an email and I can modify the script for you.

Slide: The slide that you are coding
Context: The context for that slide
Group: Which group on that slide you are coding
Tokens: Which tokens were grouped together in that group

Let's use the following participant's results as a model.  Here is their first slide:  ​

This participant's ID is B50.  They did version A of our task.  This is slide 1, and in Version A, the context on the first slide is "ata".  They made 5 groups of stimuli on this slide, so let's choose a group at random and code it, for instance the group at the left containing 19, 14, 23, 20, and 24.  Since this is the first group we're coding on this slide, we can label it group 1 under Group and put in the token numbers under Tokens.  For the token numbers, we want to separate them with commas and no spaces.  It doesn't matter what order the token numbers are in within that cell.  So now we have:

If we code the rest of the groups on this slide, we have:

Here is B50's second slide:

And below is the coding for this slide added to the spreadsheet.  Since B50 made 4 groups on this slide, we only have 1-4 under Group:

Here is B50's slide 3:

This participant's results are finished being coded!  Do this for all your participants and you'll be ready to analyze your data.  You'll need to save this spreadsheet as a tab-separated text file for use with the R script to create similarity matrices.

To choose the stimuli for a free classification task, I recommend including all related sounds if possible, such as all the vowels in that language.  If not, when it comes time to do a multi-dimensional scaling analysis and correlations with properties of the stimuli, the results may be difficult to interpret because listeners don't need to use all of the relevant dimensions to group the stimuli.  For example, if you only include front vowels in your task, it may appear in your analysis that F2, or vowel backness, is not a relevant dimension for grouping vowels, when actually it is an artifact of the stimuli you picked. 

Once you decide what sounds you want to examine, you will need to determine what phonetic contexts you want to use.  I recommend two or three different contexts, since perception may differ between them, and you'll have more data points.  In our German vowel experiment, we used both an alveolar context (/ʃtVt/) and velar context (/skVk/), and in our Finnish length experiment we used three contexts (pata, tiki, kupu).  The stimuli for each context will be on a separate slide.

Next, you'll need to figure out how many speakers you should record, which will determine how many stimuli you have per slide.  I think that 30 stimuli per slide is the upper limit, since it becomes increasingly difficult to compare all the sound files to each other the more you have.  For example, in our German experiment, since we looked at 14 different vowels, we only had 2 speakers, such that each slide contained 28 stimuli.  For our Finnish length experiment, we looked at 8 possible length templates (CVCV, CVVCV, CVVCCV, etc.), and thus we chose to have 3 different speakers, for a total of 24 stimuli on each slide.  Keep in mind that if you have few stimuli to group per slide, you'll have fewer data points for your analysis. 

When you have your list of stimuli for each slide, you'll have to decide which number each stimulus will have.  These should be randomized so that participants can't use the numbers to group the stimuli.  If you want to use our R script for analyzing results, you should create a file like the one pictured below to record which numbers and slides correspond to which stimuli.  This should then be saved as a tab-separated text file.  We have called this file a "Lookup Matrix" and our example one for the Finnish length experiment is available here.  Make sure your codes will be in the same alphabetical order across contexts.  For example, here we have our code as CVCCV_ata_M instead of patta_M, since patta_M would alphabetically be after paata_M, but kuppu_M would alphabetically be before kuupu_M.  When the R script calculates similarity, the codes are put in alphabetical order, and the contexts combined matrices are calculated by adding the matrices for each context. Thus, all of the "ata" stimuli need to be in the same order as all of the "iti" and "upu" stimuli.

We have two versions of the task, Version A with the "ata" context first and Version B with the "upu" context first.  Only the order of slides differs, rather than what is on the slides themselves.  ASlide and BSlide refer to which slide contains which context within each version.  In this case, the "ata" context is on slide 1 for Version A but on slide 3 for Version B.  IconNumber refers to the random number that each stimulus is given.  For Finnish length, Condition refers to the length template of each stimulus (e.g. stimulus 1 is [patta]).  For your own task, this may be the target vowel or consonant for that stimulus.  Code is the columns Condition, Context, and Speaker concatenated together.

​Once you have the sound files and their corresponding random numbers ready, you’ll need to insert all the sound files into PowerPoint and change their images to the appropriate number icon.  Insert the sound file and change the image for each one by one to lessen the possibility of mixing up the numbers.  A sample PowerPoint with grid can be downloaded here.  Number images 1-28 are available here.  The following instructions work with Office 365.

To insert a sound file into PowerPoint:
Go to “Insert” --> “Audio” --> “Audio on my PC”

After you’ve chosen a sound file, it will show up with a speaker icon.  You will need to change this icon to the picture of a number by clicking on the icon, clicking on "Audio Format" in the top ribbon, and then clicking on “Change Picture” and choosing the appropriate image file.

Now that you have a number icon for the sound file, you can add a border to it by clicking on "Picture Border".  Make sure the color is black and the width is 0.75.

To adjust the size of the images, I recommend specifying the height and width to ensure uniformity across the images rather than manually adjusting the size of each by eyeballing it.  Images that are 0.46” high x 0.61” wide fit nicely into the rectangles of the grid.  You can specify the size of the image on the right size of the ribbon.  Follow these steps for the sound files and numbers for all your slides and you're ready to go!

For a participant to complete a free classification task takes about 10-20 minutes, depending on how many stimuli they have to group and how obsessively they listen to everything.  You can remind participants that there is no right answer and that they can make groups of any size as long as there are at least two sound files in a group.  They can also arrange groups however they want as long as the divisions between groups are clear.  Where they put the groups on the grid, or even if they use the grid, does not matter for the analysis.  Once a participant finishes, we always check to make sure their groups are clearly delineated and that they don't have any single sound files without a group.  The PowerPoint file is then saved with the participant name for later coding.  Here are some examples of how participants have completed a free classification slide.

Additional tips:

• You can counterbalance which slide participants see first, such that half the participants get one context first (e.g. velar) (Version A) and the other other get the other context first (e.g. alveolar) (Version B).  We've never found a difference by slide order, so I wouldn't say this is strictly necessary, but it's nice to show that presentation order doesn't matter.  Our R script for analysis assumes there are two versions. 
• The correspondences between numbers and sound files should differ between the slides so that participants don't catch on and simply choose to group the same numbers as before. 
• You can change the background color so that they differ for each slide (e.g. first slide is white, second is light green).  This makes it visually easier to see what slide participants are on if you're testing them in person.   
• You may find it useful to also give participants a practice slide with 6-8 sound files with very clear differences to make sure they are following instructions.

What you'll notice is that it is zoomed out all the way in order to fit the entire file.  If you want to zoom in to see individual words, you'll need to use the controls at the bottom left of the window.  The "in" button zooms in around wherever your cursor is, in this case the middle of the sound file.  The "sel" button zooms in to a selection.  You can drag your cursor across the sound to highlight a region, then press "sel".

You may have to zoom in a few times in order to get to individual words.  Once you do, you'll see something like this:

Now you can see two different representations of the sound.  The top shows the sound wave ("waveform") and the bottom shows the energy at different frequencies ("spectrogram").  The darker the spectrogram is, the more energy there is at that frequency. The blue line is pitch, which you can toggle off and on through the "Pitch" menu at the top.  You can also toggle on and off other analysis tools, like formant tracking and glottal pulses, but you won't need to worry about those for cutting sound files.  If you want to listen to just the part you zoomed in to, click on the bottom bar labeled "Visible part" or press Tab on the keyboard to play the selection.  If you click on the bar labeled "Total duration", it will play the entire sound file.  You can stop a sound file from playing by pressing Esc on the keyboard.

Now that you're familiar with basic controls for Praat, follow the instructions in the attached pdf prepared by my colleague Ryan Lidster.  It explains how to mark individual stimuli with intervals in a TextGrid and then pull out those words as individual sound files (using a PC, I'm not sure how well everything correlates to a Mac).

Here is a summary of the steps you'll need to take:

​​1. For each stimulus, highlight close to where the sound file begins and ends, about 30-50 ms before and after, though it doesn't need to be exact.  In general, you don't want the amount of silence before and after your stimuli to vary greatly, because this will affect your interstimulus interval (ISI) later.  [Note: If you're doing an experiment that requires exact timing of the stimuli, like priming or eyetracking, you will need to be more precise in your cutting, especially at the beginning of the stimulus.  One possible thing you could do is put your left boundary exactly at the start of the word, then use the Praat script "move_left_boundary_left_ for_labeled_intervals_&_zero_cross.praat" to change the left boundary to, for example, exactly 30 ms before the start of the stimulus.  The script will also zero cross both boundaries.]

2. Before you hit enter to make the boundaries, hit Ctrl + comma and Ctrl + period (or Command + comma and Command + period with a Mac) to make sure the boundaries are at zero crossings (i.e. the where the sound wave at the top crosses over zero) so you don't get clicking noises (if you've already hit enter, you can move the boundaries to the nearest zero crossing by going to "boundary" in the top menu).

3. You should have a spreadsheet in which you've prepared the file names for each stimulus.  Depending on the type of experiment, it can get confusing to name your files as simply the (non)word that they are, because then you won't have information about the condition or contrast they're being used in, which could make analyses more difficult.  Instead, it's often useful to label your files something like condition_sound_speaker.  For example, in our study on Finnish length perception, we labeled our files like "ata1.L", meaning the /ata/ context, with the first length template (i.e. all short segments; we had a code for each template, such as "2" for the first vowel was long, "3" for the first consonant was long, etc.), and "L" for the speaker. [Edit 4/26/22: I recommend using underscores instead of periods in your file names.  It turns out some Praat scripts don't play well with periods.]  Whatever you choose, make sure it's laid out very clearly somewhere. 

4. I find it useful to highlight my stimuli in this spreadsheet as I go along putting the labels in the TextGrid.  I use green for a good token, yellow for a somewhat questionable token, and red if there was something wrong with it and I didn't even mark that word.  It's always a good idea to keep track of your progress and take note of any problems you encounter.

4. Save your TextGrid often because Praat may crash.  While you're working in the sound file, go to "File" in the top menu --> "Save TextGrid as text file".

5. Once all the labels are in, run the script "save_labeled_intervals_to_wav_sound_files.praat" by going to the Praat Objects window --> "Praat" on the top menu --> "Open Praat Script" (if you're using a Mac, go to "Open"-->"Read from file").  Make sure the sound file is open as a long sound file ("Open" --> "Open long sound file") and the TextGrid is also highlighted.  If you access the script from my website, it will open in the .txt format in the browser instead of downloading as .praat.  You can copy the text of the script, then go to the Praat Objects window --> "Praat" --> "New Praat Script", and paste it in.

6. When you run the script ("Run" in the top menu in the open script --> "Run"), don't forget to put in the folder where you want all the sound files to go (including a backwards [PC] or forwards [Mac] slash at the end of the file path).  At this point you can add a prefix or a suffix to all your files, such as labeling them with the speaker.

You should now have sound files to use in your experiment!

The recording space:

• Use a sound booth if possible, otherwise pick the quietest place you possibly can.

• Avoid large spaces with echos.

• Avoid air conditioners, fridges, heaters, etc. running in background, since these will often create an audible hum in the recording.

• Place the mic about 2 inches from their mouth and slightly to the side to prevent large spikes in amplitude from their breath hitting the mic.

The recording style:

• Model how you want the words read or play a small part of a previous recording if you want to match speed/style of speaking.  

• You need clear pauses between each word.  I cannot stress enough how important this is for cutting the stimuli later.  They're basically worthless if there is no pause between them because there will be audible coarticulation, making segmenting the words into individual sound files very difficult and causing the words to sound weirdly chopped off when you try.

• Have them read with falling intonation on each word so that when the stimuli are cut, they don't sound like questions. This is difficult for speakers since it is natural to read with list intonation, which has rising intonation on each word.  If they're really bad at it, you may need to have them repeat after you for each word (making sure they are pausing sufficiently and not overlapping your speech).  Another technique is to have them say each word a few times as in a list, so that the final iteration of the word has falling intonation.  You can also avoid list intonation by putting stimuli in a sentence. This can help create more natural-sounding stimuli, but recording takes longer and cutting stimuli is more time consuming.  If you do embed words in a context, place the stimuli between stop consonants or repeat it after a sentence (e.g. "Say X again. X.") for ease of cutting later.  We've found that sentence-final tokens are clearest for perception experiments (if clear segments are your intention).  By repeating the word at the end of a sentence, it is easier for the speaker to produce it surrounded by a pause and with falling intonation.  

• If the speaker uses creaky voice, call attention to it and have them try to lessen it.  I've noticed people tend to do it more toward the end of the recording when they're bored and tired, so giving them breaks and water might help.  

• Watch out for speakers' tendency to speed up over time.  

Making the recording:

• Have them practice part of the list and listen to/look at the recording to check the mic levels, background noise, reading speed, pauses, etc.

• Record with a 3 to 1 ratio of recordings to number of stimuli you need.  In other words, if you need one good token of a word, record the list 3 times.  It may be necessary to record even more times and coach the speaker further if you need particular phonetic properties, such as final released [t] in English or dialectal variants like [h] for /s/ in Spanish. I recommend having the speaker read the list multiple times rather than reading each word multiple times in a row because people tend to say a word the same (possibly erroneous) way when repeating it.

​For most kinds of tasks, to calculate how long it will take you'll need to take into account the number of trials and how long each trial lasts.  When figuring out how many trials you need, keep in mind that for AX, you should have an equal number of same and different trials, and with lexical decision you should have an equal number of word and nonword trials.  For ABX-like tasks you'll need to balance the order of stimuli so that all six possible orders of stimuli are present: ABA, ABB, AAB, BAB, BAA, BBA.  In other words, if your contrast is [e] vs. [i], in one trial the order will be [kek], [kik], [kek] (ABA), in another [kek], [kik], [kik] (ABB), etc.  This ensures that X is equally likely to be A or B and that the order of presentation does not influence the results.  Oddity tasks should also have the six possible orders of stimuli, plus all-the-same trials (AAA, BBB).  You can either balance the number of same vs. odd-one-out trials or balance how often each button is the correct answer (first sound is different, second sound is different, third sound is different, or all the same).  I prefer the second option, since participants are likely to hear difficult contrasts as same trials anyway.  Note that the types of trials don't have to be perfectly balanced; for example, you could do 12 different trials (4, 4, and 4 in each position) and 8 same trials.  Just make sure they aren't too disparate.  

For the number of conditions, don't forget that you need a control condition to show that your task is working.   In our experiments, we've tested up to 10 contrasts in a discrimination task.  I think this is near the upper limit, since having more data points per condition is always better, especially if you plan to do individual-level analyses, and for each condition you add, the less trials per condition you'll be able to fit in.  Around 16-20 trials per condition is a good number, which may be split into a couple different phonetic contexts.  Here's an example from our latest oddity task:

10 contrasts (e.g. /u/ vs. /y/) x 10 trials per contrast (2 AAA, 2 BBB, AAB, ABA, ABB, BAA, BBA, BAB) x 2 contexts ([tVhVt], [kVhVk]) = 200 trials

It's helpful if you map out your trial set up in Excel, like this:
​

Oddball: 

• Participants hear a sequence of repeating stimuli which change to a new category after a certain number of trials.  Participants indicate when they heard a change in the category of the stimuli.  This technique originated with studies of infants, in which a change in category is indicated by a head turn, rather than by a button press as with adults.

Sequence recall:

• Participants are trained to associate two nonwords with two different keys.  For example, in Dupoux et al. (2008) /'numi/ was matched to key 1 and /nu'mi/ was matched to key 2.  Once participants are familiarized with the keys, they hearing a sequence of these nonwords and then have to type in the sequence they heard.  These sequences are followed by the word "okay" to prevent listeners from using echoic memory to complete the task.  Sequences start with two stimuli and increase incrementally until six stimuli in a row are presented.  

​​​Additional information:
Different discrimination tasks allow for different processing strategies, and tasks can be modified to increase or decrease the amount of cognitive load.  Strange and Shafer (2008) have a nice summary of perception tasks and the effects of different experiment conditions on performance.  An increase in cognitive demand reduces the possibility that participants can use acoustic cues to complete the task.  Physically different stimuli, multiple talkers, and embedding the stimuli in a context are all ways to force participants to process the contrast phonetically or phonologically, rather than searching for an acoustic match.  Cognitive demand can also be increased through a higher memory load.  For example, at the higher sequence lengths in a sequence recall task, even the native speakers are not at ceiling.  The effect of interstimulus interval (ISI) is somewhat mixed in the literature, with some studies finding a difference in performance at different ISIs (e.g. Werker & Tees, 1984) and others not (e.g. Tyler & Fenwick, 2012).  Overall, it seems that both an extremely short ISI (0ms) and an extremely long ISI (1500ms) make the task more difficult.  In my experience, an ISI of 1000ms is already very boring, and participants are more likely to let their attention wander.  

In general, AX tasks are easiest for participants, with difficulty increasing for AXB and oddity, followed by sequence recall, particularly at higher sequence lengths.  For example, in a series of studies Dupoux and colleagues found that French learners of Spanish could discriminate a stress contrast in an AX task with a single talker, but performed worse in an ABX task with multiple talkers.  In a sequence recall task, the performance of French participants was negatively affected by an increased sequence length, as well as by the amount of phonetic variability present in the stimuli.   In our research, we've found that the results of an AXB task with a 1000ms ISI and an oddity task with a 400ms ISI were very highly correlated.  Given this, I suggest using an oddity task since 1) the task is more intuitive, so participants are less confused by it and 2) chance level is much lower (25% for oddity vs. 50% for AXB), so a higher variability in scores is possible with oddity.

Results of discrimination tasks are typically reported in terms of accuracy or d' (d-prime) (or A'). d' is often a better metric because unlike accuracy rates, it is not affected by a participant's bias to answer one way or the other, e.g. a participant that responds 'same' to all trials.
​
4. Can L2 learners identify these L2 sounds?

Identification:

• Participants listen to a stimulus and pick what they heard from various options presented to them.  These options may be in the form of single segments or (non)words.  Identification tasks have been paired with training sessions in which participants learn to identify a non-native/L2 contrast and later are tested on their accuracy in identifying the sounds in this contrast (e.g. Bradlow et al., 1999).  This task has also been used with eyetracking in the visual world paradigm, in which pictures (or orthographic representations) of a target word, a possible competitor, and distractors are displayed on the screen while participants listen to the target stimulus (Weber & Cutler, 2004).  By tracking participants' gaze as they listen to the stimuli, it is possible to gain insight into lexical activation and competition. A variation of the identification task with synthetic stimuli on a continuum plus goodness ratings (e.g. Iverson & Kuhl, 1995) has been used to evaluate the Native Language Magnet Model (NLM) (Kuhl, 1993; Kuhl et al., 2008) for L1 phonology.  Identification tasks are not well suited for the investigation of non-native perception, since participants won't have labels for non-native sounds. This task is very useful, however, for examining whether learners associate certain variants with their respective phonemes.  For example, Schmidt (2011) examined whether learners had acquired [h] as a possible variant of coda /s/ in Spanish by playing nonwords like [bahpe] and providing different possible orthographic representations to choose from.  The following is a screenshot of the identification task in Praat from Schmidt (2011, p. 207):

​5. Can L2 learners represent this contrast in their mental lexicon?

Lexical decision:

• Participants hear a stimulus and decide if it is a word or non-word.  If participants do not have a contrast between two L2 sounds in their lexical representations, they will frequently accept nonwords as words.  For example, in Daidone and Darcy (2014), English-speaking learners of Spanish often accepted the trill as a possible realization of the tap, e.g. they indicated that quierro /kjero/ was a word, when the actual word is quiero /kjeɾo/.  Lexical decision tasks are often paired with priming in order to investigate lexical activation and competition.  Most of these lexical decision tasks are cross-modal, such that participants listen to a (non)word then respond to a visual, orthographic target. If participants are faster at responding to the word after a prime that contains a different sound, e.g. L1 Dutch speakers responding faster to groove after hearing the nonword groof, they have not encoded the relevant contrast in their mental lexicon, in this case final consonant voicing (Broersma & Cutler, 2008). These primes can also be fragment primes, i.e. only the beginning of a word, such as daf from daffodil followed by the visual target deaf in order to investigate whether Dutch learners of English have encoded this vowel contrast (Broersma, 2012).  A variation of this task is with repetition priming.  If learners do not have a contrast between words in a minimal pair, hearing one word of the minimal pair will result in a faster reaction time when the second word appears in the task (e.g. Pallier, Colomé, & Sebastián-Gallés, 2003).  It is important to use the keyboard or a response box for accurate reaction times for this task.

Word Learning:

• Participants learn a small group of nonwords that contain the relevant contrast, e.g. Japanese minimal pairs like keto and ketto with singleton and geminate consonants (Hayes-Harb & Masuda, 2008).  In the word-learning phase, participants learn to associate these words with a picture of their meaning.  This is followed by practice in which participants must indicate whether each word-picture pair they are given is correct; at this stage, they do not have to be sensitive to the test contrast.  Participants who fail to reach a certain criterion (e.g. 90% accuracy) must repeat this phase until they pass.  The test phase is similar to the practice phase, but no feedback is given and word-picture pairs that require sensitivity to the test contrast are presented, e.g. hearing keto but seeing the picture for ketto.  This type of task is commonly used to evaluate the effect of different  orthographic representations on participant's ability to encode contrasts in their mental lexicon (e.g. Showalter & Hayes-Harb, 2013).

​

I hope this summary of perception tasks has been helpful!  I also recommend checking out "A Brief Primer on Experimental Designs for Speech Perception Research" by Grant McGuire at UC Santa Cruz.  As you're deciding on a perception task, keep in mind that researchers frequently pair tasks together to look at different levels of processing.  A study may examine, for example, both participants' ability to discriminate a contrast and their ability to represent this contrast in lexical representations. Previous research has found that discrimination is easier than identification, which in turn is easier than any task tapping a lexical level of processing (Díaz et al., 2012; Ingram & Park, 1997; Sebastián-Gallés & Baus, 2005).

References

​Best, C. T. (1995). A direct realist view of cross-language speech perception.  In W. Strange (Ed.), Speech perception and linguistic experience: Issues in cross-language research (pp. 171-204). Timonium, MD: York Press.

Best, C., & Tyler, M. (2007). Nonnative and second-language speech perception: Commonalities and complementarities. In O.-S. Bohn & M. Munro (Eds.), Language experience in second language speech learning: In honor of James Emil Flege (pp. 13-34). Philadelphia: John Benjamins.

Bradlow, A. R., Akahane-Yamada, R., Pisoni, D. B., & Tohkura, Y. I. (1999). Training Japanese listeners to identify English /r/ and /l/: Long-term retention of learning in perception and production. Attention, Perception, & Psychophysics, 61(5), 977-985.

Broersma, M. (2012). Increased lexical activation and reduced competition in second-language listening. Language and cognitive processes, 27(7-8), 1205-1224.

Broersma, M., & Cutler, A. (2008). Phantom word activation in L2. System, 36(1), 22-34.

Clopper, C. G. (2008). Auditory free classification: Methods and analysis. Behavior Research Methods, 40(2), 575-581.

Daidone, D. & Darcy, I. (2014).  Quierro comprar una guitara: Lexical encoding of the tap and trill by L2 learners of Spanish.  In R. T. Miller, K. I. Martin, C. M. Eddington, A. Henery, N. Marcos Miguel, A. M. Tseng, …D. Walter (Eds.), Selected Proceedings of the 2012 Second Language Research Forum (pp. 39-50). Somerville, MA: Cascadilla Proceedings Project.

Daidone, D., Kruger, F., & Lidster, R. (2015). Perceptual assimilation and free classification of German vowels by American English listeners. In The Scottish Consortium for ICPhS 2015 (Eds.), Proceedings of the 18th International Congress of Phonetic Sciences. Glasgow, UK: Glasgow University.

Díaz, B., Mitterer, H., Broersma, M., & Sebastián-Gallés, N. (2012). Individual differences in late bilinguals' L2 phonological processes: From acoustic-phonetic analysis to lexical access. Learning and Individual Differences, 22(6), 680-689. 

Dupoux, E., Sebastián-Gallés, N., Navarrete, E., & Peperkamp, S. (2008). Persistent stress ‘deafness’: The case of French learners of Spanish. Cognition, 106(2), 682-706. 

Escudero, P., & Vasiliev, P. (2011). Cross-language acoustic similarity predicts perceptual assimilation of Canadian English and Canadian French vowels. The Journal of the Acoustical Society of America, 130(5), EL277-EL283.

Flege, J. E. (1995). Second language speech learning: Theory, findings, and problems. In W. Strange (Ed.), Speech perception and linguistic experience: Issues in cross-language research (pp. 233-277). Timonium, MD: York Press.

Fox, R. A., Flege, J. E., & Munro, M. J. (1995). The perception of English and Spanish vowels by native English and Spanish listeners: A multidimensional scaling analysis. The Journal of the Acoustical Society of America, 97(4), 2540-2551.
​
Hayes-Harb, R., & Masuda, K. (2008). Development of the ability to lexically encode novel second language phonemic contrasts. Second Language Research, 24(1), 5-33.

Ingram, J. C., & Park, S. G. (1998). Language, context, and speaker effects in the identification and discrimination of English /r/ and /l/ by Japanese and Korean listeners. The Journal of the Acoustical Society of America, 103(2), 1161-1174.

Iverson, P., & Kuhl, P. K. (1995). Mapping the perceptual magnet effect for speech using signal detection theory and multidimensional scaling. The Journal of the Acoustical Society of America, 97(1), 553-562.

Iverson, P., Kuhl, P. K., Akahane-Yamada, R., Diesch, E., Tohkura, Y. I., Kettermann, A., & Siebert, C. (2003). A perceptual interference account of acquisition difficulties for non-native phonemes. Cognition, 87(1), B47-B57.

Kuhl, P. K. (1993). Innate predispositions and the effects of experience in speech perception: The native language magnet theory. In Developmental neurocognition: Speech and face processing in the first year of life (pp. 259-274). Springer: Netherlands.

Kuhl, P. K., Conboy, B. T., Coffey-Corina, S., Padden, D., Rivera-Gaxiola, M., & Nelson, T. (2008). Phonetic learning as a pathway to language: New data and native language magnet theory expanded (NLM-e). Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1493), 979-1000.

Levy, E. S. (2009). On the assimilation-discrimination relationship in American English adults' French vowel learning. The Journal of the Acoustical Society of America, 126(5), 2670-2682.

Pallier, C., Colomé, A., & Sebastián-Gallés, N. (2001). The influence of native-language phonology on lexical access: Exemplar-based versus abstract lexical entries. Psychological Science, 12(6), 445-449.

Schmidt, L. B. (2011). Acquisition of dialectal variation in a second language: L2 perception of aspiration of Spanish /s/ (Unpublished doctoral dissertation). Indiana University, Bloomington, Indiana. 

Sebastián-Gallés, N., & Baus, C. (2005). On the relationship between perception and production in L2 categories. In A. Cutler (Ed.), Twenty-first century psycholinguistics: Four cornerstones (pp. 279-292). Mahwah, NJ: Lawrence Erlbaum Associates.

Showalter, C. E., & Hayes-Harb, R. (2013). Unfamiliar orthographic information and second language word learning: A novel lexicon study. Second Language Research, 29(2), 185-200.

Strange, W., Bohn, O. S., Nishi, K., & Trent, S. A. (2005). Contextual variation in the acoustic and perceptual similarity of North German and American English vowels. The Journal of the Acoustical Society of America, 118(3), 1751-1762.

Strange, W., & Shafer, V. L. (2008). Speech perception in second language learners: The re-education of selective perception. In J. G. Hansen Edwards & M. L. Zampini (Eds.), Phonology and second language acquisition (pp. 153-192). Philadelphia, PA: John Benjamins.

Tyler, M. D., & Fenwick, S. (2012). Perceptual assimilation of Arabic voiceless fricatives by English monolinguals. In INTERSPEECH 2012 (pp. 911-914).

Tyler, M. D., Best, C. T., Faber, A., & Levitt, A. G. (2014). Perceptual assimilation and discrimination of non-native vowel contrasts. Phonetica, 71(1), 4-21.

Weber, A., & Cutler, A. (2004). Lexical competition in non-native spoken-word recognition. Journal of Memory and Language, 50(1), 1-25.

Werker, J. F., & Tees, R. C. (1984). Phonemic and phonetic factors in adult cross‐language speech perception. The Journal of the Acoustical Society of America, 75(6), 1866-1878.