Do Semantic Contributions to Early Morphological Processing Vary With Target Semantic Density and Reading Skill?

View Segments Segment :

  • Citations
  • Add to My List
  • Embed
  • Link
  • Help
  • Citations
  • Add to My List
  • Embed
  • Link
  • Help
Successfully saved clip
Find all your clips in My Lists
Failed to save clip
  • Transcript
  • Transcript

    Auto-Scroll: ONOFF 
    • 00:09

      LAURIE BETH FELDMAN: So I'll be reportingon collaborative work that looks at semantic effectson morphological processing, and I'llbe introducing a new semantic variable thatseems to be doing a lot of work for us.If I were writing the title now, and to anticipate the results,we have effects of semantic density,

    • 00:30

      LAURIE BETH FELDMAN [continued]: and prime target similarity, and morphological processing,but it's the prime semantic densitythat's playing the role rather than the targetsemantic density.So, not everyone agrees about how best to define a morpheme.

    • 00:52

      LAURIE BETH FELDMAN [continued]: By the explicit morphology account,morphological processing as revealedby patterns in a lexical decision taskrequire-- facilitation requires decomposinga morphologically complex string into its constituent parts.And that decomposition is essential

    • 01:15

      LAURIE BETH FELDMAN [continued]: in order to be able to see effectsof morphological relatedness.In this account, so, it's the reactivationof the same morpheme, and prime, and targetthat drives facilitation, and the position of that morphemeshould have no effect.And in initial stages of processing,those morphemes are treated as units of form,

    • 01:37

      LAURIE BETH FELDMAN [continued]: and only later as units of meaning.So, early in the course of processing,if you adhere to the explicit morphology formand then meaning view, you would expect comparable facilitationfor pairs like battery bat and batter bat.

    • 01:57

      LAURIE BETH FELDMAN [continued]: So, batter and bat are semanticallyrelated, battery and bat are not,although structurally it appears to besharing the same morpheme.The alternative is an emergent morphology view,where morphological effects reflect concurrent influencesof form and meaning.

    • 02:18

      LAURIE BETH FELDMAN [continued]: And when you see effects of similarity, either formsimilarity or meaning similarity,they tend to be graded.So, the prediction would be that facilitation for battery batshould be less than facilitation for batter bat.And the account would look much moreat the activation dynamics, and how form and meaning are

    • 02:39

      LAURIE BETH FELDMAN [continued]: contributing to recognition.Important would be both the similarity between the primeand the target but, also, easily accommodatedare properties of the prime and properties of the target,with respect to their relationship to other words,in the mental lexicon.

    • 03:03

      LAURIE BETH FELDMAN [continued]: By an explicit morphology account,facilitation is decomposition-based,and you can talk about levels of processing through whichthe analysis goes.So, early in processing the emphasis is on shared form.So, battery and batch would be equivalent in terms

    • 03:25

      LAURIE BETH FELDMAN [continued]: of their ability to activate the target,and only later in processing would yousee a semantic influence.So, a morphosemantic contributionwould come in later, and this morphosemantic contributionis confined to the relationship between the primeand the target.So, just to anticipate, there is no anticipation

    • 03:47

      LAURIE BETH FELDMAN [continued]: of properties prime or propertiesof the target that play an effect other than perhapsfrequency.There's an alternative view, whichis that the semantic effects come in early,and right from the beginning morphological effectsreflect both similarity of form and similarity of meaning.

    • 04:11

      LAURIE BETH FELDMAN [continued]: Here, you'd expect greater facilitation for batter batthan for battery bat.This is the form with semantics view.Most important is that the activation dynamics are notall or none, they can be graded, and theycould encompass not only the similaritybetween the prime and the target,

    • 04:31

      LAURIE BETH FELDMAN [continued]: but also, easily, properties of the target,or the prime with respect to their similarityto other words.OK, most of that work was done-- all of thatwork was done with a forward masked primed lexical decisiontask, the idea being that that captures early processing.So, the important part is that the prime is up for justa very short amount of time, and that it's masked.

    • 04:53

      LAURIE BETH FELDMAN [continued]: So, this would-- farmer farm wouldbe equivalent to my batter bat example.Semantically related, and also sharesa sequence of letters that can be treated as a morpheme.Alternative, it would be examples like corner corn.Corner appears to be morphologically structured,as does battery, but in fact they from part target share

    • 05:16

      LAURIE BETH FELDMAN [continued]: form a not meaning.Pairs like this in a forward masked lexical decision taskdo produce differences in facilitation.That's just a greater facilitationwhen both form and meaning are sharedthan when only form is shared.

    • 05:36

      LAURIE BETH FELDMAN [continued]: And we've documented this both at a long-- arelatively long masked duration, 48, but also at 34.Now, in those experiments, when we constructed the materials,the similarity was based on, let's say, two dimensions.One is the LSA similarity between primes and targets,

    • 05:57

      LAURIE BETH FELDMAN [continued]: but the other is based on rating data.We ask a different group of subjectsto determine whether these words are related in meaning as wellas in form.In the new results that I'd like to present today,I want to kind of move away from those semantic measuresto something that's more based on distributional semantics.

    • 06:18

      LAURIE BETH FELDMAN [continued]: OK, so we're interested in whether the presenceof a shared morpheme is sufficient to inducemorphological facilitation, and whatthe time course of semantic effects are,and whether those semantic effects arerestricted to the similarity between the primeand the target.OK, the new experiment's going to lookat SOAs of 300 milliseconds and 48 milliseconds.

    • 06:41

      LAURIE BETH FELDMAN [continued]: Rather than the suffixed pairs thathave been examined most frequently in the past,we're going to be looking at prefix pairs.Obviously, prefix pairs don't have initial form similarityin the way that suffix pairs do.So, farmer farm, you can see that farm right upin the front of the prime.On the other hand, if you have suffixed pairs,

    • 07:02

      LAURIE BETH FELDMAN [continued]: then the similarity is attenuated inso far as the position of the sequence of lettersthat overlaps is not always word initial.OK, we're going to move away from ratingmeasures of semantic similarity and work from two measures.One is co-occurrence similarity, and thena second is based on co-occurrence similarity,

    • 07:23

      LAURIE BETH FELDMAN [continued]: and it reflects semantic density.In this case, it's going to be of the prime.So, given the mixed background in this room,kind of a word about distributional semanticsand how we use them to capture semantic similarity.The idea is that if words appear in similar contextthey must be similar in meaning.OK, so, the little matrix there is

    • 07:46

      LAURIE BETH FELDMAN [continued]: supposed to correspond to words on the rowsand context on the columns.And we keep track of where a word occurs and do fancy thingsto reduce the dimensionality of the matrix,

    • 08:07

      LAURIE BETH FELDMAN [continued]: and normalize the vectors, and come up with a vector thatcorresponds to each of the words and the context in which itappears.And then to compare the similarity of the vectors,we look at the angle, or the cosine similarity of the twovectors.So, we're using the cosine angle as a geometric quantification

    • 08:32

      LAURIE BETH FELDMAN [continued]: of how similar two words are in meaning.So, just to give you a depiction,here are four morphological relatives of the word act--actuate, transact, and action.Action has the highest cosine similarity.Actuate has the lowest cosine similarity.OK, so it's a way to capture how similar

    • 08:53

      LAURIE BETH FELDMAN [continued]: the prime and the target are when they share a morpheme.Here's a [INAUDIBLE] depiction of the same semantic measure.OK, so we have act and four of its morphological relatives.So, again, it's just-- and it's not even to scale,but you get the idea that the larger numbers aregetting bigger circles there.OK, so, in the data that I'm going to show you

    • 09:13

      LAURIE BETH FELDMAN [continued]: we have targets, and they're each preceded by two primes.Those two primes share a morpheme,and differ on cosine similarity.Some of the values are not intuitive,but I'll kind of let you-- so, account and miscount.

    • 09:34

      LAURIE BETH FELDMAN [continued]: Account, by the distributional accounts,is more similar to count than is miscount.OK, semantic density reflects the average cosine similarityof the 20 closest neighbors to the target,and so we can look at this density measure for targets.OK, and here's some lists of them.

    • 09:56

      LAURIE BETH FELDMAN [continued]: We can go back to that.We can also look at it for primes.So, again, it's the 20 top co-occurrence neighborsof the word, and they don't have to be morphologically related.They can be, but they're not always.OK, so this is enacts semantic neighbors,average density 0.78.Transact semantic neighbors, average density 0.62.

    • 10:20

      LAURIE BETH FELDMAN [continued]: Just as a rule, those numbers tendto be lower than the numbers for suffix primes.So, the result. Between subject's manipulation of SOA,vocabulary and spelling tests tossed in there,but they're not doing much work for us in our models.First long SOA priming data, transact act and act act.

    • 10:41

      LAURIE BETH FELDMAN [continued]: OK, so we have our prime types, wehave our continuous measures of semantic similarity,and we have two SOAs.This is the general additive mixed model.Most interesting here is going to be that tensor product thatlooks at the interaction between cosine similarityand prime density.But, let me just say that these semantic effects are only

    • 11:02

      LAURIE BETH FELDMAN [continued]: occurring in the long SOA condition,and they're occurring such that the two morphological primesare different than the unrelated prime, but let me unpack that.OK, first, the 300 SOA data.The two morphological primes are differentthan the unrelated prime, and we hadthis provocative interaction between the twosemantic measures only at the long SOA.

    • 11:23

      LAURIE BETH FELDMAN [continued]: At the shorter SOA, we got nada.So, there's no effect of prime type,there's no effect of the semantic density.OK, so here is the main piece of data that I have for you.On the x-axis here, we have the similaritybetween the prime and the target only,captured by cosine similarity.On the y-axis, we have prime semantic density,

    • 11:45

      LAURIE BETH FELDMAN [continued]: the closeness of the 10 top neighbors of the prime.For those of you who are not familiar with these pictures,green means fast, brown means slow.OK, so, what you can see here is that the influenceof prime similarity is much stronger

    • 12:07

      LAURIE BETH FELDMAN [continued]: when the prime density is low.Where do you get the most gradations in color?When prime density is low, then youget the effects of semantic similarity.OK, if we look the other way we seethat where you get the most gradation in colorswith respect to prime semantic densityis when the cosine similarity is relatively high.

    • 12:27

      LAURIE BETH FELDMAN [continued]: So, there's interactions between the two semantic measures.What happens at the low SOA, at the short SOA condition?Not a lot.No morphological facilitation with prefixprimes at the 48 millisecond SOA.OK, so, we have semantic contributionsto morphological priming that are varyingwith semantic density, and under some conditions

    • 12:50

      LAURIE BETH FELDMAN [continued]: the semantic density of the primecan offset the semantic similarity between the primeand the target.So, just to kind of summarize in here,the presence of a shared morphemeis not always sufficient to producemorphological facilitation at a short SOA,and this calls into question decompositionas an account of early morphological processing.

    • 13:13

      LAURIE BETH FELDMAN [continued]: Of course, there's attenuated form overlapand when primes are prefixed, and the prefix prime targetcosine values tend to be lower than the suffixed ones.OK, when the SOA is long, not only prime target similarity,but also something about the activationamong the semantic neighbors of the prefixed word

    • 13:34

      LAURIE BETH FELDMAN [continued]: are slowing down reaction times in a sense, competingwith activation of the target.OK, so, conclusions-- the prefix wordsand the suffixed morphological relativesare not working the same way.So, I want to suggest that an account of activationof morphological processing based on a shared morpheme

    • 13:55

      LAURIE BETH FELDMAN [continued]: must be inadequate.Semantic similarity is important at the long SOA.It interacts with the similarity between the primeand the target.So, we're getting the activation dynamicsfrom the prime modulating lexical processingat the long SOA.OK, and so, as anticipated at the beginning,

    • 14:17

      LAURIE BETH FELDMAN [continued]: we have semantic contributions to morphological primingthat vary with prime density and with prime target similarity.And I'll just stick in there that we haveno effects of reading skill.That's to say, reading skill contributed nothingto our model over and above a random effect of differencesbetween subjects.Thank you.[APPLAUSE]

Do Semantic Contributions to Early Morphological Processing Vary With Target Semantic Density and Reading Skill?

View Segments Segment :


Professor Laurie Beth Feldman presents her research into how test subjects process words with a shared morpheme. Cognitive processing speed varies depending on how semantically similar words are, and depending on whether the shared morpheme is at the prefix or suffix position.

SAGE Video Forum
Do Semantic Contributions to Early Morphological Processing Vary With Target Semantic Density and Reading Skill?

Professor Laurie Beth Feldman presents her research into how test subjects process words with a shared morpheme. Cognitive processing speed varies depending on how semantically similar words are, and depending on whether the shared morpheme is at the prefix or suffix position.

Copy and paste the following HTML into your website

Back to Top