認知心理学4. short-term working memory：短期記憶

認知心理学もくじ

Information processing so far

• sensory memory (感覚記憶):
  • limited capacity
  • limited duration 250ms
• iconic memory.
• pattern recognition
  • attention resource to determine which source we are giving attention to.
• short-term memory

The Atkinson & Shiffrin Modal Memory Model

アトキンソンとシフリンは、外界から入力された情報（刺激）は、最初に自動的（無意識的）に感覚登録器(sensory register)に入力され、その情報は感覚記憶(sensory memory)としてごく短時間だけ保持されると考えました。視覚刺激の感覚記憶は『アイコニック・メモリー(iconic memory)』と呼ばれ、その持続時間はスパーリング(Sperling)の実験によると約500ミリ秒以内であるとされています。聴覚刺激の感覚記憶は『エコイック・メモリー(echoic memory)』と呼ばれ、その持続時間はグルックスバーグとコワン(Glucksberg & Cowan)の実験によると約5秒以内であるとされています。本人が意識しない間に自然に外界から入ってくる情報は、感覚登録器で感覚記憶となりますがその持続時間は極めて短いことが分かります。

感覚登録器(sensory registers)に自動的に入力された情報の中で、選択的注意 (selective attention)を向けられた情報は、『短期貯蔵庫(STS)』に格納されて情報保持の持続時間が約15〜30秒に延長されるとアトキンソンとシフリンはいいます。短期貯蔵庫に情報を一時的に保存することによって、人間は『意図的な情報処理の選択』を行うことが出来るようになります。情報の短期貯蔵庫が存在することによって、リハーサル(rehearsal)やコーディング(coding)といった記銘処理をしてその情報を更に長い時間保持するのか、それともその情報をリハーサルせずにそのまま忘却してしまうのかを選択できるというわけです。
(短期記憶のメカニズムを説明する基礎理論)

• Structural Features
  • sensory store
  • STM: capacity 7 ±2
  • LTM: unlimited capacity

• Control Processes
  • Attention influences on number, duration, by maintenance rehearsal
  • e.g.: notecard memorizing is very poor strategy because not thinking about semantic meanings. this is only good for increasing duration of memory.
• elaboration
  • encoding deeper semantic meanings.
  • different from rehearsal.
  • understanding meaning behind the concept
• retrieval

The Atkinson & Shiffrin Multistore Memory Model (1969)

	SIS	STM	LTM
Forgetting	Decay	Interference	Retrieval Failure
Duration	~250 ms	10-30 s	lifetime
Capacity	Quite Large	7±2 chunks	Unlimited
Code	Simple sensory features	Acoustic/semantic	Semantic

• STM:
  • interference of other info coming.
  • can expand using maintenance rehearsal
  • Acoustic: what is the word sounds like.

Duration of STM

• Peterson & Peterson Task (1959)
  • Subjects hear 3 letters (CHJ)
  • Subjects hear a random number (506)
  • Count backwards (by 3s)
  • Upon seeing a cue (i.e., light) try to repeat letters

Results of Peterson Task

• As we can see, when the inter stimulus interval was 0 subjects could recall the 3 letters nearly 100% of the time.
• At 3 sec. recall is still good around 80% correct.
• At 6 sec recall is o.k. above 50%
• However, we see a rapid decline from 9 sec. on. At 9 sec it has dropped from 60% to below 40.
• And at 18 sec we see that their ability to recall the letters is gone.
• problem: letters do not have any meaning.

Forgetting In Short Term Memory

• 2 theories
  1. decay theory
     1. Proposal that information is spontaneously lost over time, even when there is not interference from other material (CP 76)
  2. interference theory
     1. proposal that forgetting occurs because other material interferes with the information in memory (CP 76)
• But how do we know whether forgetting in STM is do decay or interference.
  • If forgetting is due to decay, then recall should be determined by the length of the retention interval.
  • If forgetting is due to interference, then recall should be determined by the number of interfering items.

• Waugh & Norman (1965) Probe Digit Task
• Waugh & Norman tested whether forgetting was due to interference or decay.
• Using a probe digit task they presented subjects with lists of 16 single digits.
• The last digit in every list was the probe digit and occurred exactly once earlier in the list.
• Task: Report the digit following the probe digit.
• If the test item occurred late in the list there were fewer interfering digits. If the test item occurred early in the list then there were many interfering items.
• Waugh and Norman also varied the rate at which Ss heard the digits. Digits were presented at a rate of either 1 digit per sec or 4 digits per sec.
• Few interfering digits
  • 5646446578323278
• Many interfering digits
  • 3746844657323278
• 8の場合は前にその次にあった３と答える
• Decayなら、早く数字をpresentしたほうがいい結果を出すはずだ。
• Rate: 1 digit per sec. or 4 digits per sec.

Results of Waugh & Norman

• Between decay and interference, interference wins.
  • Decay: cannot prevent loss.
  • Interference: improve retention by structuring learning.
• And this good news. If forgetting were due to decay then we would be unable to prevent the loss of info. But because forgetting is due to interference we can improve retention by structuring learning to minimize interference.
  • もしdecayが理由なら人はどうしようもないが、Interferenceが理由であるため、structuring learningによってretentionをimproveできる。
• problem: words do not have any meaning

2 types of Interference

• retroactive interference (RI)
  • Forgetting that occurs because of interference from material encountered after learning. (CP 78)
  • occur after the event.
  • 新しい携帯を買ったら、昔の携帯の番号を忘れた。
• proactive interference (PI)-
  • occur before the event.
  • Forgetting that occurs because of interference from material encountered before learning (CP 78)
  • 古い電場番号のせいで今の番号を覚えられない。
• Release from proactive interference
  • Wickens ,Born and Allen demonstrate the release from PI phenomenon by having subjects remember 3 numbers or 3 words for 20 sec. while performing another task at the same time to keep from rehearsing.
  • Reducing proative interference by having information be dissimilar from earlier material. (CP 78)
  • 似たようなことを連続して勉強しない
• Wickens, Born, & Allen (1962).
  • Remember 3 numbers or 3 words for 20 sec. while performing another task (4 trials).
  • Experimental Group: First three trials from the same category 4th trial different category.
  • Control Group: All four trials from same category.
  • For the experimental group the first three trials were from the same category and the 4th trial was from a different category. EXPLAIN

The control group received the same category for all 4 trials.

Release from PI

• We can compensate for rapid forgetting by:
  • ordering information into the proper sequence.
  • presenting similar information at different times.
• But What about STM capacity limitation

Capacity of STM

• memory span (the number of correct items that people can immediately recall from a sequence of items (CP 80)) or digit span task-

T Y N J A S D
G V R O K Z F M P 
U X W L F T H Y J
U F N W S A K

• So we’ve talked about how information gets out of STM, but how much information can be in there in the first place?
• Most often, this is studied in the laboratory by memory span procedures: word span, digit span, operation span. Some tasks are simple span tasks (digit span, word span) that measure the size of STM.
• The task most often used in determining the number items that can be held at one time in memory is the memory span task also called the digit span task.

For example participants would read each row of letters once and then try to recall the letters in the correct order.

• Its probably not very hard to recall a string of 7 letters such as row 1 and 4. But it would be much more difficult to recall a nine letter string as seen in rows 2 and 3.
• So this gives us a measure of a persons memory span and is known as a simple span task.
• And what we find is that people can hold an average of 7 items in STM at any one time.

• absolute judgment task
  • Identifying stimuli that vary along a single, sensory continuum (CP 80)
  • different loudness of the sound. low-high.
• Present Ss with stimuli that can be easily discriminated and assign value.
  • For example if there were 7 stimuli the softest would be assigned a 1 and the loudest a value of 7.
• Task: correctly identify the value of each stimulus.
• Although the results vary a little depending on the sensory modality the upper limit seems to be 7 items plus or minus 2
• youtube video

Increasing memory span

• "The magical number seven, plus or minus two: Some limits on our capacity for processing information." (Miller, 1956).
• chunk
  • items that have been combined to form a unit in LTM.
  • a cluster of tems that has been stored as a unit in LTM (CP 81)
• Chunking allows us a way around the STM capacity limitations.
• For example, we can see that this letter string contains nine chunks at the moment. And recalling these chunks would be very difficult for two reasons. One is that it is at the upper limits of our STM capacity, and secondly these chunks have no meaning to us.
  • FB-IT-W-AC-I-AIB-M-N-ASA
• But if we chunk this letter string differently we find not only are there fewer chunks to be remembered but now the chunks are meaningful to us, and therefore it would be much easier for us to not only hold these chunks in STM but store them in LTM as well.
  • FBI-TWA-CIA-IBM-NASA
• How big is a chunk?
  • More experience you have within a certain domain, such as football, or basketball or whatever you can combine more feature together into a meaningful unit.

Expert knowledge and STM capacity

• The classic study done on this was done by a Dutch psychologist named de Groot in the 40s and 50s.
• He showed participants chess boards from 20 moves into actual matches for 5 seconds. He then had the participants recreate the board after the pieces were taken off. The participant would replace the pieces, the experimenter would remove the incorrect ones, and the P would go again- up to 12 times. The number of trials it took to perfectly recreate the board was the D.V. Some of the participants were chess masters, others knew how to play...

• Master Chess Players: can recreate 90% of the board on first try.
• Regular Chess Players: only 40% on first try.
• How do we know that master chess players just aren’t better a guessing?

de Groot (1965)

• To test this possibility he had both the masters and the novices simply guess where the pieces were located without ever having seen the board.
• Graph: Chess masters after 5 sec exp. Are getting 90% right on the first trial and they are pretty much at 100% from trial six on.
• Weak players after 5 sec exposure begin at 40% but don’t get any higher then 95% and that takes them till trial 10.
• But when both the master and weak players guess there is not much of a difference between the two. Master chess players show slightly better performance over weak players, but the difference is not significant.

In summary

• Forgetting in STM is due to interference, NOT decay.
• retroactive interference is caused by information occurring after the event.
• proactive interference is caused by info occurring before the event.
  • retroはどうしようもない, proはdisimilarでなんとかなる。
• By making items dissimilar we can get a release from proactive interference.
• expand capacity by chunking(chunk)

review

• forgetting is due to
  • sensory memory: decay
  • STM: interference of other incoming info.
• The magical number seven
• chunking(chunk)

Memory Codes

• acoustic code
  • A memory code based on the sound of the stimulus (CP 83)
  • actually has meaning.
• semantic code
  • A memory code based on the meaning of the stimulus (CP 83)
• STM and LTM can handle both acoustic, semantic,

Acoustic Codes

• 2 types related to verbal rehearsal

1. speed of pronunciation
   1. how fast people can express
2. retrieval
   1. how fast people can recall

Verbal Rehearsal

• acoustic confusion:
  • An error that sounds like the correct answer (CP 85)
• If acoustic confusions occur, an error would be more likely to involve substitution of a letter from the same group than substitution of a letter from a different group.
• Conrad found that 75% of the errors involved one of the other four letters belonging to the same acoustic group, and 25% of the errors involved one of the five letters in the other acoustic group.
• It is particularly easy for acoustic confusion to occur when all letters in a sequence sound alike.
• Try to recall the letters in each of the two following rows.
• You should find that the letters in the second row are easier to recall than the letters in the first row.

G Z D B C P T V
M J Y F H R K O

• But how do they occur?

• phoneme:
  • any of the basic sounds of a language that are combined to form speech (CP 85)

• It is convenient to use phonemes to account for acoustic confusions because words that sound alike usually have some phonemes incommon.
• Let's look again at the tow sets of letters in Conrad's experiment.
• The names of the letters in the set FMNSX (ef, em, en, es, ex) have the same initial phoneme, but their second phoneme differs.
• The letters in the set BCPTV (be, se, pe, te, ve) all share a common phoneme, but have different first phonemes.
• The major assumption of a model proposed by Laughery is that each of the auditory components representing an item can be independently forgotten.
• In other words, if a name consists of two phonemes, a person might remember one phoneme but not the other. The model also assumes that the auditory components can be forgotten at different rates; the decay rates were determined from experimental result.

(CP 85)

G Z D B C P T V

• Laughery (1969)
  • auditory components can be independently forgotten.
  • i.e.,if a name consists of two phonemes, a person might remember one phoneme but not the other.

Acoustic Codes and Reading

• subvocalizing:
  • Silently speaking to oneself (CP 86)
• Levy (1978)
  • Subjects had to read a short passage.
  • And continuously count from 1 - 10.

''An emergency''
The hospital staff paged the busy doctor.
The solemn physician distressed the anxious mother.
The sobbing woman held her unconscious son.
A speeding truck had crossed the mid-line.
Her oncoming car was hit and damaged.
Her child had plunged through the windshield.
The medical team strove to save him.

• lexical alteration:
  • Substituting a word with similar meaning for one of the words in a sentence (CP 86)
  • e.g.:The solemn physician distressed the anxious woman.
• semantic alteration:
  • Changing the order of words in a sentence to change the meaning of the sentence (CP 86)
  • e.g.:The solemn mother distressed the anxious physician.
• people are very bad at this regardless which alterations they are using

• However
  • Suppressing subvocalization (subvocalizing) did not interfere when subjects listened to the sentences.
    • because while reading have to convert visual code to acoustic code, but listening does not need that conversion.
  • Using paraphrased sentence (using different words to express the same ideas in a sentence (CP 87)) improves performance as well.
    • e.g.: The solemn doctor upset the anxious mother.

Recognition

• memory set:
  • A set of items in STM that can be compared against a test item to determine if the test item is stored there (CP 89)
• Involves two processes:
  1. encode: (create acoustic)
     1. to create a visual or verbal code for a test item so it can be compared with the memory codes of items stored in STM (CP 89)
  2. scan: (find if it actually appears in STM)
     1. To sequentially compare a test item with items in STM to determine if there's a match (CP 89)

• Sternberg (1966)

Sternberg(1966)は、一桁の数字を単時間提示した後、新たな数字を提示し、それが前に提示されたものと同じかどうかを再認してもらい、同じであれば、ｙのボタンを、異なればｎのボタンをおしてもらうことによって、反応時間を測定した。その結果、短期記憶における想起では、並列走査、直列打ち切り走査、直悉皆走査のうち、直列悉皆走査が行われていることを発見した。
(記憶のメカニズム II)

• • 5 3 6 8 (memory set)
  • Probe: 6
• I.V.: number of items in memory set.
• D.V.: time to make yes or no decision.
• responding "no" took little bit longer than responding "yes".

Result

• exhaustive search:
  • a search that continues until the test item is compared with all items in the memory set (CP 89)
• self-terminated search:
  • A search that stops as soon as the test item is successfully matched to an item in the memory set (CP 89)

5  3  6  8

• このようなセットが与えられ、3を探せと言われたとき、見つけた時点でサーチを終わるか（self-terminated search）、セットを最後まで見てから一致するかどうかを調べるか(exhaustive search)
• use exhaustive search, because if people use self-terminated search, they have to compare every time they see a new input, which takes longer time than just get all information first, then compare.

Working Memory

• Baddeley & Hitch (1974)
  • semi independent slave system

1）まず中央実行系Central Executiveです。これは注意管理システムan attentional control systemです。作動記憶の中でもっとも重要なコンポーネントで、処理容量が限定されており、認知的に最も困難な課題を扱うときに使われます。
2）それを２つの補助システムslave systemsが補佐します。音声ループphonological loopと視覚・空間スケッチパッドvisuo-spatial sketchpadと言われるものです。中央実行系はこれらの２つの従属システムを制御します。目的とする作業や活動がスムーズに行われるよう全体を見渡し、２つのスレーブシステムに作業を割り振ったり、必要な記憶容量を確保したりします。中央実行系はあくまでも注意制御システムであって、情報の保持機構ではないので、実際の情報は音韻ループや視空間スケッチパッドで貯蔵されることになります。（ここでもう一度確認しておいてください。作動記憶とは従来の短期記憶のことです。ただ、作動記憶の場合には単なる貯蔵だけでなく、そこで「処理」も行なうという点が違うのです。）

音韻ループは、文章の理解などの言語的な情報処理に関わるもので、リハーサルによって情報を保持します。音韻ループは、音声情報を保持する受動的な音韻性短期貯蔵(phonological short-term store, passive phonological store)と、能動的に音声情報を反芻する構音コントロールプロセス(articulatory control process)という下位システムが含まれています（図を参照）。音声情報は3つの経路を通じて音韻性短期貯蔵に入ります。?直接音声を聴取することで?subvocal articulationを通じて間接的に?長期記憶に貯蔵されている音声情報を通じて間接的に。

他方、視覚・空間イメージなどの言語化できない情報の処理を行なうメカニズムが、視空間スケッチパッドです。これも視覚・空間情報を貯蔵する受動的な視覚キャッシュ(visual cache)と、映像を頭の中で反芻する能動的なインナースクライブ(inner scribe) から構成されています。
(作動記憶と通訳研究)

1. 1. phonological loop:
      1. a component of Baddeley's working memory model that maintains and manipulates ''acoustic information'' (CP 91)
      2. responsible for maintaining and manipulating ''speech-based information''
   2. visuospatial sketchpad
      1. A compoinent of Baddeley's working memory model that maintains and manipulates ''visual / spatial information'' (CP 91)
      2. responsible for maintaining and manipulating visual or spacial information.
   3. central executive
      1. a component of Baddeley's working memory model that ''manages the use of working memory'' (CP 92)
      2. responsible for selecting strategies and integrating information

Phonological Loop

• Two components:

1. phonological store:
   1. for holding verbal information
2. rehearsal mechanism
   1. for keeps the information active in the phonological store

• Evidence
  • Phonological Similarity
    • Lists of words that sound similar are more difficult to remember than words that sound different. Semantic similarity (similarity of meaning) has comparatively little effect, supporting the assumption that verbal information is coded largely phonologically in working memory.
  • Articulatory suppression

構音抑制
これは被験者が’the’とか’bla’とか、無関係な言葉をくりかえしながら、視覚的に提示された数字列を記憶しようとするようなことを言います。これまでの作動記憶のモデルでは、構音抑制、つまり無関係な言葉を繰り返し発することでphonological loopを絶えず一杯の状態にしておき、（この場合は数字の）リハーサルに使わせないようにすること
(作動記憶と通訳研究)

• • • Memory for verbal material is impaired when people are asked to say something irrelevant aloud. This is assumed to block the articulatory rehearsal process, thereby leaving memory traces in the phonological loop to decay.
  • Transfer of info between slave systems.
    • With visually presented items, adults usually name and sub-vocally rehearse them, so the information is transferred from a visual to an auditory code. Articulatory suppression prevents this transfer, and in that case the above mentioned effect of phonological similarity is erased for visually presented items.
  • Neuropsychological evidence:
    • phonological store and rehearsal occur in different brain regions.
    • A defective phonological store explains the behavior of patients with a specific deficit in phonological short-term memory. Aphasic patients with dyspraxia are unable to set up the speech motor codes necessary for articulation, caused by a deficiency of the articulatory rehearsal process

(STM = working memory)

Visuospatial Sketchpad

• Logie (1995) Two components:
  1. Visual Cache
     1. stores information about form and color.
  2. Inner Scribe
     1. deals with spatial and movement information. It also rehearses information in the visual cache and transfers information to the central executive.

• Evidence
  • Less interference between a visual task and a spatial task then two visual tasks.
  • Brain damage can effect one but not the other.
  • Visual tasks activate left hemisphere and spatial tasks activate the right hemisphere.

Central Executibe

• The central executive is a flexible system responsible for the control and regulation of cognitive processes.
• It has the following functions:
  1. binding information from a number of sources into coherent episodes
  2. coordination of the slave systems
  3. shifting between tasks or retrieval strategies
  4. selective attention and inhibition
• falsifiable?
  • Recent research on executive functions suggests that the 'central' executive is not as central as conceived in the Baddeley & Hitch model. Rather, there seem to be separate executive functions that can vary largely independently between individuals and can be selectively impaired or spared by brain damage.

Evidence for distinct systems.

• Main task: recreate chess board (uses V.S.)
• Secondary tasks:
  • Make an irrelevant sound repeatedly (suppresses P.L.)
  • Random letter generation (suppresses C.E.)
  • Pressing buttons in a predetermined pattern (suppresses V.S.)
  • Control condition (no secondary task)
• Result

Working V. Short-term Memory

エビングハウスの忘却曲線の記憶研究を嚆矢とする記憶分野の研究活動の発展によって、現在では、短期記憶を『作動記憶（作業記憶, working memory）』という概念で表現することもあります。短期記憶の概念では、アトキンソンやシフリンが仮定した『情報の短期的な貯蔵庫』としての意味合いが強くなっていますが、作動記憶（ワーキングメモリー）の概念では、読書・計算・コミュニケーション・記憶課題への回答・学習行動などの認知活動で、情報がどのように操作されて利用されるのかという『情報処理機能としての記憶』の意味合いが強くなっています。
(短期記憶のメカニズムを説明する基礎理論)

• Is WM and STM the same?
  • STM is part of working memory.
  • Baddeley says yes.
  • Engle and Turner (1989) say no.

Engle and Turner (1989)

• Operation Span task.
  • combines verification of brief mathematical equations such as "2+6/2 = 5?" with memory for a word or a letter that follows immediately after each equation.
  • IS (4-2) + 4 = 16? PAINT
• O-span scores predict scores on many other tasks:
  • SAT, ACT, and other IQ tests
  • GPA
• Central executive = controlling attention

Revised Working Memory Baddeley (2000)

• If the central executive allocates attention where is information integrated?
• episodic buffer
  • a storage system that can integrate memory codes from different modalities. (CP 94)
  • It explains how you could mentally form a visual map from verbal directions. (道を尋ねる例など)

Newer ideas

• Unsworth & Engle (2007)
• Working memory is needed for two main tasks:
  1. Overcoming automatic processing (e.g. stroop effect)
  2. Maintaining information in primary memory and retrieving information from secondary memory

認知心理学もくじ

• 認知心理学もくじ