In [1]:
from __future__ import print_function
import os
import time
import numpy as np
import tensorflow as tf
import pandas as pd
from collections import defaultdict
from sklearn.metrics import roc_auc_score, accuracy_score
import nltk
from correct_text import train, decode, decode_sentence, evaluate_accuracy, create_model,\
get_corrective_tokens, DefaultPTBConfig, DefaultMovieDialogConfig
from text_correcter_data_readers import PTBDataReader, MovieDialogReader
%matplotlib inline
In [2]:
root_data_path = "/Users/atpaino/data/textcorrecter/dialog_corpus"
train_path = os.path.join(root_data_path, "movie_lines.txt")
val_path = os.path.join(root_data_path, "cleaned_dialog_val.txt")
test_path = os.path.join(root_data_path, "cleaned_dialog_test.txt")
model_path = os.path.join(root_data_path, "dialog_correcter_model_testnltk")
config = DefaultMovieDialogConfig()
In [3]:
data_reader = MovieDialogReader(config, train_path)
In [4]:
train(data_reader, train_path, val_path, model_path)
Reading data; train = /Users/atpaino/data/textcorrecter/dialog_corpus/movie_lines.txt, test = /Users/atpaino/data/textcorrecter/dialog_corpus/cleaned_dialog_val.txt
---------------------------------------------------------------------------
KeyboardInterrupt Traceback (most recent call last)
<ipython-input-4-c4f4a34e9f4a> in <module>()
----> 1 train(data_reader, train_path, val_path, model_path)
/Users/atpaino/github/deep-text-correcter/correct_text.pyc in train(data_reader, train_path, test_path, model_path)
138 "Reading data; train = {}, test = {}".format(train_path, test_path))
139 config = data_reader.config
--> 140 train_data = data_reader.build_dataset(train_path)
141 test_data = data_reader.build_dataset(test_path)
142
/Users/atpaino/github/deep-text-correcter/data_reader.pyc in build_dataset(self, path)
125 # dropouts.
126 for _ in range(self.dataset_copies):
--> 127 for source, target in self.read_samples(path):
128 for bucket_id, (source_size, target_size) in enumerate(
129 self.config.buckets):
/Users/atpaino/github/deep-text-correcter/data_reader.pyc in read_samples(self, path)
113 """
114 for source_words, target_words in self.read_samples_by_string(path):
--> 115 source = [self.convert_token_to_id(word) for word in source_words]
116 target = [self.convert_token_to_id(word) for word in target_words]
117 target.append(EOS_ID)
/Users/atpaino/github/deep-text-correcter/data_reader.pyc in convert_token_to_id(self, token)
77 :return:
78 """
---> 79 token_with_id = token if token in self.token_to_id else \
80 self.unknown_token()
81 return self.token_to_id[token_with_id]
KeyboardInterrupt:
In [3]:
data_reader = MovieDialogReader(config, train_path, dropout_prob=0.25, replacement_prob=0.25, dataset_copies=1)
In [5]:
corrective_tokens = get_corrective_tokens(data_reader, train_path)
In [6]:
import pickle
with open(os.path.join(root_data_path, "corrective_tokens.pickle"), "w") as f:
pickle.dump(corrective_tokens, f)
In [8]:
import pickle
with open(os.path.join(root_data_path, "token_to_id.pickle"), "w") as f:
pickle.dump(data_reader.token_to_id, f)
In [5]:
sess = tf.InteractiveSession()
model = create_model(sess, True, model_path, config=config)
Reading model parameters from /Users/atpaino/data/textcorrecter/dialog_corpus/dialog_correcter_model/translate.ckpt-41900
In [7]:
# Test a sample from the test dataset.
decoded = decode_sentence(sess, model, data_reader, "you must have girlfriend", corrective_tokens=corrective_tokens)
Input: you must have girlfriend
Output: you must have a girlfriend
In [7]:
decoded
---------------------------------------------------------------------------
NameError Traceback (most recent call last)
<ipython-input-7-e6ad1ea29283> in <module>()
----> 1 decoded
NameError: name 'decoded' is not defined
In [6]:
decoded = decode_sentence(sess, model, data_reader,
"did n't you say that they 're going to develop this revolutionary new thing ...",
corrective_tokens=corrective_tokens)
Input: did n't you say that they 're going to develop this revolutionary new thing ...
Output: did n't you say that they 're going to develop this revolutionary new thing ...
In [9]:
decode_sentence(sess, model, data_reader, "kvothe went to market", corrective_tokens=corrective_tokens, verbose=False)
Out[9]:
['kvothe', 'went', 'to', 'the', 'market']
In [10]:
decode_sentence(sess, model, data_reader, "blablahblah and bladdddd went to market", corrective_tokens=corrective_tokens,
verbose=False)
Out[10]:
['blablahblah', 'and', 'bladdddd', 'went', 'to', 'the', 'market']
In [11]:
decode_sentence(sess, model, data_reader, "do you have book", corrective_tokens=corrective_tokens, verbose=False)
Out[11]:
['do', 'you', 'have', 'a', 'book']
In [10]:
decode_sentence(sess, model, data_reader, "the cardinals did better then the cubs", corrective_tokens=corrective_tokens, verbose=False)
Out[10]:
['the', 'cardinals', 'did', 'better', 'than', 'the', 'cubs']
In [6]:
# 4 layers, 40k steps
errors = evaluate_accuracy(sess, model, data_reader, corrective_tokens, test_path)#, max_samples=1000)
Bucket 0: (10, 10)
Baseline BLEU = 0.8354
Model BLEU = 0.8492
Baseline Accuracy: 0.9090
Model Accuracy: 0.9354
Bucket 1: (15, 15)
Baseline BLEU = 0.8826
Model BLEU = 0.8595
Baseline Accuracy: 0.8055
Model Accuracy: 0.8149
Bucket 2: (20, 20)
Baseline BLEU = 0.8880
Model BLEU = 0.8216
Baseline Accuracy: 0.7301
Model Accuracy: 0.6689
Bucket 3: (40, 40)
Baseline BLEU = 0.9097
Model BLEU = 0.6357
Baseline Accuracy: 0.5981
Model Accuracy: 0.2283
In [9]:
# 4 layers, 30k steps
errors = evaluate_accuracy(sess, model, data_reader, corrective_tokens, test_path)#, max_samples=1000)
Bucket 0: (10, 10)
Baseline BLEU = 0.8368
Model BLEU = 0.8425
Baseline Accuracy: 0.9110
Model Accuracy: 0.9303
Bucket 1: (15, 15)
Baseline BLEU = 0.8818
Model BLEU = 0.8459
Baseline Accuracy: 0.8063
Model Accuracy: 0.8014
Bucket 2: (20, 20)
Baseline BLEU = 0.8891
Model BLEU = 0.7986
Baseline Accuracy: 0.7309
Model Accuracy: 0.6281
Bucket 3: (40, 40)
Baseline BLEU = 0.9099
Model BLEU = 0.5997
Baseline Accuracy: 0.6007
Model Accuracy: 0.1607
In [13]:
# 4 layers, 20k steps
errors = evaluate_accuracy(sess, model, data_reader, corrective_tokens, test_path)#, max_samples=1000)
Bucket 0: (10, 10)
Baseline BLEU = 0.8330
Model BLEU = 0.8335
Baseline Accuracy: 0.9067
Model Accuracy: 0.9218
Bucket 1: (15, 15)
Baseline BLEU = 0.8772
Model BLEU = 0.8100
Baseline Accuracy: 0.7980
Model Accuracy: 0.7437
Bucket 2: (20, 20)
Baseline BLEU = 0.8898
Model BLEU = 0.7636
Baseline Accuracy: 0.7366
Model Accuracy: 0.5370
Bucket 3: (40, 40)
Baseline BLEU = 0.9098
Model BLEU = 0.5387
Baseline Accuracy: 0.6041
Model Accuracy: 0.1117
In [16]:
errors = evaluate_accuracy(sess, model, data_reader, corrective_tokens, test_path)#, max_samples=1000)
Bucket 0: (10, 10)
Baseline BLEU = 0.8341
Model BLEU = 0.8516
Baseline Accuracy: 0.9083
Model Accuracy: 0.9384
Bucket 1: (15, 15)
Baseline BLEU = 0.8850
Model BLEU = 0.8860
Baseline Accuracy: 0.8156
Model Accuracy: 0.8491
Bucket 2: (20, 20)
Baseline BLEU = 0.8876
Model BLEU = 0.8880
Baseline Accuracy: 0.7291
Model Accuracy: 0.7817
Bucket 3: (40, 40)
Baseline BLEU = 0.9099
Model BLEU = 0.9045
Baseline Accuracy: 0.6073
Model Accuracy: 0.6425
In [15]:
for decoding, target in errors:
print("Decoding: " + " ".join(decoding))
print("Target: " + " ".join(target) + "\n")
Decoding: you beg for mercy in a second .
Target: you 'll beg for mercy in a second .
Decoding: i 'm dying for a shower . you could use the one too . and we 'd better check that bandage .
Target: i 'm dying for a shower . you could use one too . and we 'd better check that bandage .
Decoding: whatever ... they 've become hotshot computer guys so they get a job to build el computer grande ... skynet ... for the government . right ?
Target: whatever ... they become the hotshot computer guys so they get the job to build el computer grande ... skynet ... for the government . right ?
Decoding: did n't you say that they 're going to develop this revolutionary a new thing ...
Target: did n't you say that they 're going to develop this revolutionary new thing ...
Decoding: bag some z ?
Target: bag some z 's ?
Decoding: sleep . it 'll be a light soon .
Target: sleep . it 'll be light soon .
Decoding: well , at least i know what to name him . i do n't suppose you 'd know who father is ? so i do n't tell him to get lost when i meet him .
Target: well , at least i know what to name him . i do n't suppose you 'd know who the father is ? so i do n't tell him to get lost when i meet him .
Decoding: we got ta get you to doctor .
Target: we got ta get you to a doctor .
Decoding: hunter killers . patrol machines . a build in automated factories . most of us were rounded up , put in camps ... for orderly disposal .
Target: hunter killers . patrol machines . build in automated factories . most of us were rounded up , put in camps ... for orderly disposal .
Decoding: but outside , it 's a living human tissue . flesh , skin , hair ... blood . grown for the cyborgs .
Target: but outside , it 's living human tissue . flesh , skin , hair ... blood . grown for the cyborgs .
Decoding: you heard enough . decide . are you going to release me ?
Target: you 've heard enough . decide . are you going to release me ?
Decoding: okay . okay . but this ... cyborg ... if it metal ...
Target: okay . okay . but this ... cyborg ... if it 's metal ...
Decoding: you go naked . something about the field generated by living organism . nothing dead will go .
Target: you go naked . something about the field generated by a living organism . nothing dead will go .
Decoding: ca n't . nobody goes home . nobody else comes through . it just him and me .
Target: ca n't . nobody goes home . nobody else comes through . it 's just him and me .
Decoding: i see . and this ... computer , thinks it can win by killing the mother of its enemy , kill- ing him , in effect , before he is even conceived ? sort of retroactive abortion ?
Target: i see . and this ... computer , thinks it can win by killing the mother of its enemy , kill- ing him , in effect , before he is even conceived ? a sort of retroactive abortion ?
Decoding: skynet . a computer defense system built for sac-norad by cyber dynamics . modified series 4800 .
Target: skynet . a computer defense system built for sac-norad by cyber dynamics . a modified series 4800 .
Decoding: a year 2027 ?
Target: the year 2027 ?
Decoding: with one thirty a second under perry , from '21 to '27 --
Target: with the one thirty second under perry , from '21 to '27 --
Decoding: why do n't you just stretch out here and get some sleep . it take your mom 's a good hour to get here from redlands .
Target: why do n't you just stretch out here and get some sleep . it 'll take your mom a good hour to get here from redlands .
Decoding: lieutenant , are you sure it them ? maybe i should see the ... bodies .
Target: lieutenant , are you sure it 's them ? maybe i should see the ... bodies .
Decoding: i already did . no answer at the door and the apartment manager 's out . i keeping them there .
Target: i already did . no answer at the door and the apartment manager 's out . i 'm keeping them there .
Decoding: that stuff two hours cold .
Target: that stuff 's two hours cold .
Decoding: you got ta be kidding me . the new guys 'll be short-stroking it over this one . one-day pattern killer .
Target: you got ta be kidding me . the new guys 'll be short-stroking it over this one . a one-day pattern killer .
Decoding: give me a short version .
Target: give me the short version .
Decoding: because it 's fair . give me the next quarter . if you still feel this way , vote your shares ...
Target: because it 's fair . give me next quarter . if you still feel this way , vote your shares ...
Decoding: it 's probably will . in fact , i 'd go so far as to say it 's almost certainly will , in time . why should i settle for that ?
Target: it probably will . in fact , i 'd go so far as to say it almost certainly will , in time . why should i settle for that ?
Decoding: stock will turn .
Target: the stock will turn .
Decoding: you want to know what it is ? what 's it all about ? john . chapter nine . verse twenty-five .
Target: you want to know what it is ? what it 's all about ? john . chapter nine . verse twenty-five .
Decoding: i only mention it because i took a test this afternoon , down on montgomery street .
Target: i only mention it because i took the test this afternoon , down on montgomery street .
Decoding: christine ! mister van orton is valued customer ...
Target: christine ! mister van orton is a valued customer ...
Decoding: a single ?
Target: single ?
Decoding: there 's another gig starting in saudi arabia . i just a walk-on this time though . bit-part .
Target: there 's another gig starting in saudi arabia . i 'm just a walk-on this time though . bit-part .
Decoding: no ! you take another step , i shoot ! they 're trying to kill me ...
Target: no ! you take another step , i 'll shoot ! they 're trying to kill me ...
Decoding: listen very carefully , i 'm telling the truth ... this is a game . this was all the game .
Target: listen very carefully , i 'm telling the truth ... this is the game . this was all the game .
Decoding: that 's gun . that 's ... that 's not automatic . the guard had an automatic ...
Target: that gun . that ... that 's not automatic . the guard had an automatic ...
Decoding: take a picture out .
Target: take the picture out .
Decoding: yeah . first communion . are n't i little angel ?
Target: yeah . first communion . are n't i a little angel ?
Decoding: let me go get some clothes on . we talk , okay ? be right back .
Target: let me go get some clothes on . we 'll talk , okay ? be right back .
Decoding: i 'm tired . i 'm sorry , i should go . i 've been enough of nuisance .
Target: i 'm tired . i 'm sorry , i should go . i 've been enough of a nuisance .
Decoding: they said five hundred . i said six . they said man in the gray flannel suit . i think i said , you mean the attractive guy in the gray flannel suit ?
Target: they said five hundred . i said six . they said the man in the gray flannel suit . i think i said , you mean the attractive guy in the gray flannel suit ?
Decoding: i have a confession to make . someone gave me six-hundred dollars to spill a drinks on you , as a practical joke .
Target: i have a confession to make . someone gave me six-hundred dollars to spill drinks on you , as a practical joke .
Decoding: maitre d ' called you christine .
Target: the maitre d ' called you christine .
Decoding: i know owner of campton place . i could talk to him in the morning .
Target: i know the owner of campton place . i could talk to him in the morning .
Decoding: fresh shirt ...
Target: a fresh shirt ...
Decoding: investment banking . moving money from a place to place .
Target: investment banking . moving money from place to place .
Decoding: what 's the c .r .s . ?
Target: what 's c .r .s . ?
Decoding: this is a c .r .s .
Target: this is c .r .s .
Decoding: their ladder here .
Target: there 's a ladder here .
Decoding: this is n't attempt to be gallant . if i do n't lift you , how are you going to get there ?
Target: this is n't an attempt to be gallant . if i do n't lift you , how are you going to get there ?
Decoding: are you suggesting we wait till someone 's finds us ?
Target: are you suggesting we wait till someone finds us ?
Decoding: `` ... wait for help . '' wait for help . i 'm not opening that specifically warns me not to .
Target: `` ... wait for help . '' wait for help . i 'm not opening a door that specifically warns me not to .
Decoding: read what it says : `` warning , do < u > not < /u > attempt to open . if elevator stops , use the emergency ... ``
Target: read what it says : `` warning , do < u > not < /u > attempt to open . if elevator stops , use emergency ... ``
Decoding: long story . i found this key in the mouth of wooden harlequin .
Target: long story . i found this key in the mouth of a wooden harlequin .
Decoding: how do you know that way ?
Target: how do you know that 's the way ?
Decoding: it 's run by company ... they play elaborate pranks . things like this . i 'm really only now finding out myself .
Target: it 's run by a company ... they play elaborate pranks . things like this . i 'm really only now finding out myself .
Decoding: you got to be kidding .
Target: you 've got to be kidding .
Decoding: i do n't think he breathing .
Target: i do n't think he 's breathing .
Decoding: a bad month . you did exact the same thing to me last week .
Target: a bad month . you did the exact same thing to me last week .
Decoding: yeah , yeah . she 's called a cab . said something about catching plane .
Target: yeah , yeah . she called a cab . said something about catching a plane .
Decoding: oh , god yes please . thanks , man . i take you up on that .
Target: oh , god yes please . thanks , man . i 'll take you up on that .
Decoding: this ... ? oh , this is just ... this is bill .
Target: this ... ? oh , this is just ... this is the bill .
Decoding: baby , they were all over the house with metal detectors . they switched your gun with look-alike , rigged barrel , loaded with blanks . pop-gun .
Target: baby , they were all over the house with metal detectors . they switched your gun with a look-alike , rigged barrel , loaded with blanks . pop-gun .
Decoding: you dodged bullet .
Target: you dodged a bullet .
Decoding: c .r .s . who do you think ? jesus h . , thank your lucky charms . to think what i 've almost got you into .
Target: c .r .s . who do you think ? jesus h . , thank your lucky charms . to think what i almost got you into .
Decoding: it 's profound life experience .
Target: it 's a profound life experience .
Decoding: you 've heard of it . you 've seen other people having it . they 're entertainment service , but more than that .
Target: you 've heard of it . you 've seen other people having it . they 're an entertainment service , but more than that .
Decoding: they make your life fun . there 's only guarantee is you will not be bored .
Target: they make your life fun . their only guarantee is you will not be bored .
Decoding: not after i done with it . actually , i 've been here . in grad-school i bought crystal-meth from the maitre d ' .
Target: not after i 'm done with it . actually , i 've been here . in grad-school i bought crystal-meth from the maitre d ' .
Decoding: that 's why it 's a classic . come on , man ... how 'bout hug ... ?
Target: that 's why it 's a classic . come on , man ... how 'bout a hug ... ?
Decoding: how much is it ? a few thousand , at least . a rolex like that ... lucky for you 've missed it .
Target: how much is it ? a few thousand , at least . a rolex like that ... lucky for you they missed it .
Decoding: i told you , they hired me over the phone . i 've never met anyone .
Target: i told you , they hired me over the phone . i never met anyone .
Decoding: i do n't want money . i 'm pulling back curtain . i 'm here to meet the wizard .
Target: i do n't want money . i 'm pulling back the curtain . i 'm here to meet the wizard .
Decoding: tell them the cops are after you ... tell them you got to talk to someone , i 'm threatening to blow the whistle .
Target: tell them the cops are after you ... tell them you 've got to talk to someone , i 'm threatening to blow the whistle .
Decoding: they own the whole building . they just move from the floor to floor .
Target: they own the whole building . they just move from floor to floor .
Decoding: look , it was just a job . nothing personal , ya know ? i play my part , improvise little . that 's what i 'm good at .
Target: look , it was just a job . nothing personal , ya know ? i play my part , improvise a little . that 's what i 'm good at .
Decoding: that 's right -- you 're left-brain the word fetishist .
Target: that 's right -- you 're a left-brain word fetishist .
Decoding: one guarantee . payment 's entirely at your brother discretion and , as a gift , dependent on your satisfaction .
Target: one guarantee . payment 's entirely at your brother 's discretion and , as a gift , dependent on your satisfaction .
Decoding: your brother was a client with our branch . we do a sort of informal scoring . his numbers were outstanding . sure you 're not hungry at all ... ? tung hoy , best in chinatown ...
Target: your brother was a client with our london branch . we do a sort of informal scoring . his numbers were outstanding . sure you 're not hungry at all ... ? tung hoy , best in chinatown ...
Decoding: key ?
Target: the key ?
Decoding: nobody 's worried about your father .
Target: nobody worried about your father .
Decoding: there 's been a break in . lock this door and stay here . do n't move muscle .
Target: there 's been a break in . lock this door and stay here . do n't move a muscle .
Decoding: i do n't know what you 're talking about . what happened ?
Target: i do n't know what you 're talking about . what 's happened ?
Decoding: did alarm go off ? the house ... they ... you did n't see ... ?
Target: did the alarm go off ? the house ... they ... you did n't see ... ?
Decoding: then then .
Target: goodnight then .
Decoding: okay . i think he into some sort of new personal improvement cult .
Target: okay . i think he 's into some sort of new personal improvement cult .
Decoding: dinner in the oven .
Target: dinner 's in the oven .
Decoding: there was incident a few days ago ... a nervous breakdown , they said . the police took him . they left this address , in case anyone ...
Target: there was an incident a few days ago ... a nervous breakdown , they said . the police took him . they left this address , in case anyone ...
Decoding: what 's trouble ?
Target: what 's the trouble ?
Decoding: mister ... seymour butts .
Target: a mister ... seymour butts .
Decoding: what 's the gentleman , maria ?
Target: what gentleman , maria ?
Decoding: i would n't mention following , except he was very insistent . it 's obviously some sort of prank ...
Target: i would n't mention the following , except he was very insistent . it 's obviously some sort of prank ...
Decoding: i send your regrets . honestly , why must i even bother ?
Target: i 'll send your regrets . honestly , why must i even bother ?
Decoding: the hinchberger 's wedding .
Target: the hinchberger wedding .
Decoding: invitations : museum gala .
Target: invitations : the museum gala .
Decoding: nice touch . does a game use real bullets ... ?
Target: nice touch . does the game use real bullets ... ?
Decoding: it 's what they do . it 's like ... being toyed with by a bunch of ... depraved children
Target: it 's what they do . it 's like ... being toyed with by a bunch of ... depraved children .
Decoding: find out about a company called the c .r .s . consumer recreation services .
Target: find out about a company called c .r .s . consumer recreation services .
Decoding: someone 's playing hardball . it 's complicated . can i ask favor ?
Target: someone 's playing hardball . it 's complicated . can i ask a favor ?
Decoding: how 's the concerned should i be ?
Target: how concerned should i be ?
Decoding: that you 've a involved conrad ... is unforgivable . i am now your enemy .
Target: that you 've involved conrad ... is unforgivable . i am now your enemy .
Decoding: what happened ...
Target: what 's happened ...
Decoding: modelling small-group dynamics in formation of narrative hallucinations . you brought us here to scare us . insomnia , that was just a decoy issue . you 're disgusting .
Target: modelling small-group dynamics in the formation of narrative hallucinations . you brought us here to scare us . insomnia , that was just a decoy issue . you 're disgusting .
Decoding: come on . these are the typically sentimental gestures of depraved industrialist .
Target: come on . these are the typically sentimental gestures of a depraved industrialist .
Decoding: the children . children hugh crain built the house for . the children he never had .
Target: the children . the children hugh crain built the house for . the children he never had .
Decoding: obsessive worrier . join club . and you ? i 'd guess ...
Target: obsessive worrier . join the club . and you ? i 'd guess ...
Decoding: so why did you need the addam family mansion for a scientific test ?
Target: so why did you need the addam 's family mansion for a scientific test ?
Decoding: -- how much is this car 's worth ?
Target: -- how much is this car worth ?
Decoding: you do n't really believe it haunted ... do you believe in ghosts ?
Target: you do n't really believe it 's haunted ... do you believe in ghosts ?
Decoding: so could you ! is this some fucked up the idea of art , putting someone else 's name to a painting ?
Target: so could you ! is this some fucked up idea of art , putting someone else 's name to a painting ?
Decoding: and why did n't marrow tell < u > us < /u > ? does n't he a trust women ? that fuck .
Target: and why did n't marrow tell < u > us < /u > ? does n't he trust women ? that fuck .
Decoding: nah , you 're going crazy with doubt , all of your mistakes are coming back up the pipes , and it 's worse than nightmare . --
Target: nah , you 're going crazy with doubt , all of your mistakes are coming back up the pipes , and it 's worse than a nightmare . --
Decoding: not the way you 've constructed your group , it just not ethical !
Target: not the way you 've constructed your group , it 's just not ethical !
Decoding: children want me . they 're calling me . they need me .
Target: the children want me . they 're calling me . they need me .
Decoding: i looked at theo . she had look on her face .
Target: i looked at theo . she had a look on her face .
Decoding: i was n't thinking about my mother bathroom .
Target: i was n't thinking about my mother 's bathroom .
Decoding: so ... smell ... is ... smell is sense that triggers the most powerful memories . and memory can trigger a smell .
Target: so ... smell ... is ... smell is the sense that triggers the most powerful memories . and a memory can trigger a smell .
Decoding: in the bathroom in my mother 's room , toilet was next to old wooden table . it smelled like that wood .
Target: in the bathroom in my mother 's room , the toilet was next to an old wooden table . it smelled like that wood .
Decoding: cold sensation . who felt it first ?
Target: the cold sensation . who felt it first ?
Decoding: i really ... honored to be part of this study , jim .
Target: i 'm really ... honored to be part of this study , jim .
Decoding: nell . good enough . and i jim .
Target: nell . good enough . and i 'm jim .
Decoding: that ? that 's a hill house .
Target: that ? that 's hill house .
Decoding: here 's how they 're organized . groups of five , very different personalities : scored all over the kiersey temperament sorter just like you asked for . and they all score high on insomnia charts .
Target: here 's how they 're organized . groups of five , very different personalities : scored all over the kiersey temperament sorter just like you asked for . and they all score high on the insomnia charts .
Decoding: you hear the vibrations in the wire . there 's magnetic pulse in the wires , you feel it . i could test it .
Target: you hear the vibrations in the wire . there 's a magnetic pulse in the wires , you feel it . i could test it .
Decoding: but experiment was a failure .
Target: but the experiment was a failure .
Decoding: he wandering around house , and nell heard him . she thought it was ghosts . let 's go look for him again .
Target: he 's wandering around the house , and nell heard him . she thought it was ghosts . let 's go look for him again .
Decoding: i 'll take her with me to university tomorrow . i ca n't believe i read the test wrong . i did n't see anything that looked like she was suicidal .
Target: i 'll take her with me to the university tomorrow . i ca n't believe i read the test wrong . i did n't see anything that looked like she was suicidal .
Decoding: no , but nell been here longer than i have .
Target: no , but nell 's been here longer than i have .
Decoding: rene crain . up there . rope . ship 's hawser . hard to tie . do n't know how she 's got it .
Target: rene crain . up there . rope . ship 's hawser . hard to tie . do n't know how she got it .
Decoding: mrs . dudley be waiting for you .
Target: mrs . dudley 'll be waiting for you .
Decoding: that 's a good question . what is it about fences ? sometimes a locked chain makes people on both sides of fence just a little more comfortable . why would that be ?
Target: that 's a good question . what is it about fences ? sometimes a locked chain makes people on both sides of the fence just a little more comfortable . why would that be ?
Decoding: well , i 've never lived with a beauty . you must love working here .
Target: well , i 've never lived with beauty . you must love working here .
Decoding: nell , it makes sense . it 's all makes sense . you and i , we were scaring each other , working each other up .
Target: nell , it makes sense . it all makes sense . you and i , we were scaring each other , working each other up .
Content source: Raja-VP/dl-text-corrector
Similar notebooks: