clean-data (#1)

- Clean data (55d1991543e12c7a03ebea8f0a72824d5726b808)

100M/aochildes.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4a9ade69911e4d7351e5d1427c2e415a61f63357068d88631cac56502e0f4654
-size 18211974

100M/bnc_spoken.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:42c1a30d5923d3b86272d553bcdc4f989a0b5e58c649249665080bd749916ccf
-size 42472678

100M/cbt.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:13cfc7743cea6752b5f22fa49c8977292428e176b11963d3906f10528e491401
-size 25742364

100M/children_stories.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e0cd311ad21fa042f021a680820f60fd9599f4bc394f3b6e33fc7e20073247fd
-size 17447597

100M/gutenberg.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ece0df262b6d10472dea4be608974a21cb11520844e79128544ac41b0288384f
-size 54796262

100M/open_subtitles.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:758897a04d43c584f4969fabfcfd80aa3e37df1cbfa843719326bd2b0db6c756
-size 166489947

100M/qed.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ffa520f03fedbd5f6fd36babfeecc68061a6c455f5f708712ed572a404118609
-size 56962048

100M/simple_wikipedia.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f94bfa6f1f15e4b7fcac833133fbaa5c34565f7a18c6f36ab89f9d68391b1ca4
-size 87972498

100M/wikipedia.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e43a7fd062cf5ead3f50b2c4b9662cbed0e9afe1ddb2e354fbfd2f1941bc2cef
-size 61643655

10M/open_subtitles.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8ce0277990b9977ba3242b865fdd496e7fb94fa9531674a2721f9a4c11abbc3a
-size 16450052

BabyLM.py

@@ -65,9 +65,9 @@ class BabyLM(datasets.GeneratorBasedBuilder):
             train_data_dir = "100M"
 
         urls_to_download = {
-            "train": [f"{train_data_dir}/{fn}" for fn in filenames],
-            "dev": [f"dev/{fn}" for fn in filenames],
-            "test": [f"test/{fn}" for fn in filenames]
+            "train": [f"clean/{train_data_dir}/{fn}" for fn in filenames],
+            "dev": [f"clean/dev/{fn}" for fn in filenames],
+            "test": [f"clean/test/{fn}" for fn in filenames]
         } 
 
         downloaded_files = dl_manager.download_and_extract(urls_to_download)

clean_data.py

@@ -0,0 +1,230 @@
+""" Script used to clean the data. """
+
+import os
+import re
+from nltk import tokenize
+
+def clean_aochildes(lines):
+    """ For aochildes, we just remove the space between the punctuation mark and the final word """
+    new_lines = []
+    for line in lines:
+        new_lines.append(line[:-3] + line[-2:])
+    return new_lines
+
+def clean_bnc_spoken(lines):
+    """ For bnc_spoken, we lowercase """
+    new_lines = []
+    for line in lines:
+        new_lines.append(line.lower())
+    return new_lines
+
+def clean_cbt(lines):
+    """ For cbt, we lowercase and normalise punctuation """
+    punctuation = ['.', ',', '?', '!', ':', ';', '(', ')', '[', ']', '{', '}', '"', "'", '“', '”', '—', '–']
+    new_lines = []
+    for line in lines:
+        new_line = line.lower()
+        new_line = new_line.replace(": ' ", ":  \"")
+        new_line = new_line.replace("''", "\"")
+        new_line = new_line.replace(" '\n", "\"\n")
+        new_line = new_line.replace(" ' ", "\" ")
+        new_line = new_line.replace(" `` ", "  \"")
+        new_line = new_line.replace("` ", " \"")
+        new_line = new_line.replace("`", "\"")
+        new_line = new_line.replace("’", "\"")
+        for punct in punctuation:
+            new_line = new_line.replace(f" {punct}", punct)
+        new_lines.append(new_line)
+    return new_lines
+
+def clean_children_stories(lines):
+    """ For children_stories, we lowercase and split long lines into sentences """
+    new_lines = []
+    for line in lines:
+        sentences = [s + '\n' for s in tokenize.sent_tokenize(line.lower().strip()) if s != '']
+        new_lines.extend(sentences)
+    return new_lines
+
+def clean_gutenberg(lines):
+    """ For gutenberg, we lowercase, remove italics, group lines into paragraphs and then split into sentences """
+    # Get paragraphs
+    paragraphs = []
+    paragraph = ""
+    for line in lines:
+        # Remove italics
+        tmp_line = line.lower().strip().replace('_','')
+        if tmp_line == "" and paragraph != "":
+            paragraphs.append(paragraph[:-1] + '\n')
+            paragraph = ""
+        else:
+            paragraph += tmp_line + " "
+    
+    # Split into sentences using NLTK
+    new_lines = []
+    for paragraph in paragraphs:
+        sentences = [s + '\n' for s in tokenize.sent_tokenize(paragraph) if s != '']
+        new_lines.extend(sentences)
+    return new_lines
+
+def clean_open_subtitles(lines):
+    """ For open_subtitles, we lowercase, remove subtitle dashes and fix the lowercase 'l' problem """
+    punctuation = ['.', ',', '?', '!', ':', ';', '(', ')', '[', ']', '{', '}', '"', "'", '“', '”', '—', '–', ' ', '\n']
+    new_lines = []
+    for line in lines:
+        new_line = line.lower()
+        if new_line[0:2] == "- ":
+            new_line = new_line[2:]
+        if new_line[0] == "-":
+            new_line = new_line[1:]
+        new_line = ' ' + new_line
+        for punct in punctuation:
+            new_line = new_line.replace(f" l{punct}", f" i{punct}")
+            new_line = new_line.replace(f" lm{punct}", f" im{punct}")
+            new_line = new_line.replace(f" lf{punct}", f" if{punct}")
+        new_line = new_line.replace(' lc', ' ic')
+        new_line = new_line.replace(' ld', ' id')
+        new_line = new_line.replace(' lj', ' i j')
+        new_line = new_line.replace(' ln', ' in')
+        new_line = new_line.replace(' lp', ' ip')
+        new_line = new_line.replace(' lr', ' ir')
+        new_line = new_line.replace(' ls', ' is')
+        new_line = new_line.replace(' isd', ' lsd')
+        new_line = new_line.replace(' lt', ' it')
+        new_line = new_line.replace(' lt', ' it')
+        new_line = new_line.replace(' lv', ' iv')
+        new_lines.append(new_line.strip() + '\n')
+    return new_lines
+
+def clean_qed(lines):
+    """ For qed, we lowercase and normalise punctuation, remove words contained in parentheses,
+    remove lines that arejust character's names and fix the lowercase 'l' problem"""
+
+    new_lines = []
+    for line in lines:
+        # Before lowercasing, check if the words in the line are uppercase containing lowercase 'l' instead of 'I' and fix accordingly
+        words = line.split()
+        for i, word in enumerate(words):
+            if word.replace('l','I').isupper() and 'l' in word and word != 'I\'ll':
+                words[i] = word.replace('l', 'I')
+        new_line = ' '.join(words).lower()
+        new_line = new_line.replace(' lc', ' ic')
+        new_line = new_line.replace(' ld', ' id')
+        new_line = new_line.replace(' lj', ' i j')
+        new_line = new_line.replace(' ln', ' in')
+        new_line = new_line.replace(' lp', ' ip')
+        new_line = new_line.replace(' lr', ' ir')
+        new_line = new_line.replace(' ls', ' is')
+        new_line = new_line.replace(' isd', ' lsd')
+        new_line = new_line.replace(' lt', ' it')
+        new_line = new_line.replace(' lt', ' it')
+        new_line = new_line.replace(' lv', ' iv')
+
+        # Skip lines that are just character names, e.g. "AMY GOODMAN:"
+        if len(new_line.strip()) < 1 or (len(words) <= 3 and new_line.strip()[-1] == ':'):
+            continue
+
+        # Remove subtitle dashes
+        if new_line[0:2] == "- ":
+            new_line = new_line[2:]
+        if new_line[0] == "-":
+            new_line = new_line[1:]
+
+        # Remove substrings contained within circular or square parantheses (screen descriptions)
+        pattern = r'\([^)]*\)'
+        new_line = re.sub(pattern, '', new_line)
+        pattern = r'\[[^)]*\]'
+        new_line = re.sub(pattern, '', new_line)
+        new_line = new_line.replace('"', '\'')
+
+        # Remove strange characters
+        new_line = new_line.replace('#','')
+        new_line = new_line.replace('*','')
+
+        new_line = new_line.strip()
+        if new_line != "":
+            new_lines.append(new_line + '\n')
+    return new_lines
+
+def clean_simple_wikipedia(lines):
+    """ For simple_wikipedia, we lowercase, remove empty lines and article names and split paragraphs into sentences."""
+    new_lines = []
+    next_line_is_article_name = False
+    for line in lines:
+        if next_line_is_article_name:
+            next_line_is_article_name = False
+            continue
+        if line.strip() == "":
+            next_line_is_article_name = True
+            continue
+        sentences = [s + '\n' for s in tokenize.sent_tokenize(line.lower()) if s != '']
+        new_lines.extend(sentences)
+    return new_lines
+
+def clean_switchboard(lines):
+    """ For switchboard, we lowercase """
+    new_lines = []
+    for line in lines:
+        new_line = line.lower()
+        new_lines.append(new_line)
+    return new_lines
+
+def clean_wikipedia(lines):
+    """ For wikipedia, we lowercase, remove empty lines and article names and split paragraphs into sentences.
+     We also remove lines that seem to be figure names or table entries. """
+    new_lines = []
+    for line in lines:
+        new_line = line.strip()
+        words = new_line.split()
+        
+        # Remove empty lines and article names
+        if new_line == "":
+            continue
+        if new_line[0] == "=" and new_line[-1] == "=":
+            continue
+
+        # Filter out lines that seem to be figure names or table entries
+        all_numeric = True
+        all_uppercase = True
+        for word in words:
+            if not word.isnumeric():
+                all_numeric = False
+            if not word[0].isupper():
+                all_uppercase = False
+        if all_numeric or all_uppercase:
+            continue
+    
+        # Split into sentences using NLTK
+        sentences = [s + '\n' for s in tokenize.sent_tokenize(new_line.lower()) if s != '']
+        new_lines.extend(sentences)
+    return new_lines
+
+CLEAN_FUNCTIONS = {'aochildes' : clean_aochildes, 'bnc_spoken' : clean_bnc_spoken, 'cbt' : clean_cbt, 'children_stories' : clean_children_stories, 'gutenberg' : clean_gutenberg, 'open_subtitles' : clean_open_subtitles, 'qed' : clean_qed, 'simple_wikipedia' : clean_simple_wikipedia, 'switchboard' : clean_switchboard, 'wikipedia' : clean_wikipedia}
+FOLDERS = ['10M', '100M', 'dev', 'test']
+
+if __name__ == "__main__":
+
+    # Read all text files from directory "BabyLM"
+    all_files = []
+    for folder in FOLDERS:
+        for root, dirs, files in os.walk(f"original/{folder}"):
+            for file in files:
+                if file.endswith(".txt"):
+                    all_files.append(os.path.join(root, file))
+
+    for file in all_files:
+        print(file)
+        with open(file, 'r') as f:
+            lines = f.readlines()
+
+        # Get the corpus name
+        corpus_name = os.path.basename(file).split('.')[0]
+
+        # Clean the data
+        if CLEAN_FUNCTIONS[corpus_name] is not None:
+            lines = CLEAN_FUNCTIONS[corpus_name](lines)
+
+        # Write the new file
+        new_file = file.replace('original', 'clean')
+        os.makedirs(os.path.dirname(new_file), exist_ok=True)
+        with open(new_file, 'w') as f:
+           f.writelines(lines)

dev/open_subtitles.txt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:72ee06e6fcbdb01b845c465be7da4089f373ac96bc3e3216e139ccca7fe92369
-size 15745512