Title: Free construction of a free dictionary of synonyms using computer science
1Free construction of a free dictionary of
synonyms using computer science
- Viggo Kann and Magnus RosellKTH, Stockholm
- Talk given by Viggo at Amherst College November
11, 2006
2Examples of English synonyms
- Smith A Dictionary of Synonymous Words in the
English Language 1889 - CLASS. Order. Rank. Degree. Classification.
Grade.
- Websters Dictionary of Synonyms 1942
- classify. Alphabetize, pigeonhole, assort, sort.
Ana. Order, arrange, systematize, methodize,
marshal.
3Goals
- To construct a Swedish dictionary of synonyms as
a list of synonymous pairs - I dont want to work a lot
- I dont want to pay anyone to work
- The resulting list should be free
4Ideas
- Automatically construct a large set of word pairs
that might be synonyms - Use ten thousands of people, who are each willing
to make a small contribution without payment, to
check the word pairs
5More ideas
- Use the Lexin on-line Swedish-English dictionary
web site, that had 9 millions (now 17 M) of
lookups each month - Users visit Lexin to translate words, and are
thus probably motivated to help me - Each time a user makes a lookup, give her the
opportunity to decide whether two words are
synonyms or not
6My plan
- Construct lots of possible synonyms
- Sort out bad synonym pairs automatically
- Ask lots of users if the rest of the pairs are
good synonyms - Analyze the gradings done by the users and decide
which pairs to keep
7Step 1 Construct lots of possible synonyms
- If we have access to a Swedish-English dictionary
SE and an English-Swedish dictionary ES, try to
translate each word to English and back again to
Swedish - (w,v) ?y y?SE(w) ? v?ES(y) or(w,v) ?y
y?SE(w) ? y?SE(v) - 616 000 word pairs were generated
8Step 2 Sort out bad synonym pairs automatically
- Use RI (Random Indexing)Kanerva, Kristoferson,
Holst 2000to measure the distance between words
represented in a large vector space - Keep pairs that have small enough distance in the
vector space
9Random Indexing
- Each word w is assigned a random label vector Lw
of thousand elements - For each word w construct a context vector Cw by
adding the random vectors for the words appearing
in the context of each occurrence of w in a large
corpus
10Random Indexing settings
- Context 4 words to the left and 4 to the
rightStop words were removed - Dimensionality 1800
- 5 corpora from different domains were used, for
example newspapers and medical texts
11Number of pairs for different cos thresholds (435
000 of 616 000 pairs occurred in corpus)
12Step 3 Ask lots of users if the rest of the
pairs are good synonyms
- When a user has sent a word to the Lexin
dictionary he receives the translation followed
by a question like - Are 'spread' and 'lengthen' synonyms? Answer
using a scale from 0 to 5 where 0 means 'I dont
agree' and 5 means 'I do fully agree', or answer
'I dont know'
13After answering the user may
- grade new randomly chosen word pair
- look up word in the synonym dictionary
- suggest new synonymous word pair
- download synonym dictionary in XML
14(No Transcript)
15Step 4 Analyzing the gradings done by the users
- 1.2 millions gradings were made in less than 2
months - Grading statistics were analyzed on several
occasions - Some users sent comments
16Keeping the users happy!
- Many users said that there were too many bad
pairs - Lots of pairs were graded 0 (not at all synonyms)
by all users. After some weeks 25 000 such pairs
were removed. Later 60 000 more pairs were
removed, improving the quality of the remaining
pairs considerably.
17User gradings first two months
18More interesting gradings 2006
19Distribution of mean gradings of word pairs after
two months
20Distribution of mean gradings of word pairs 2006
21Analysis of the pairs graded 0Distance (cosine)
in RI space
22Some statistics (November 2006)
- 2.5 M user gradings done
- 67 000 pairs (graded 2) in dictionary
- 90 000 pairs suggested by users
- 50 000 unique pairs suggested
- 14 000 of them have been accepted
23Example Synonyms to klass (class)
- 5 rang (grade)rank (rank)slag (kind)
- 4 kategori (category)stånd (social
class)årskurs (grade) - 3 fack (sphere)grad (degree)grupp
(group)kvalitet (quality)nivå (level)ordning
(order)
- 3 skikt (layer)sort (sort)standard
(standard)stil (style) - 2 storleksordning (magnitude)typ (type)
- 1 poäng (point)stadga (stability)
- 0 uppdrag (mission)utbilda (educate)
24How to prevent abuse?
- Many gradings of a word pair are needed before
its considered to be good - The pair to be graded is randomly picked from a
very large list - Word pairs suggested by users are spell checked
before they are added to the very large list
25People's definition of synonymy
- Exact meaning of 'synonym' wasnt defined
- Users will grade using their intuitive
understanding of the concept of synonymy and the
words in the pair - The produced dictionary will use the people's own
definition of synonymy Hopefully this is exactly
what they want!
26The peoples synonym dictionary on the web
- http//lexin.nada.kth.se/cgi-bin/synlex
27Lessons learned
- The list of suggested synonyms should be huge
- Try to improve the quality of the list
automatically as much as possible,Random
indexing is useful for this, also try tagging and
using other dictionaries - Use the 0 answers early to remove bad pairs that
only irritate the users